(Studies in Brain and Mind 2) John Bickle (Auth.) - Philosophy and Neuroscience - A Ruthlessly Reductive Account-Springer Netherlands (2003)

Philosophy and Neuroscience
Studies in Brain and Mind

Volume 2
Series Editors '

John W. Bickle , University of Cincinnati, Cincinnati, Ohio
Kenneth J. Sufka , University of Mississippi, Oxford, Mississippi
Philosophy and
N euroscience
A Ruthlessly Reductive Account
by
John Bickle
University of Cincinnati, U.S.A.
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.
Library of Congress Cataloging-in-Publication Data

Bickle, Jobn.
Philosophy and neuroscience: a ruthlessly reductive account / by John Bickle.
p. ; crn. -- (Studies in brain and rnind ; 2)
Includes bibliographical references and index.
ISBN 978-1-4020-1302-7
ISBN 978-94-010-0237-0 (eBook)
DOI 10.1007/978-94-010-0237-0
1. Neurosciences--Philosophy. 1. Title. Il. Series.
[DNLM: 1. Psychophysiology. 2. Mind-Body Relations (Metaphysics) 3. Neuroscience.
4. Philosophy. 5. Psychologica1 Theory. WL 103 B583p 2003]
RC343.B43 2003
612.8'01--dc21
2003040140
ISBN 978-1-4020-1302-7
Printed on acid-free paper
AlI Rights Reserved

2003 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2003
Softcover reprint ofthe hardcover Ist edition 2003
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming,
recording or otherwise, without written permis sion from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
To Marica
To Caroline, Kat, and Margaret
Family
"And yet, there are philosophers who refuse to acknowledge scientific

philosophy as a philosophy, who wish to incorporate its results into an
introductory chapter of science and claim that there exists an independent
philosophy, which has no concern with scientific research and has direct
access to truth. Such claims, I think, reveal a lack of critical judgment. Those
who do not see the errors of traditional philosophy do not want to renounce its
methods or results and prefer to go on along a path which scientific
philosophy has abandoned. They reserve the name of philosophy for their
fallacious attempts at a superscientific knowledge and refuse to accept as
philosophical a method of analysis designed after the patterns of scientific
inquiry.
What is required for a scientific philosophy is a reorientation of
philosophic desires."
--Hans Reichenbach (1957), The Rise ofScientific Philosophy, 305.
CONTENTS
Preface
Chapter One: From New Wave Reduction to New Wave Metascience
1. Why Cellular and Molecular Neuroscience?
2. Background: The Intertheortic Reduction Reformulation of the
Mind-Body Problem
3. Revolts Against Nagel's Account
3.1 "Radical" Empiricism (and Patrick Suppes)
3.2 Schaffner's General Reduction (-Replacement) Paradigm
3.3 Hooker's General Theory of Reduction
4. Extending Hooker' s Insight: New Wave Reduction
4.1 Handling Multiple Realizability
4.2 New Wave Reduction
5. WWSD? (What Would Socrates Do?)
5.1 Problems for New Wave Reductionism
5.2 New Wave Metascience
Nores
xiii
6
10
10
15
16
21
21
26
29
29
31
40
Chapter Two: Reduction-in-Practice in Current Mainstream Neuroscience 43

I. A Proposed "Psychoneural Link"
44
2. Two Psychological Features of Memory Consolidation
46
3. LTP is Discovered
52
3.1 From Hebb's Neuropsychological Speculations, 1949, to
Norway, 1973
52
3.2 Some Basic Cellular Neuroscience
53
3.3 Back to Norway, 1973
61
4. Molecular Mechanisms of LTP: One Current Model
62
4.1 Early Phase LTP
63
4.2 Late Phase LTP
67
5. But is This Really Memory (Consolidation)?
75
5.1 Declarative Memory
76
5.2 Biotechnology Solves a Long-Standing Methodological
81
Problem in LTP-Memory Research
5.3 An Experimental Link Between Molecules and Behavior:
PKA, CREB, and Declarative Long-Term Memory Consolidation 88
6. The Nature of "Psychoneural Reduction" at Work in Current Mainstream (Cellular and Molecular) Neuroscience
95
Nores
102
Chapter Three: Mental Causation, Cognitive Neuroscience, and Multiple

Realization
1. The Problem of Mental Causation
2. Letting Neuroscientific Practice be Our Guide
3. What About Cognitive Neuroscience?
3.1 "Levels" Questions Within Neuroscience
3.2 Searching For the Cellular Mechanisms of the Sequential
Features of Higher Cognition
3.3 Cognitive Neuroscientific Resources to the Rescue:
Biological Modeling and Functional Neuroimaging
3.4 Philosophical Lessons From Transdisciplinary Neuroscience
4. Putnam's Challenge and the Multiple Realization Orthodoxy
5. Molecular Mechanisms of Nondeclarative Memory Consolidation in
Invertebrates
5.1 Single-Gene Fly Mutants for Associative Learning
5.2 Consolidating Nondeclarative Memory in the Sea Slug
6. Evolutionary Conservatism at the Molecular Level: The Expected
Scope of Shared Molecular Mechanisms
7. Consequences For Current Philosophy of Mind
Nores
107
107
111
115
115
117
121
128
131
136
136
141
149
157
158
Chapter Four: Consciousness

1. Prefrontal Neurons Possess Working Memory Fields
2. Construction and Modulation of Memory Fields: From Circuit
Connectivities to Receptor Proteins
3. Explicit Attention and Its Unremarkable Effects on Individual Neuron
Activity
4. Single-Cell Neurophysiology and the "Hard Problem"
4.1 Chalmers on Easy Versus Hard Problems of Consciousness
4.2 Neuroscientific Background: Wilder Penfield's Pioneering
Use of Cortical Stimulation
5. Inducing Phenomenology From Visual Motion to Somatosensory
Flutter ... And Beyond?
5.1 Results from William Newsome's Lab
5.2 Results from Kenneth Britten's Lab
5.3 Results from Ranulfo Romo's Lab
6. The Strange Case of Phenomenal Externalism
7. The "Hard Problem" and the Society for Neuroscience Crowd
Notes
163
165
194
194
200
203
206
212
213
Bibliography
217
171
178
189
189
190
LIST OF FIGURES
Figure 2.1 Average number of anticipatory runs into the "safe" box
50
Figure 2.2 A schematic neuron
55
Figure 2.3 A schematic neuron action potential
56
Figure 2.4 Effects of molecular mechanisms of E-LTP
65
Figure 2.5 Mechanisms of postsynaptic nitric oxide production and
retrograde transmission
68
Figure 2.6 Early steps in the molecular mechanisms inducing L-LTP
69
Figure 2.7 Basic constituents of a gene
72
Figure 2.8 Specific gene targets of phosphorylated CREB
73
Figure 2.9 Squire and colleagues current division of memory systems
77
Figure 2.10 Odor pairs in transitivity and symmetry memory tasks
81
Figure 2.11 Visual metaphor for the structure of intertheoretic reduction 101
Figure 3.1 Vector subtraction in eye movement space
120
Figure 3.2 Results with a computer simulation of our neurocomputational
model
123
Figure 3.3 Visual display during a four-step saccade sequencing trial
125
Figure 3.4 "Detrended" normalized composite mean BOLD signal
intensities in three regions
127
Figure 3.5 Molecular mechanisms of synaptic facilitation in Drosophila 140
Figure 3.6 Molecular mechanisms of synaptic facilitation in Aplysia
142
Figure 3.7 Fisher's model of adaptive evolution
153
Figure 4.1 Gross anatomy of primate prefrontal cortical region
Figure 4.2 Gross anatomy of sensory portions of primate cortex
Figure 4.3 Simplified circuit diagram of neural regions in "dorsal" and
"ventral" visual processing streams
Figure 4.4 Tuning curves of an orientation-selective V4 neuron
Figure 4.5 Newsome's measure of motion stimulus strength
Figure 4.6 Psychometric function relating responses to preferred motion
direction
Figure 4.7 Heading direction visual stimulus
169
182
183
185
195
197
202
PREFACE
As a work that combines philosophy and (what was then) current

science, I have always admired David Hull's (1974) Philosophy of Biological
Sciences in the Prentice Hall "Foundations of Philosophy Series." Hull's
writing is crisp, his presentation is focused, and-most importantly-the
scientific details, clearly described, drive the philosophy. I kept these features
of Hull's book constantly before me as I wrote this manuscript. I hope others
see in this work those same features, along with the addition of more science
from the primary literature, as befitting a book that seeks to do more than
introduce students to a philosophical area.
Neuroscience and its implications have not gone unnoticed by either
professional philosophers or the educated general public. How could they
have? We move closer every day to actually having something that human
beings have speculated about for centuries, a purely physical account of
behavioral causes. Yet it has struck me for close to a decade that most selfdescribed "philosophers of neuroscience" and "neurophilosophers"-people
with a professional stake in keeping up with the actual science-focus on the
wrong levels of research, theory, and experiment. Furthermore, this mistaken
emphasis by specialists produces negative repercussions in the larger
philosophy of mind/cognitive science community. Without question, neural
network modeling and computer simulation, functional neuroimaging, and
neuropsychological and neurological assessments are central neuroscientific
pursuits. But one glance at an influential neuroscience textbook, or a short
perusal through recent titles of colloquia talks delivered at a Ph.D.-granting
Neuroscience department, or--even more dramatically-one visit to the
week-long Society for Neuroscience annual meeting, should convince anyone
that the amount of research being done in "cognitive neuroscience" and the
amount we can safely be said to "know" at that level pales in comparison to
the amount going on and already discovered in the discipline's cellular and
molecular core. I'll begin developing this theme right off the bat in Chapter
One, but this entire book is at bottom an extended argument that higher-level
theorists of mind, especially philosophers, should reorient their interests
"down levels" in the neurosciences. Or, short of that, they should realize that
the mainstream core of the current science, the part on which all the higher
level cognitive neuroscientific investigations ultimately depend, has a "ruthless" reductionism built directly into its practice. Furthermore, at this cellular/
molecular level, we "know a lot about how the brain works" and we are
increasingly able to manipulate specific behaviors by intervening directly with
these cellular processes and intracellular pathways. This is no longer just the
xiv
PHILOSOPHY AND NEUROSCIENCE
fare of science fiction, but of Cell, Journal of Neurophysiology, Science,

Nature, and Proceedings ofthe National Academy ofSciences.
The status of "ruthless reductionism" in current mainstream
neuroscience contrasts sharply with its status in philosophy, even in the tip of
the "analytic" branch that over the last century embraced scientific philosophy. In his Presidential Address to the American Philosophical Association
(delivered orally in 1989), noted reductionist philosopher Jaegwon Kim
remarked:
Perhaps as a result of the singular lack of success with which
our earlier reductionist efforts have been rewarded, a negative
image seems to have emerged for reductionisms in general.
Many of us have the feeling that there is something rigid and
narrow-minded about reductionist strategies. Reductionisms,
we tend to feel, attempt to impose on us a monolithic, straitjacketed view of the subject matter, the kind of cleansed and
tidy picture that appeals to those obsessed with orderliness
and discipline.... Perhaps, too, reductionists are out of step
with the intellectual style of our times : we strive for patterns
of life and thought that are rich in diversity and complexity
and tolerant of disagreement and multiplicity. We are apt to
think that the real world is a messy place and resists any
simplistic drive, especially one carried on from the armchair,
toward simplification and unification. In fact, the word
"reductionism" seems by now to have acquired a negative,
faintly disreputable flavor-at least in philosophy of mind.
Being a reductionist is a bit like being a logical positivist or a
member of the Old Left-an aura of doctrinaire naivete hangs
over him. (1993, 265-266)
Kim's assessment of philosophical orthodoxy remains correct to this day.
The motivation guiding this book is that this orthodoxy remains
because (scientific) reduction is misunderstood. That diagnosis is, of course,
not original with me. But my new prescription for the malady is. I now
contend that the only way to overcome this misunderstanding is to show what
scientific reductionism is in practice-the experiments it motivates, the
results these experiments have yielded, and the way they are interpreted-in
the mainstream branch of a "hot" reductionist discipline. I leave it to readers,
both philosophers and scientists, to decide if this prescription works .
Realistically, however, can I expect philosophers and cognitive
scientists to wade through as much cellular and molecular detail as I have
included in this book? I hope so. There is a growing schism in both
philosophy of mind and philosophy of science, between metaphysically
minded and normatively prescriptive philosophers versus philosophers willing
to countenance scientific practice and results as scientists present them. I'm
PREFACE
xv
trying to push the agenda of the second camp one step beyond where it has
been pushed so far by philosophers interested in neuroscience. I seek to push
it into core neuroscience circa 2002. If successful, this will widen philosophy's schism. But that might not be such a bad thing. Perhaps it is time to
cleave philosophy of mind, philosophy of science, and philosophy of
particular sciences (like psychology, cognitive science, neuroscience, and
biology) into separate disciplines: one that, although mindful of scientific
practices and results, remains tied to perennial metaphysics and epistemology;
the other a part of science itself. This attitude reflects the rationale behind the
Reichenbach quote that serves as this book's epigram, as well as the first part
of my title. This is a book on philosophy and neuroscience, not philosophy of
neuroscience. Does the neuroscience overshadow the philosophy? As readers
will see throughout this book, I take on questions and arguments that have
been put forward by philosophers; only I do so while limiting myself to the
resources of recent cellular and molecular neuroscience. I also strive for the
"synoptic vision" of all of neuroscience that lies implicit in its mainstream
cellular and molecular core. In one sense I do leave things "entirely up to
science," but in another I am putting together the individual pieces that
science provides to make explicit the "bigger picture" that most scientists
leave implicit. That's "philosophy" enough for me.
My subtitle comes from my colleague, Robert Richardson. Those who
know Bob know of his ruthless wit. Appropriately, he heard my original,
boring subtitle, A Thoroughly Reductive Account, and immediately suggested
the much punchier form. Continuing discussions with a number of my
colleagues in Philosophy at the University of Cincinnati-in particular,
Richardson, Don Gustafson, Christopher Gauker, and Tom Polger-helped
me clarify arguments (and strengthened my conviction that I was on the right
track!). The Neuroscience Graduate Program at the University of Cincinnati
College of Medicine, of which I am very proud to be a part, keeps me up on
the latest scientific developments and trends through its weekly visiting
speaker's seminar. My scientific collaborators on a recent functional
neuroimaging project-Scott Holland at the University of Cincinnati and
Childrens Hospital, Cincinnati, Malcolm Avison at the University of
Kentucky Medical Center, and Vince Schmithorst at Childrens Hospital,
Cincinnati-have all commented helpfully on these themes over many
discussions. My current Ph.D. student, Anthony Landreth, assisted with this
manuscript in numerous ways. I tried out earlier versions of this material in a
Fall 2000 seminar at the University of Cincinnati, and thank Ph.D. students
from Philosophy, Biology, and Neuroscience for helpful questions and
comments.
I've presented talks that became pieces of this book at many
conferences and colloquia over the past few years, so thanks to all who've
made me think things through again with their questions and comments. This
group includes especially Ken Aizawa (Centenary College), Louise Antony
(Ohio State University), Ansgar Beckerman (Univresity of Bielefeld), Luc
xvi
Faucher (University of Quebec at Montreal), Owen Flanagan (Duke

University), Nicolas Georgalis (East Carolina University), Melvin Goodale
(University of Western Ontario), Valerie Hardcastle (Virginia Tech), Terry
Horgan (University of Arizona), Huib Looren de Jong (Vrije University
Amsterdam), Jaegwon Kim (Brown University), Michael Quante (University
of Muenster), Maurice Schouten (University of Tilburg), Sven Walter (Ohio
State University), and Omit Yalcin (East Carolina University). A number of
philosophers, neuroscientists, and psychologists commented on this manuscript's penultimate draft and numerous improvements resulted. This group
includes Jim Bogen (University of Pittsburgh), Carl Craver (Wahsington
University in St, Louis), Dingmar van Eck (Vrije University Amsterdam),
Trent Jerde (University of Minnesota), Huib Looren de Jong (Vrije University
Amsterdam), Maurice Schouten (University of Tilburg), John Symons
(University of Texas at El Paso), and Kenneth Sufka (University of
Mississippi). Thanks also to graphics artist David Winterhalter, who adapted
and created some of the figures. A Publications grant from the University of
Cincinnati Taft Memorial Fund supported production of the illustrations.
The biggest thanks of all, of course, go to Marica Bernstein and my
family, Caroline, Kat, and Margaret Cooper. Marica not only created and
adapted some of the figures (right before the deadline for the final manuscript!), but also helped me write some of the scientific sections covering
contemporary molecular genetics and, as always, was a critical respondent to
every idea presented here. Together we live the excitement of our shared
scientific interests, and I can't imagine a better partner to share my professsional and personal lives. They, like Marica and I, meld into one.
November 2002
Cincinnati, Ohio
CHAPTER ONE
FROM NEW WAVE REDUCTION TO NEW

WAVE METASCIENCE
This book is about contemporary neuroscience. More specifically, it

works with detailed examples drawn from current research to express that
discipline's reductive aspirations, aims, and potential. This reductionism holds
important consequences for some "hot" issues in contemporary philosophy of
mind . Even more specifically, this book is about the nature of reduction at
work in the mainstream core of the current discipline, cellular and molecular
neuroscience.
Questions arise immediately. Why look for philosophical lessons in
current cellular and molecular neuroscience? Why does the nature of
reduction at work in this area warrant interest, philosophical or scientific?
And why try to provide this account by focusing on detailed examples of
current research instead of, say, articulating a general account of scientific
reduction and applying it to them? These are questions I' II tackle in this
introductory chapter.
1 WHY CELLULAR AND MOLECULAR

NEUROSCIENCE?
Some philosophers (of mind and science) and cogmtive scientists
regularly keep abreast of developments in contemporary neuroscience.
Patricia Churchlands landmark Neurophilosophy (1986) provided an explicit
defense of this interdisciplinary attention. In the book's "General
Introduction," she writes: "In a way, nothing is more obvious than that
philosophers of mind could profit from knowing at least something of what
there is to know about how the brain works. After all, one might say, how
could the empirical facts about the nervous system fail to be relevant to
studies in the philosophy of mind?" (1986, 4). Even philosophers who don 't
share Churchlands exuberance for neuroscience can agree with this much. It
only requires that they take discoveries in our current sciences as relevant for
some philosophical issues. Of course, there are philosophers who reject even
this. They are beyond the pale (of this book) .
J. Bickle (eds.), Philosophy and Neuroscience
Kluwer Academic Publishers 2003
However, neuroscientifically astute philosophers and cognitive scientists have almost universally ignored the "cellular and molecular wave" that
swept through neuroscience over the past two decades. Instead, they've
focused on "cognitive neuroscience," the "interdisciplinary melding of studies
of the brain, of behavior and cognition, and of computational systems that
have properties of the brain and that can produce behavior and cognition"
(Kosslyn 1998, 158). Investigative techniques here range from state-of-the-art
functional neuroimaging to traditional neuropsychological measures to computational modeling in massively interconnected neural networks. It isn't
surprising that philosophers (and cognitive scientists) with neuroscientific
proclivities are attracted to this branch of the discipline. First, there is the
nearly universal intuition among high-level theorists that "levels" considerations and relations are crucial to understanding the mind-brain. In fact, the
mind-brain need not be thought of as special in this regard. Most philosophers
of biology assume a similar view about the importance of levels in the study
of higher-level biological phenomena; the "philosophy of molecular biology"
is hardly a recognized area. In addition to these "levels" intuitions, there is
also familiarity. Philosophers are at home with cognitive neuroscience's
descriptions of behavior and cognition, and with the types of behavior and
cognition these scientists investigate. Grasping the experimental methods isn't
even much of a professional stretch . The physics of functional neuroimaging
are daunting, but even practicing cognitive neuroscientists who employ these
methods tend to leave their physics to physicists and concentrate on the
behavioral and control tasks and the functional interpretation of analyzed data.
The basic concepts of neurocomputational techniques and their mathematics
are readily presented geometrically, making them comprehensible to anyone
with some quantitative background (Churchland and Sejnowski 1992).
Finally, the relevance of cognitive neuroscientific theories and explanations
for philosophical (and cognitive scientific) issues is usually readily apparent.
Many philosophers are interested in "naturalizing" intentionality, consciousness, and the like. Cognitive neuroscience appears to be a direct scientific
analog of philosophical "naturalizing" projects. In short, the levels of theory
and explanation inhabited by contemporary cognitive neuroscience are nearby
those of scientifically inspired philosophy of mind. So why search through
other branches of current neuroscience for philosophical consequences and
implications?
The principal reason is straightforward: neuroscience's "mainstream"
currently lies elsewhere. It lies in cellular physiology and molecular biology.
This "revolution" began two decades ago and now is in full swing. It is in
keeping with the ascendance of molecular techniques and investigations in
biology generally and is now reflected clearly in principal neuroscience
textbooks. Consider a single example. A decade ago, in the introduction to the
NEW WAVE REDUCTION TO NEW WAVE METASCIENCE
third edition of their monumental Princ iples of Neural Scien ce, Eric Kandel ,
Jame s Schwartz, and Thomas Jessell asserted the promi se of investigating the
molecular mechanisms of mind: "The goal of neural science is to understand
the mind , how we perceive, move , think , and remember. In the previous
editions of this book we stressed that important aspects of behavior could be
explained at the level of individual nerve cells. ... Now it is possible to
address these questions directly on the molecular level" (1991 , xii). Do notice
the first sentence. The ultimate explanandum of neuroscience is mind, not
some behavioral or ersatz laboratory substitute. By 1991, the search was
already on for its molecular mechanisms and their experimental verification,
to the extent that this focu s had already made the discipline's general
textbooks.
By the text 's recent fourth edition, and after another decade of
cellular and molecular research, these same authors were ready to announce
mind-to-molecules " linkages" not just as research promises, but rather as
accomplished results :
This book .. . describes how neural science is attempting to
link molecules to mind-how proteins responsible for the
activities of individual nerve cells are related to the
complexity of neural processes. Today it is possible to link
the molecular dynamics of individual nerve cells to representations of perceptual and motor acts in the brain and to
relate these internal mechanisms to observable behavior.
(2000 ,3-4)
The chain of explanations envisioned by the se authors is nothing less than a
reduction of mind to molecules, through interposed "cognitive" and cellular
levels. It should be noted explicitly that Principles of Neural Scien ce remains
the standard comprehensive textbook in the field .
Two lessons from these passages-and from similar passages that
occur in introductory chapters in most of neuroscience's current texts-are
.crucial. First, according to these prominent neuroscientists speaking with the
authority of textbook authors, some observable behaviors have already been
explained at the level of molecular mechanisms. Second, the guiding aim of
"mainstream" neuroscience is the discovery of these mind-to-molecules
"linkages." So by limiting attention to cognitive neuroscience only, by
ignoring the cellular and molecu lar core , philosophers and cognitive scientists
are getting off the neuroscience train before the current end of its explanatory
line. Throughout this book I will argue explicitly that techniques of cognitive
neuroscience are an essential part of discovering mind-to-molecules
" linkages." But some of what these techniques reveal has already been carried
"further down" to cellular, synaptic , and ultimately molecular biological

mechanisms; these existing reductions reveal an essentially heuristic role for
higher-level scientific investigations; and there is genuine empirical promise
that more "ruthless reductions" will be coming forth . Defending these
assertions is this book 's principal goal.
When one limits his or her attention to cognitive neuroscience, one
not only misses some of neuroscience's most celebrated recent results. One
also misses the core of the current discipline: the problems, methods, and
results that occupy the day-to-day work of the greatest percentage of the
28,000+ scientists who belong to the Society for Neuroscience and thereby
identify themselves professionally as neuroscientists. The Society is "the
largest professional society dedicated to the study of the nervous system."
Regular membership is open to "any scientist ... who has done meritorious
research relating to the nervous system" (www.sfn .org/mernb/factjsheet.
html), Its web site offers a searchable database of the abstracts from the
13,000+ slide talks and posters presented at the most recent annual meeting
(at this writing , of the 2001 annual meeting, November 4-11, in San Diego,
CA) (http://sfn.scholarone.com/, available as a link from www.sfn.org).This
resource is a gold mine for surveying "hot" areas in current neuroscience.
What is hot now? Searching for cellular and molecular mechanisms and
attempting to relate these mechanisms directly to observable behavior. For
example, if one conducts a standard search of this database on the two themes
most dominated by cellular physiological and molecular biological experimental techniques, namely "Development" and "Synaptic Transmission and
Excitability," searching for "Any subtheme," one receives 4698 abstracts
(respectively, 1818 and 2880) . On the other hand, if one searches the theme
most closely associated with cognitive neurosci entific techniques, namely
"Cognition and Behavior," one receives 1873 abstracts. But when one further
limits this last category to the subtheme, "Human cognition and behavior," the
number drops to 476 abstracts; and many of these are purely behavioral
studies that don't purport to offer "cognitive" explanations for the data
revealed. T his is anecdotal evidence, of course, but the upshot is clear. When
it comes to basic scientific, not-purely-clinical, research , the search for
cellular and molecular mechani sms dominates among bench neuroscientists.
Amateur sociology aside, there is another way that philosophers and
cognitive scientists become misinformed when they limit their attention to
cognitive neuroscience. Searching for mind-brain connections at the level of
regional neural activation patterns, neuropsychological measures, or activation vector s across neural networks isn't as "ruthlessly reductionistic" as
attempting to explain behavior-and manipulate it experimentally in animal
preparations-at the level of intracellular signaling pathways within and
between the individual neurons constituting the network. The "parts" are
smaller in the latter theories and there is intuitively a larger gap between
starting and ending "levels," between explananda and explanans. One will
thereby misunderstand the nature and extent of reductionism in current
neuroscience if one knows nothing about the research and theory now taking
place to forge mind-to-molecules "linkages."
So this book's first task is to reveal the scientific details of some
accomplished mind-to-molecules "linkages" and to evaluate the explanatory
potential of this "ruthless reductionism" for behavior and cognition generally.
Only by considering these details can one grasp the reductionism that
characterizes the core branch of current neuroscience. One might end up
disagreeing with my positive arguments for the explanatory potential of
current and foreseeable reductionist cellular and molecular neuroscience. To
tip my hand, I will argue for its explanatory potential all the way up to
features of consciousness (in Chapter Four). But no one should be mistaken
about the factual existence of a ruthless and audacious reductionism that
informs neuroscience's current cutting edge. If I can communicate that, I will
at least break the popular but mistaken myth among philosophers and
cognitive scientists that reductionism is "dead." On the contrary: it is alive
and thriving, at the very core of one of the hottest (and best funded) scientific
disciplines. Perhaps I can even convince some that this "ruthless reductionism" is the correct way to pursue a science of mind, given all we know
and can do now.
One final remark needs to be emphasized before we start doing
philosophy and neuroscience in earnest. I've already referred to current
cellular and molecular neuroscience as trendy, "hot," and "well-funded." It is
easy to misunderstand the significance I place on these features. I am well
aware that even many neuroscientists find these terms unseemly and are
squeamish about the sociology of their discipline that these terms reflect. (I
myself do not and am not, but that is a mere psychological fact about me that
has no bearing on the arguments in this book.) I am also well aware that there
are numerous (sociological) explanations for why this branch of the discipline
is so prominent, including funding opportunities for expensive research
equipment, novel (and very cool) technological toys to play around with , the
larger Zeitgeist that propels molecular biology in general at present-add your
personal favorite here. However, I am NOT committing a simple-minded
fallacy that equates a discipline's temporal popularity and external funding
trough with a beeline to Truth or the fruitfulness of its approach!
Instead, my appeal to the centrality of cellular and molecular
neuroscience within the discipline serves two purposes in setting up this
book's project (and in motivating particular steps in the argument later on) .
First, it stresses the fact that "naturalistic" philosophers of mind, committed to
the continuity of philosophical theses and arguments with neuroscientific
work (including some who describe their research as "philosophy of

neuroscience") have been focusing myopically and virtually without
exclu sion on work that constitutes a tiny island of that discipline; and given
the "higher level" location of that tiny island, these philosophers have not
seen how ruthlessly reductive that the bulk of neuroscience actually is circa
the turn of the 21 51 century. A central task of this book (beginning in Chapter
Two) is to demon strate explicitly this ruthle ss reductionism-rather than just
assert it, as I am doing now . Second, even though I acknowledged multiple
explanations for why an approach can become trendy, hot, and funding rich , I
am also convinced that in this particular case the "on the right track" reason is
among these. If you are convinced that the sociological factors exhaust the
reasons for cellular and molecular neuroscience's "hot" and "well-funded"
current status , then this book probably isn't for you. However, I am also well
aware of a demand to argue for this thesis, and that will consist in explaining
not just some results of current cellular and molecular neuroscience, but of the
experimental methods, their rationale, and the case for their expected
fruitfulness. It will also consist in showing how these scientific resources can
be put together to address important philosophical concerns. In other words ,
the rest of this book constitutes an argument for the scientific and the
philosophical legitimacy of neuroscience's cellular and molecular revolution.
For those who are squeamish about terms like "trendy," "hot," and "wellfunded ," or for those who assume when they see such terms that a simpleminded fallacy about the justifiability of an approach is on offer, ear-marking
this page might be in order, to remind themselves about the significance and
the challenge to argument that I place on these features.
2 BACKGROUND: THE INTERTHEORETICREDUCTION REFORMULATION OF THE MIND BODY

PROBLEM
Although the nature and extent of reduction in current cellular and
molecular neuroscience is unfamiliar to most philosophers of mind (and
cognitive scientists), the concept of scientific reduction is not. Theories of
intertheoretic reduction from philosophy of science have been mobilized and
criticized by philosophers of mind and psychology over the past four decades.
The project I am pursuing in this book is a self-conscious attempt to break an
impa sse in current philo sophy of mind over the importance of scientific
reduction for addressing "the mind-body problem." To set up my project, I
will first explain how philosophy of mind got into this impasse (an impas se
which includes my previous "new wave reductionism" in Bickle 1998). Then
I'll suggest how my new approach and its larger philosophical motivations
purport to break this impasse. Besides, this episode in recent philosophy is
interesting enough to warrant presenting to this book' s wider interdisciplinary
audience.'
An old philosophical question about "the relationship between mind
and body" lies at the bottom of any discu ssion of reduction in the
psychological and brain sciences. Neuroscientists have not missed this
foundation . Introductory chapters of general neuroscience textbooks routinely
begin with a quick overview of philosophical views about mind, usually
starting with Rene Descartes' interactive dualism of mental and physical
substances. Unfortunately, many of these discussions depart from
philosophy's history before the middle of the 20 th century, by the time that
substance dualism had lost almost all popularity with philosophers in the
"an alytic " tradition. From introductory chapters in neuroscience texts, one can
obtain the mistaken impression that materialist or physicalist views about
mind barely exist in philosophy? Paul Churchland's masterful (1987)
introductory philosophy of mind text is a welcome panacea. After presenting
standard arguments against dualist views of mind , Churchland writes:
"Arguments like these have moved most (but not all) of the professional
community to embrace some form of materialism" (1987 , p. 21) . That
"professional community" includes many philosophers in the "analytic"
tradition.
How does an interest in scientific reduction fit into recent philosophy
of mind ? The traditional philosophical mind-body problem is ontological.
"Ontology" is a piece of fancy Greek philosophical jargon for the study of
being, of "what there is," of the fundamental constituents and categories of
existence. The mind-body problem is about the fundamental nature of the
conscious and cognitive mind and its relation to physical events. At bottom, it
is a clash between two conflicting intuitions:
1. That the nature and core properties of mental phenomena guarantee
that they cannot be identical to physical (i.e. , neural) events ;
2. That the domain of the mental should ultimately be brought within the
scope of our otherwise comprehensive and wholly physical scientific
worldview.
Physicalism amounts to giving pride of place to intuition #2.
But how can one argue for this view? How can one support such an
account as more than "blind faith" or "mere opinion"? One way, fashionable
for nearly four decades now, is to construe intuition #1 as resting on
allegiance to a psychological theory whose central theoretical posits (beli efs
and desires, or propositional attitudes generally) figur e in generalizations that
explain and predict observable behavior. The "nature and core properties of
the mental," paraded as "conceptually autonomous" by defenders of intuition
# 1, are said to depend upon this theory's constitutive principles and
generalizations. One can then reformulate the traditional philosophical mindbody problem as, first and foremost, a question about the intertheoretic
relationships obtaining between this psychology and its counterparts or
successors from the physical sciences (like neuroscience). Ontological
conclusions become secondary to and dependent upon the nature of the crosstheory relations that obtain. The ontology of mind is thus treated on a par with
similar cross-theoretic conclusions in science-like, e.g., the relationship
between temperature and mean molecular kinetic energy. Historically in
science, intertheoretic reduction has been the key relation thought to yield
cross-theoretic conclusions about the entities and properties posited by
potentially reducible theories. Depending upon whether or how an
intertheoretic reduction goes, scientists draw conclusions ranging from
the autonomy of a theory's objects and properties (e.g., electrical

charge is not identical to any combination of an object's mechanical
properties),
cross-theoretic identity (e.g., visible light is electromagnetic radiation
with wavelength between .35 and .75flm),
conceptual revision (e.g., mass is a two-place relation between an
object and countless reference frames, not a one-place property of
objects), or
elimination (e.g., there is no such thing as phlogiston).
It has also been fashionable among some physicalists to claim that

something called "folk psychology" provides the theory with which our
familiar ontology of mind is affiliated. Folk psychology is supposed to be the
rough-and-ready patchwork of generalizations that underwrite our everyday
explanations and predictions of each others ' behavior. Its principal theoretical
posits are beliefs and desires, which mediate between sensory inputs and
behavioral outputs.' There is, however, no need to saddle our ontology of
mind with "folk psychology." One can give our mentalistic categories and
kinds all the resources and sophistication of scientific psychology and ask
how those theories are relating to developing neuroscientific counterparts.
Friends of the mental can help themselves to any psychological theory they
see as fit to champion, be it commonsensical or sophisticatedly scientific."
Call this approach the Intertheoretic Relation (lR) reformulation of
the mind-body problem. Its guiding hope is that by reorienting the traditional
philosophical issue away from its ontological focus, making ontological
conclusions secondary to the prior intertheoretic reduction question, we might
overcome the deadlock over intuitions # 1 and #2 in a way that makes relevant
some of the rich and rigorous resources of zo" century philosophy of science
and the contemporary cognitive and brain sciences. Scientific clarity hopefully emerges from philosophical murk.
So it is not surprising that soon after Ernest Nagel (1961, chapter 11)
published his groundbreaking theory of intertheoretic reduction, it was
adopted by physicalist philosophers of mind. Physicalists especially liked his
linking the "temperature-is-mean molecular kinetic energy" identity to the
intertheoretic reduction of classical thermodynamics to statistical mechanics.
Nagel's resources became so prevalent that critics took themselves to be
attacking the entire psychoneural reductionist program by pointing out
difficulties that "special sciences" (like economics and psychology) pose for
Nagel's account. For example, in footnote 2 of his ([1974] 1981) Jerry Fodor
asserted-without argument-that "many of the liberalized versions of reductionism suffer from the same base defect as what I shall take to be the classic
form of the doctrine" ((1974] 1981, 322). The "classic form" was Nagel's
account, which had been published only thirteen years prior to Fodor's essay.
(Some works become "classics" very quickly!)
In the spirit of logical empiricism-the paradigm of mid-20 th century
philosophy of science to which he adhered and partly defined-Nagel's
theory of intertheoretic reduction purported to be completely general. It
purported to apply to every specific case in science and its history. It held that
the reduction of one theory to another consists of a logical deduction
(derivation) of the laws or principles of the reduced theory T R from the laws
or principles of the reducing or basic theory T B. In interesting cases
("heterogeneous cases," as Nagel called them), Tn's descriptive vocabulary
lacks terms contained in T R. For example, "pressure" and "temperature" do
not occur in statistical mechanics or microphysics. The logical derivation in
such cases also requires various "correspondence rules" or "bridge principles"
(BP) to connect the disparate vocabularies, at least one for each term of T R
not included in Tn. (We'll hear more about BPs in the next section.)
Eschewing niceties and many details, we can represent Nagel's account as
follows:
T B & BP (as necessary)
logically entails
TR.
Also in the spirit of logical empiricism, Nagel characterized T R, T B, and BP
"syntactically," as sets of statements or propositions distinguished by their
logical form.
10
Reformulated in light of Nagel's theory , the traditional philosophical

mind-body problem becomes: Will future brain science develop theories (Tns)
from which, with appropriate BPs, the generalizations of psychological
theories (T RS) are logically derivable? In light of this reformulation,
physicalism asserts that theories from physical science will occur at the end of
chains of intertheoretic reductions, from psychology down to e.g., neuroscience (with intermediaries possibly in between). And each link in this chain
must meet the demands that Nagel's account places on the intertheoretic
reduction relation, namely , logical derivation with the help of bridge
principles to connect the disparate vocabularies (in these obviously
heterogeneous cases.)
What does the Nagel-supplemented IR reformulation accomplish?
What do we achieve by reformulating the traditional mind-body problem in
this fashion? First, we replace the murky notion of "ontological reduction"
with intertheoretic reduction, a rigorous, scientifically grounded alternative.
Physicalists now can appeal to examples of reduction from the history of
science to illuminate and compare the nature of purported relations between
theories from psychology and the physical sciences. Taking intertheoretic
reduction as the central issue provides clear and justified verdicts about the
variety of philosophical arguments brought to bear on the traditional mindbody problem. Evidence and arguments relevant to deciding for or against
predicted future intertheoretic reductions in science are legitimate; evidence
and arguments irrelevant to these issues are not.' This methodological lesson
was taken to be significant by scientifically inspired philosophers. Philosophical questions about mind could finally be addressed from the perspective of a
rigorous philosophy of science.
3 REVOLTS AGAINST NAGEL'S ACCOUNT

3.1 "Radical" Empiricists (and Patrick Suppes)
Unfortunately, just as philosophers of mind began applying Nagel's
theory of intertheoretic reduction to the mind-body problem, that theory came
under decisive attack within the philosophy of science. Constructed within the
logical empiricist program, Nagel 's theory incorporated that program's
strengths and weaknesses qua philosophy of science. One weakness was
stressed increasingly throughout the 1960s: its assumption about the
continuity of scientific progress. Reduction as deduction of reduced theory T R
from reducing theory Tn reflects this assumption. Propositional logic requires
that if some principles of T R (the conclusion of the logical derivation) are
11
false, then at least one of the premises must be false. Careful historical
analysis-an emerging cottage industry throughout 1960's philosophy of
science-of some "textbook" scientific reductions revealed that principles of
T RS are often false. Falling bodies near the surface of the earth do not exhibit
uniform vertical acceleration over any finite interval. Yet this assumed
uniformity is central to Galilean physics. Galilean physics is empirically false.
It does not describe correctly the behavior of falling objects in any portion of
the actual world. Yet the reduction of Galilean physics to Newtonian
mechanics is a "textbook" historical case. On Nagels' account, however, this
requires some premise of the deduction to be false, some law or generalization
of Newtonian mechanics or some bridge principle linking a term unique to
Galilean physics to some Newtonian counterpart. This requirement stands in
contradiction to their assumed truth (at least at the time the reduction was
achieved).
As careful history of science flourished , philosophers noticed that
even the case that Nagel used to illustrate his theory turned out to involve a
significantly false reduced T R. The equilibrium thermodynamics-to-statistical
mechanics and microphysics reduction is actually a limit reduction, and the
limits in which the laws of equilibrium thermodynamics can be derived are
never actually realized (e.g. , an infinite number of gas particles whose
diameter divided by the average distance between particles is only negligibly
greater than zero). At best, equilibrium thermodynamical explanations
approximate the actual microphysical events and their statistical distributions.
Second, many key thermodynamical concepts fragment into distinct statistical
mechanical/microphysical concepts. Clifford Hooker (1981) nicely demonstrates this point for 'entropy.' Third , a diachronic view of this case (its
development over actual historical time) reveals mutual feedback between
reduced and reducing theories. Problems confronting classical thermodynamics (the reduced theory T R ) spurred the application and development of
statistical approaches. And the "injection" of statistical mechanics (part of the
reducing theory T B) back into classical thermodynamics resulted in more
accurate predictions."
What consequences do these features have for the explanations,
categories, and theoretical posits of classical thermodynamics? Hooker is
explicit: "In a fairly strong sense thermodynamics is simply conceptually and
empirically wrong and must be replaced" (1981, 49) . This quote reflects one
important criticism of Nagel's account of reduction. Intertheoretic reductions
in actual science typically correct the reduced T RS. Beyond a point, these
corrections make untenable the "reduction as deduction of the T R" account.
Emerging logical and historical problems for Nagel's account spurred
a number of alternative theories of intertheoretic reduction. Yet even before
these problems became prominent, and spurred by some general criticisms of
12
the underlying logical empiricist account of theory structure, Patrick Suppes

(1956, chapter 12, 1965) characterized scientific theories semantically, as sets
of models sharing some set-theoretic structure. In turn he characterized
intertheoretic reduction not as syntactic derivability but as set-theoretic
isomorphism (the formal analog of "sameness of structure"). Karl Popper
(1962), Paul Feyerabend (1962) and Thomas Kuhn (1962) all developed
independent accounts of reduction that insisted (or at least implied) that
besides explaining why the T R works in situations where it does , the T B must
also explain why the T R fails in other expected applications. This is their
sense in which the reducing theory corrects the reduced. Feyerabend (1962)
famously expressed this contention by denying that reduction involves
deduction in any capacity. Instead , he insisted that "incommensurability"
between T Rand T B and "ontological replacement"-a new account of the
ultimate constituents of the universe-were the central intertheoretic relations.
He went so far as to call for philosophers of science to abandon the search for
any formal or "objective" account of intertheoretic reduction or scientific
progress. Any attempt to characterize reduction or progress formally would
bastardize the actual history of scientific practice.
In keeping with the trend in 1960's philosophy of mind, Suppes and
Feyerabend both applied their alternative accounts of scientific reduction to
the reformulated mind-body problem. Suppes writes : "The thesis that psychology may be reduced to physiology would be for many people appropriately
established if one could show that for any model of a psychological theory it
was possible to construct an isomorphic model within physiological theory"
(1965,59). Feyerabend writes :
In the course of the progress of knowledge, we may have to
abandon a certain point of view and the meanings connected
with it-for example, if we are prepared to admit that the
mental connotations of mental terms may be spurious and in
need of replacement by a physical connotation according to
which mental events, such as pain , states of awareness, and
thoughts, are complex physical states of either the brain or the
central nervous system, or perhaps the whole organism.
(1962,30)
He advocated this view for "all so-called mental states." Armed with
Feyerabend's account of scientific reduction, the philosopher of mind's
agenda is to "develop a materialistic theory of human beings." Such a result
would "force us to abandon the mental connotations of the mental terms, and
we shall have to replace them with physical connotations" (1962, 90). With
13
Feyerabend's account of scientific reduction, "eliminative materialism"

received its first serious expression and defense.
Eliminative materialism remains deeply controversial in philosophy
of mind. It has few current adherents. This current status makes it enlightening to look back at the writings of some famous "central state" or "mindbrain identity" theorists throughout the mid-1960s. As their arguments came
under attack, Feyerabend's early eliminativism and the "radical empiricist"
philosophy of science that underwrote it looked increasingly attractive. For
example, just four years after his influential (1959) paper that brought the
mind-brain identity theory to a wider philosophical audience, J.J.c. Smart
claims to be
attracted to P.K. Feyerabends contention that in defending
materialism we do not need to show its consistency with
ordinary [psychological] language, any more than in
defending the general theory of relativity we need to show its
consistency with Newtonian theory. ... Feyerabend is perhaps
therefore right in arguing that the scientific concept of pain
does not need to be (and indeed should not be) even
extensionally equivalent with ordinary language. (1963, 660)
Four years later Smart clarified his (cautious) change in view. He admitted to
being even closer to Feyerabend, both in philosophical methodology and
eliminativist conclusion. In an attempt to stave off an "ordinary language"
criticism of his famous "topic-neutral translation" argument for the mindbrain identity theory, Smart writes:
I am even doubtful now whether it is necessary to give a
physicalist analysis of sensation reports. Paul Feyerabend
may be right in his contention that common sense is
inevitably dualistic, and that common sense introspective
reports are couched in a framework of a dualistic conceptual
scheme.... In view of Bradley's criticisms of my translational
form of the identity thesis, I suspect that I shall have to go
over to a more Feyerabendian position. (1967 , 91)
Smart was not the only famous identity theorist attracted to
Feyerabend's philosophy of science and eliminative materialism. In a
postscript written ten years after his famous essay defending an identity
theory, Herbert Feigl writes : "I now agree with Smart (and perhaps with
Feyerabend) that within the conceptual frame of theoretical natural science
genuinely phenomenal (raw feel) terms have no place " (1967, 141). He cites a
14
scientific analogy that became prevalent in later eliminativist writings : the

prop erties of common sense physical object s vis-a-vis their "successor
concepts" from microphysics. He concludes that "the phenomenal predicates
used in the description of after-images, sensations, feelings, emotions, moods ,
etc., are to be replaced by the (as yet only sketchily known) neurophysiological and ultimately microphysical characterizations" (1967 , 141-142; my
emphasis). Thus there was a time, not too long ago, when influential
physicalist philosophers of mind did not reject eliminative materialism out of
hand, or the theory of reduction that supported it. It is interesting that these
writings have been lost among recent philosophers of mind , who do tend to
reject eliminative materialism out of hand.
These shifts by famous mind-brain identity theorists toward
Feyerabend's philosophy of science (theory of scientific reduction) and
eliminative materiali sm hold two lessons for us. First, they demonstrate how
intractably dualistic our psychological concepts appear to be. It is not just
extremely difficult to find a meaning preserving "translation" for mental terms
within a physical language. It is probably impossible. So if we aspire to an
explanation of human behavior in physical (i.e., neuroscientific) terms, we
might do well to leave psychological conceptions to one side and not let them
bias our views about what a successful neuroscience must look like or answer
to. Second, these capitulations by identity theori sts to Feyerabend show once
again how attracted physicalist philosophers of mind have been to resources
from the philosophy of science. Every account of intertheoretic reduction that
philosophers of science took seriously was subsequently adopted by philosophers of mind in attempts to reformulate the traditional mind-body problem.
Consensus in Anglo-American philosophy of science, however, was
less sympathetic to Feyerabend. Most found his accounts too radical, too
dismissive of precise, formal resources to illuminate scientific concepts and
historical episodes. Even so radical a critic of orthodox logical empiricism as
Thomas Kuhn (1962) maintained that the majority of scientists spend the
majority of their careers doing "normal science ," i.e., puzzle-solving within
an accepted paradigm. Parts of Kuhnian "normal science" sound like the
views advocated by logical empiricists. Even cases of wholesale theory
change or "scientific revolution" seem to approximate the formal interth eoretic relations proposed by Nagel and other logical empiricists. Feyerabend
dismissed the possibility of accounting for this sense of "approximation," but
many Anglo-American philo sophers of science proceed ed on the assumption
that something like it could be clarified. At this point, the intertheoretic
reduction literature in the philosophy of science diverged in multiple
directions. I am going to trace only one of these: the search for accounts that
incorporated weakened Nagel ian conditions designed to capture features of
scientific histor y and practice emphasized by Feyerabend, Kuhn and other
15
radical empiricists. I am fully aware that other themes were stressed in the
reduction literature, by luminaries like Lawrence Sklar, Thomas Nickles,
William Wimsett, and Robert Causey. I leave it to fans of these philosophers
to compare and contrast features of psychology's relation to neuroscience
built on one of these alternatives, to the one I am about to develop.
3.2 Schaffner's General Reduction (-Replacement) Paradigm

Kenneth Schaffner's (1967) General Reduction Paradigm, later
developed more fully and renamed the General Reduction-Replacement
(GRR) model (Schaffner 1992), was the first explicit attempt to conciliate
these features. Schaffner's model includes conditions of intertheoretic
connectability and derivability (of reduced theory T R from reducing theory
T B ) that yield Nagel's exact conditions as a special case. But it also includes
"corrected" versions of reducing and reduced theories (T B * and T R *, respec tively). This weakens the general conditions, enabling connectability and
derivability to hold between T B* and T R* in cases where (actual) T B corrects
(actual) T R by making more accurate predictions in the latter's domain of
application (Schaffner 1992, 321). Furthermore, T R and T R * stand in a
relation of "strong analogy." Hence since corrected T R* is derivable from T B
(or T B *) (Schaffner's weakened condition of derivability), the reducing theory
indicates why the reduced "worked as well as it did historically" and explains
the reduced theory's domain "even when T R is replaced " (Schaffner 1992,
321). These weakened notions thus "allow the 'continuum' ranging from
reduction as subsumption to reduction as explanation of the experimental
domain of the replaced theory" (Schaffner 1992, 320) . Cases that closely
approximate Nagel's conditions group around the first pole; cases with
features emphasized by the radical empiricists group around the second. Both
orthodox logical and radical empiricist intuitions are thereby accommodated.
(However, it must be noted explicitly that Schaffner has yet to explicate the
relation of "strong analogy" between corrected T R* and actual T R.)
One of Schaffner's arguments in support of his GRR model will
sound familiar. He writes: "The flexibility of the GRR model is particularly
useful in connection with discussions concerning current theories that may
explain 'mental' phenomena" (1992, 320). The IR reformulation of the mindbody problem lives on! Schaffner shows in some detail how his model applies
to an example of (potential) reduction of actual psychology to neurobiology,
describing the cellular mechanisms of short-term and long-term learning as
revealed by studies on the sea slug, Aplysia californica (1992, 323-329).7
Although the cellular explanations Schaffner discusses are now somewhat
dated (we'll take up recent discoveries in Chapter Three), the lessons he
16
stresses for psychoneural reductions remain topical and directly relevant for
my current concern. According to Schaffner (1992) , the following are key
features of this specific case:
Reduced and reducing theories do not involve laws akin to those in

"textbook" cases of reduction from physics (329).
The reducing complex is an intricate system of causal generalizations
with varying scopes of applicability (from nervous systems in general
to specific types of neural processes). These generalizations are not
framed within the vocabulary of one specific science (e.g.,
biochemistry), but rather are characteristically interlevel (containing
terms from , e.g., biochemistry, molecular biology, cellular
neurophysiology, neuroanatomy, and behavioral psychology) (330) .8
When a phenomenon described at one level (e.g., the behavioral, as
"sensitization") gets explained in lower-level terms (e.g., cellular
mechanisms), the former description is mapped into the lower-level
vocabulary (330-331).
It is by virtue of this last feature that Schaffner's GRR conditions of

connectability and derivability are generalizations of Nagel's. At first glance,
Schaffner's GRR model appears supple enough to handle the special
complexities and details that actual psychology-to-neurobiology reductions
generate; yet it retains Nagelian-inspired conditions on the intertheoretic
reduction relation. And it achieves this consilience using an example that
emphasizes actual scientific details (at least as they stood in the late 1980s) far
beyond the extent that is typical in philosophy of science or mind .
3.3 Hooker's General Theory of Reduction

Clifford Hooker (1981) offers another account that amounts (in part)
to a weakened set of Nagel-inspired conditions. Like Nagel and Schaffner,
Hooker insists that intertheoretic reduction involves deduction, with the
reducing theory Tn serving among the premises. But unlike Nagel, the
conclusion of the derivation is not the reduced theory T R; and unlike
Schaffner, it is not a corrected version T R* of T R. Instead , what gets deduced
is an image In of T R, specified within the conceptual framework and
vocabulary of the reducing theory Tn. In matches the domain of application
and explanatory scope of T R and the generalizations comprising the former
mimic the logical (syntactic) structure of the latter . Simplifying, and ignoring
some niceties, we can express Hooker's account with the following
schematic, which I adapt with minor changes from Paul Churchland (1985):
17
Tn (& boundary conditions, limiting assumptions, as needed)

logically entails
In (a set oftheorems of [restricted] Tn)
e.g. , (x)(Ax -7 Bx), (x)((Bx & Cx) -7 Dx)
which is relevantly isomorphic to ("analogous to")
TR
e.g. , (x)(Jx -7 Kx), (x)((Kx & Lx) -7 Mx)

The schema is meant only to illustrate the "analog relation" between
In and T R. It is not intended to prov ide or even ground an analysis of the
relation (which, incidentally, neither Hooker nor Churchland has ever
provided). "Boundary conditions" and "limiting assumptions" restrict the
applicability of the Tn 's generalizations so as to isolate falsity in the T R.
Consider the reduction of false Galilean physics to Newtonian mechanics
(presumed true at the time of the reduction). We can conjoin with the
Newtonian principles either a counterfactual assumption describing conditions
near the surface of the earth that permit uniform vertical accelerations over a
finite interval, or a counterfactual assumption that limits the applicability of
Newton's laws to moving bodies that only fall distances negligibly greater
than zero. From this reducing complex-Newtonian mechanics Tn and the
counterfactual boundary condition or limiting assumption-the image (In)
that mimics the expl anatory scope of Galilean physics (T R) can be derived.
The falsity of the latter is explained by and hence confined to the
counterfactual boundary condition or limiting assumption.
It is important not to confuse Hooker's deduced image In with
Schaffner's corrected version of the reduced theory T R *. Hooker's In is
characterized completely within the fram ework and vocabulary of Tn;
Schaffner's T R * is a corrected version of T R , and so is characterized (at least
partly) out of the resources and vocabulary of the reduced theory. This
difference yields the very different ways that Hooker and Schaffner attempt to
capture radical empiricist insights within a modified Nagelian account of
intertheoretic reduction. Every topic discussed in the remainder of this section
contrasting the two accounts hinges on this difference."
Hooker readily acknowledges the radical empiricist insights built into
his account about science's actual history and progress. But his guiding
intuition about intertheoretic reduction is explicitly Nagelian: "W hile the
construction of In within Tn might be a complicated affair-[boundary
conditions] might be fearfully complex (cf. biological reductions), counterfactual (e.g ., assume continuity), nece ssarily counterfactual qua realization
(e.g., "force free"), and so on-the ultimate relation between Tn and In
remains straightforward deduction" (1981, 49). Even his justification of this
18
feature is Nagelian: "T B continues to directly explain I B and this is the basis
for TB's indirect explanation of TR's erstwhile scientific role" (ibid.) .
According to Hooker, deduction is necessary to capture the explanatory unity
of T R and T B, which remains an explicit condition of his account. This is a
condition he shares with logical empiricists like Nagel.
Noti ce that , unlike either Nagel 's or Schaffner's account, the premises
of the deduction partly constituting a Hooker reduction do not contain bridge
principl es or correspondence rules. These are not needed. Image I B is already
specified within (a restricted portion of) the T B. There are no disparate
vocabularies to bridge across premises and conclusion in the deductive component. Structures analogous in some ways to Nagelian bridge principles
appear in a second stage of a Hooker reduction, involving I B and the T R. But
these components are only ordered pairs of terms that indicate the substitutions in I B that yield the actual generalizations of T R, or approximations of
the actual generalizations if that is all that a given case permits. By themselves, these ordered pairs imply neither synonymy (sameness of meaning)
nor coextension (sameness of reference) of terms , nor material identity. Thus
one central difficulty with Nagel's account vanishes: that of specifying the
logical status of bridge principles in reductions that falsify the T R.
Earlier in this section we saw that Schaffner's (1992) generalizations
of Nagel 's conditions of connectability and derivability yielded a spectrum of
possible reduction outcomes, ranging from ones in which Nagel's actual conditions are closely approximated to others displaying features emphasized by
radical empiricists. Hooker's account yields a similar spectrum, ranging from
"relatively smooth" to "extremely bumpy" intertheoretic reductions. A case's
location on this spectrum depends on the "amount of correction" implied to
the TR, which in turn amounts to the "closeness of fit" obtaining between the
derived image I B and it. Cases approximating Nagel ' s conditions fall near the
"smooth" endpoint. (However, it remains crucial to bear in mind that on
Hooker's account, the T R itself is never deduced, not even in the "smoothest"
cases . Rather, it is always the target of a kind of complex mimicry.) The
derived I B mimics TR's explanatory scope in the latter's domain of application, is strongly analogous in logical structure to the TR, and its derivation
from the reducing theory TBrequires few counter-to-fact boundary conditions
or limiting assumptions. Historically, the physical optics-to-electromagnetism
reduction seems to reflect these conditions. Cases involving features emphasized by radical empirici sts fall toward the bumpy endpoint. Here an I B only
weakly analogous to T R can be derived from TB, and this only with the help
of numerous and wildly counterfactual boundary conditions and limiting
assumptions. Historically, the phlogi ston theory-to-oxidative chemistry reduction seems to reflect these conditions. "Mixed" reductions sharing some
features of both extremes fall on the spectrum separating these two endpoints.
19
Ambiguous historical cases for Nagel's logical ernpmcist account, like the
equilibrium thermodynamics-to-statistical mechanics and microphysics
reduction, seem to reflect these conditions.
In short, a case's location on Hooker's intertheoretic reduction
spectrum depends on the "amount of correction" implied to the reduced
theory T R, captured in two conditions: the strength of analogy between the
syntactic structures of the general izations comprising I B and T R, and the
number and counterfactual nature of the boundary conditions and limiting
assumptions necessary to derive such an I B from T B.
If we are to employ Hooker's account to reformulate the philosophical mind-body problem, then it must also provide some account of when
cross-theoretic identities are justified. Otherwise the ontological aspect of the
traditional problem cannot be captured. Unlike Nagel's and Schaffner's
accounts, Hooker's doesn 't employ cross-theoretic bridge principles BP, the
obvious resource for providing this. Instead, a reduction's relative smoothness
justifies cross-theoretic identities. For not only do historical intertheoretic
reductions line up on a spectrum; so do the cross-theoretic ontological
consequences drawn from them. The latter range from entity and property/
event identities (visible light is electromagnetic radiation with wavelength
between 0.35-0.75 urn) to significant conceptual revision (temperature T of a
gas is only identical to its mean molecular kinetic energy in an empirically
unrealizable mathematical limit) to outright elimination (there is no such thing
as phlogiston). When we layout the location of historical reductions along
these two spectra-the intertheoretic reduction "amount of correction"
spectrum and the ontological consequences spectrum-we find a rough
isomorphism. A case's location on the intertheoretic reduction spectrum
correlates closely with its location on the ontological consequences
spectrum. 10
This observation suggests a strategy for predicting the cross-theoretic
consequences of a developing or potential intertheoretic reduction. First discover where on the intertheoretic reduction spectrum the case appears to be
headed, in terms of the "amount of correction" being implied to T R. How
equally explanatory and structurally analogous to T R is an image I B derivable
within TB? How numerous and wildly counterfactual are the boundary
conditions or limiting assumptions needed to effect the derivation? Which
historical reductions does the case seem most closely to resemble in these
respects? Answers will locate the developing or potential case on Hooker's
intertheoretic reduction spectrum. The predicted cross-theoretic ontological
conclusions (identity, revision, elimination) will then be those obtaining at the
roughly isomorphic location on the other spectrum. The isomorphism across
the two spectra that grounds this strategy is inspired directly by actual
reductions from the history of science. Even the "autonomist" or anti-
20
reductionist about 1'R and its posits can be satisfied. His or her position
predicts that the reduction relation won 't obtain, i.e., that no appropriate I B
will be derived within 1'B that is equipotent to 1'R .
Notice finally that the spaces between the "smooth" and "bumpy"
endpoints on the intertheoretic reduction spectrum and the "retention" and
"replacement" endpoints on the ontological consequences spectrum provide
for the possibility of "revisionary" results (Bickle 1992b, 1998 , chapter 6). A
variety of historical scientific reductions serve as useful precedents: e.g.,
classical thermodynamics-to-statistical mechanics and microphysics and
classical mechanics-to-general relativity theory . In both cases, the
generalizations of 1'R at best are approximated by those of the equipotent
image I B derivable within T B. Key explanatory posits of 1' R fragment into a
variety of different posits of T B in different explanatory contexts. And mutual
co-evolutionary interaction between 1'R and T B increases the explanatory
scopes of both . Concerning the theoretical posits of psychology (the reduced
theory 1' R in the envisioned psychoneural reductions), revisionary physicalism
predicts enough conceptual change to rule out cross-theoretic identities with
neurophysiological posits. It differs in this way from a mind-brain identity
theory. However, revisionary physicalism also denies that psychological kinds
will undergo the radical elimination that befell, e.g., phlogiston and caloric
fluid. Instead, one group of cognitive representation concepts (the kinds
employed in psychological explanations of behavior) will be replaced by a
different group of cognitive representation concepts (the kinds emerging from
developing neuroscience). Exactly this type of result obtains in historical
revisionary reductions. Relativity theory still posits length, mass, velocity
concepts, just not the specific ones of classical mechanics. If revisionary
psychology-to -neuroscience reductions obtain, these will yield enough
conceptual change to rule out strict cross-theoretic psychoneural identities.
But they will not yield wholesale elimination of psychological kinds akin to
the caloric fluid/phlogiston variety.' :
An IR reformulation of the traditional philosophical mind-body
problem grounded on Hooker's theory of intertheoretic reduction looks
promising. Perhaps it affords the best resources for articulating physicalism
and defending it against classic philosophical objections. This remains to be
seen, however, because Hooker's account (as presented so far) faces three
outstanding problems . First, Hooker himself nowhere applies his account to
detailed potential reductions of actual psychological to neuroscientific
theories. Second, we've seen nothing in a Hooker-inspired IR reformulation
(as presented so far) that addresses the most influential philosophical criticism
of psychophysical reduction, the " multiple realizability" argument. Third,
Hooker's theory of reduction is subject to two serious criticisms from within
the philosophy of science. It is hand waving about detailed applications to
21
historical cases of scientific redu ction , leaving the key concept of an image In
and the "analog relation" between In and T R without a clear illustration. And
as Hooker (1981) himself admits, his "analog relat ion" lacks precise
formulation. "New wave" psychoneural reduction seeks to redress these
shortcomings (Bickle 1998) .
4 EXTENDING HOOKER'S INSIGHTS: NEW WAVE

REDUCTION
4.1 Handling Multiple Realizability
As physicalists about mind began adopting intertheoretic reduction,
Hilary Putnam and Jerry Fodor (among others) began emphasizing the
problem of multiple realizability. Putnam published a number of papers on
this topic throughout the 1960s. 12 Fodor ([1974] 1981; 1975, chapter 1)
extended these arguments. They contend that a given mental type (property,
state, event) is, or at least can be, realized by a variety of distinct physical
kinds sharing nothing significant at the level of physical description ; and,
more controversially, that this fact spells doom for psychoneural reduction
(and for psychophysical reduction generally). Putnam' s favorite example was
pain. The same pain state seems ascribable to creatures with very different
nervous systems: humans, rats, octopi, etc. Perhaps the same pain state can be
ascribed even to beings lacking earthly nervous systems, whose physical
mechanisms differ completely: silicon-based space aliens, appropriately programmed digital computers, etc. But then any postulated type-identity or
"reduction" of pain to a single one of its multiple physical realizers is false. 13
The multiple real izability argument remains central to non reductive
physicalism, the orthodoxy in current philo sophy of mind (see , e.g., LePore
and Loewer 1989, Horgan 1993) . If reductionism is to be a live option, the
multiple realizability argument must be addressed. Can Hooker's account of
intertheoretic reduction help?
Potentially, yes. Mo st initial replies to Putnam and Fodor appealed
explicitly to resources from theories of intertheoretic reduction . Robert
Richardson (1979) pointed out that Nagel (1961) himself countenanced
merel y conditional bridge principles. Although Nagel 's detailed examples
employed biconditional ("if and only if") bridge principles, all the
"connectability" that his condition of derivability requires is one-way
conditionals: For all x, if 13x then Rx, where '13 x' is a predicate of the
reducing theory Tn and ' Rx' is a predicate of the reduced theory T R , e.g ., if x
is in brain state b then x is in psychological state p. Conditional bridge
22
principles are consistent with the multiple realizability of T R posits within T B.

Multiple realizability only nixes conditionals in the other direction: if Rx then
Bx, e.g., if x is in psychological state p then x is in brain state b. So even
Nagel 's "classical" theory of reduction can handle the multiple realizability of
reduced kinds.
Another popular reductionist reply rests upon an insight first noted by
David Lewis (1969). Intertheoretic reductions are typically domain-specific.
Lewis himself offered a common sense, non-scientific example to illustrate
his observation, but Berent Enc (1983) and Patricia Churchland (1986),
among others, have pointed out that domain specificity obtains in the
"textbook" reduction of classical thermodynamics to statistical mechanics and
microphysics. Temperature in a gas is mean molecular kinetic energy.
Temperature in a solid is a different statistical mechanics/microphysical
property, mean maximal molecular kinetic energy, because the molecules in a
solid are bound up in lattice structures and restricted to a range of vibratory
motions. And so on for other physical states (e.g., plasmas). Temperature is
multiply realized in distinct statistical mechanical/microphysical states, and
yet it is a central reduced kind in a paradigm intertheoretic reduction from the
history of science. Clearly, multiple realizability by itself is not sufficient to
block a scientific reduction-unless one is willing, on philosophical grounds
alone, to part with scientific practice and deny that classical thermodynamicsto-statistical mechanics and microphysics is a "genuine" intertheoretic reduction. That move will strike many scientifically inspired philosophers as
privileging prior epistemological commitments over the endeavor one seeks to
understand. Jaegwon Kim (1993) even builds this domain specificity directly
into his concept of reductive bridge principles. In "local reductions" the crosstheoretic bridge principles have the form, "For all x, if Sx, then Bx if and only
if Rx," where 'Sx' is a predicate denoting a type of structure, e.g., if x is homo
sapiens, then x is in brain state b if and only if x is in psychological state p.
Multiple realizability only implies that different 'Bx' s will occur in the
embedded biconditional for different structure types, but this is consistent
with structure type-specific "local reductions." According to Kim, local
reductions "are the rule rather than the exception in all of science, not just in
psychology" and are "reductions enough ... by any reasonable scientific
standard and in their philosophical implications" (1993, 257) .
However, this strategy does not handle all types of multiple
realizability. Since Fodor ([1974] 1981), psychophysical anti-reductionists
have emphasized a more radical sense of the relation. Call the sense introduced by Putnam "multiple realizability across physical structure types," in
that distinct type s of physical structures are said to realize a given mental kind
differently. The domain specificity reply disarms an anti-reductionist argu ment employing this sense . The multiple realizability premise might be true,
23
but the anti-reductionist conclusion does not follow validly from it. But now
consider "multiple realizability within a token system across times," in which
a single instance of a cognitive system realizes a given mental type in
different physical states at different times." The plasticity of mammalian
brains-in responding to trauma, changing task demands, developmental
processes, and the neural mechanisms of learning-suggests that this more
radical sense is genuine. Ned Block (1978) once suggested that narrowing the
scope of psychological generalizations via domain specificity to handle
Putnam's sense of multiple realizability would render comparative psychology across species problematic, a seemingly legitimate endeavor. (Not to
mention routine methodologies in experimental psychology and neuroscience
using animal models of human capacities!). Block's point takes on added
urgency when we consider that a psychology narrowed enough in its scope of
applicability by domain specificity to handle the more radical sense of
multiple realizability would contain generalizations applicable only to individuals at times. Surely that much domain specificity would render psychology
insufficient to accommodate even the most minimal conditions on the generality of science.
However, actual scientific practice and its history give reductionists
ammunition against anti-reductionist arguments built on this more radical
sense . Berent En<; (1983) and I (Bickle 1992a, 1998, chapter 4) have both
suggested that this radical sense is present in historical cases of intertheoretic
reduction, including the paradigm classical thermodynamics-to-statistical
mechanics and microphysics case. For any token aggregate of gas molecules,
there is an indefinite number of microphysical realizations of a given
temperature-a given mean molecular kinetic energy. So even this radical
sense by itself appears not to block an intertheoretic reduction. Pertaining
specifically to psychoneural reductions, I've argued that this radical sense is
emerging in the potential reduction of propositional attitude psychology to
connectionist theories of representational content (Bickle 1995a, 1998,
chapter 4). These replies again grant the multiple realizability premise
(interpreted in the more radical sense) and challenge the validity of the antireductionist conclusion said to follow from it. Their success remains
conditional on accepting the standard scientific interpretation of the thermodynamics-to-statistical mechanics and microphysics case as a genuine
intertheoretic reduction; but that antecedent does not seem problematic if we
let scientific practice be our guide. This is the reduction cited and explained in
numerous physical chemistry textbooks.
Hooker himself expresses a similar attitude toward multiple
realizability in his long paper on intertheoretic reduction, writing :
24

It is often argued that , e.g., cognitive psychology cannot be
reduced to neurophysiology because the former crossclassifies the latter: any number of different systems (from
brains to machin es to leprechauns passing notes) could
realize the same funct ional or compu tational theory . It helps
to remove the intellectual dazzle of this fact to realize that
this is true of any functionally characterized system. The
same cross -classifications turn up within such prosaic fields
as electrical engineering (cf. " is an amplifier of gain A" vis-avis particular circuit diagrams) and physics (cf. "is a high
energy electron source" vis-a-vis quantum specification).
(1981,505)
His last example clearly admits of the more radical sense of multiple
realizability. There are multiple distinct quantum realizations of the same
token high-energy electron source over time. What is required to handle
multiple realizability of psychological on phys ical kinds within a reductionist
account is a better understanding of intertheoretic reduction across all of
science. Hooker continues: "In these cases, the issue is not whether reduction
is possible, but how it goes. The same appli es, I hold, between theoretical
domains as well" (1981 , 505) . Any theory of intertheoretic reduction must
handle multiple realizability of reduced on reducing kinds , or that theory is
insufficient for science generally, not just for psychology and neuroscience.
To accommodate this feature, Hooker (1981 , Part III) supplements his general
theory of reduction with an account of "token-token" reduction. His account
builds multiple realizability (including, potentially, in token systems acro ss
times) directly into the reduction relation. He illustrates this addition with a
brief discuss ion of some features of the emerging reduction of tran smission to
molecular genetics (Hooker 1981, Part III). 15
Each reduct ionist reply I've scouted accepts the truth of the multiple
realizability premise and challenges the validity of the anti-reductionist
conclusion. But the truth of the premise has also been challenged, especially
the one employing the more radical sense . Regarding neural plasticity,
regained function is often compromised following serious neural damage.
Persons can still, e.g. , talk, manipulate spatial representations , or move their
limbs, but their performance is typically degraded. This fact gives rise to
tricky que stions about individuating mental types. Do these distinct pre- and
post-plasticity neural events realize the same mental kind ? This worry can be
thought of as a specific version of a general challenge first raised by Nick
Zangwell (1992) . He claims that multiple realizability across species or
structure types is "not proven " because no argument is ever given for the
identity of mental types acros s all cases.
NEW WAVE REDUCTION TO NEW WAVE MET ASCIENCE
25
Kim (1993, chapter 16) suggests and I (Bickle 1998, chapter 4)

elaborate that a guiding methodology in contemporary neuroscience assumes
continuity of underlying physical mechanisms, both within and across
individuals and species. This assumed continuity is more than mere analogy
and informs most experimental techniques, research paradigms, and theoretical conclusions. Special empirical techniques and statistical analyses exist to
control for idiosyncratic activity on individual trials, e.g., subtraction
techniques in neuroimaging (see Posner and Raichle 1994). If the radical
sense of multiple realizability really obtained to the degree stressed by antireductionists, the experimental techniques of contemporary neuroscience
should have borne little fruit. But clearly these techniques are effective and
not hopelessly naive. Why has a detailed study of the macaque's visual
system been so instructive for learning about the human 's? Why has positron
emission tomography (PET) and functional magnetic resonance imaging
(fMRI) revealed common areas of high metabolic activity across and within
individual humans, down to a current resolution of less than 1 mm and
promising to go lower as techniques and analytical tools improve?" All of
this is indirect evidence from lower level sciences that the kinds postulated by
psychological theories might not be as radically multiply realized as current
anti-reductionists imagine.
I now think that there is even direct evidence from recent cellular and
molecular neuroscience against the truth of the multiple realization premise
that does the real work in Putnam's argument and its legacy. Neural plasticity
itself turns out to be governed by common mechanisms across both
individuals and species. It follows a regular progression during development,
learning, and following damage to principal structures. Shared molecular
mechanisms subserve it in all its forms and across the widest of gaps among
biological species. I now think that this direct attack is the definitive reply to
this influential anti-reductionist argument. But this attack depends upon a
careful analysis of Putnam's original argument and its legacy, and upon
details from the cellular and molecular investigations I will describe in the
next two chapters. We thus won 't be in a position to see these empirical
implications for this philosophical issue until near the end of Chapter Three
below. For now, you'll just have to take my word that I will attempt to cash
this paragraph's promissory note.
This subsection is not intended to be a comprehensive review of the
multiple realizability literature, or a definitive case for rejecting this popular
anti-reductionist argument. Here I am only indicating how psychophysical
reductionists have borrowed resources from both theories of intertheoretic
reduction, Hooker's included, and empirical science to address this challenge.
Psychoneural reduction is not Dead On Arrival. Even granted this much,
however, we still confront a key problem. Is there an account of intertheoretic
26
reduction sufficient for science generally and capable of handling the special
complexities of potential psychology-to-neuroscience reductions? Or is an IR
reformulation of the mind -body problem ultimately a dead-end for this less
popular reason?
4.2 New Wave Reduction

So far I've stressed the attractions of Clifford Hooker's account of
intertheoretic reduction. Now it is time to examine its shortcomings. Although
Hooker (1981) mentions numerous historical cases when presenting his
theory, he never applies it to the quantitative details of any example. Nor does
he analyze or even illustrate his key concept of an image I B much beyond the
schema I presented in section 3 above. This omission leaves the central
novelty of his account mysterious. Second, as he admits, he is unable to
formulate precisely his equally central concept of the "analog relation"
between I B and reduced theory T R. This latter concept grounds the "amount of
correction" implied to T R via its reduction to reducing theory T B, and so in
turn Hooker's entire intertheoretic reduction spectrum between "smooth" and
"bumpy" endpoints. He laments : "Unhappily, I can think of no neat formal
conditions which would intuitively separate the two" (1981, p. 223). He hints
that "category-theoretic methods" might ultimately give some quantitative
account of "comparative preservation indices" for a theory 's "theoretically
relevant properties" and subsequently for its posited entities and events (1981 ,
p. 224). With such formal measures, one could judge quantitatively how
"corrective" a given reduction is. But he admits that "all of this is very programmatic and as yet lacking in deep yet simple insight" (1981, 224). To date,
Hooker has not addressed this lacuna in print.
These problems occupy crucial chapters in my first book (Bickle
1998, chapters 2 and 3). I address the first shortcoming by reformulating in
Hooker-reduction terms the mathematical derivation of the classical ideal gas
law (pV = nrT) from Avogadro's hypothesis, the universal gas constant,
kinetic-theoretic assumptions about the nature of microparticles, and
principles of Newtonian mechanics (Bickle 1998, chapter 2). One mathematical rearrangement to standard "textbook" treatments yields an equation
containing only quantitative expressions from statistical mechanics and
microphysics that mimics exactly the ideal gas law (Bickle 1998,34-39). This
result illustrates the structure of a Hooker image I B in an actual reduction
from the history of science.
To address the second shortcoming of Hooker's theory, I adopt a
fundamentally different account of the structure of scientific theories (Bickle
27
1998, chapter 3). Hooker adopts (implicitly) the familiar "syntactic" view
from logical empiricism. A theory is a set of propositions (generalizations)
characterized by thei r syntactic structures and relations. I replace this view
with a "semantic" account of theory structure, drawing on work from the
"structuralist" program in philosophy of science. This program was initiated
by Joseph Sneed (1971) and Wolfgang StegmUller (1976), and was presented
most completely and rigorously in Balzer et al. (1987). It built explicitly upon
earlier work by Patrick Suppes (1956, 1965, discussed briefly in section 3
above), who characterized a theory's basic structure not as a set of sentences
or propositions but rather as a set of models meeting specific set-theoretic
conditions. The structuralist literature contains two developed accounts of
intertheoretic reduction, one constructed self-consciously to capture Nagel's
intuitions (Balzer et al. 1987, chapter 5), the other constructed to capture
Thomas Kuhn's alternative (Mayr 1976). One explicit goal in my 1998 book
was to build Hooker's concept of an image In into a structuralist-inspired
account of theory structure and intertheoretic reduction. The technical (settheoretical) details of my account are complex and don't bear repeating here,
but the basic idea is straightforward. Instead of characterizing intertheoretic
reduction in terms of syntactic derivations, the "new wave" approach
construes the relation as the construction of an image (Hooker's In) of the settheoretic structure of models of the reduced theory T R within the set
comprising reducing theory T B.
What does this shift to semantic resources accomplish? My theory
provides precise, semi-formal accounts of "amount of correction" and
"location on the intertheoretic reduction spectrum." These were the very
shortcomings that Hooker himself lamented. The structuralist concept of a
"blur," which I extended to apply to intertheoretic relations (like reduction),
provides a rough cardinal estimate of the "amount of correction" implied to
the T R in specific cases. Once again , I used a detailed, semi-formal analysis of
the equilibrium thermodynamics-to-statistical mechanics and microphysics
case to illustrate this application (Bickle 1998, chapter 3, section 5). Application of another structuralist resource, "ontological reduction links" (Moulines
1984), provided my account with an answer to a famou s objection that
Schaffner (1967) leveled against set-theoretic approaches to reduction.
Focusing explicitly on Suppes (1956), Schaffner contended that such accounts
are too weak . They cannot rule out cases of nonreduction where the theories
accidentally stand in the set-theoretic relations said to comprise intertheoretic
reduction . But with the "new wave" intertheoretic reduction relation characterized in part by "links" between elements of the domains and relations
comprising the individual models of the reduced and reducing theories, this
"too weak to be adequate" challenge is met. The "global" reduction relation
across reduced and reducing theories' sets of (potential) models is itself
28
composed of relations between the empirical base sets and relations that
partially define each , relations that amount to either set identity or
replacement (Bickle 1998, chapter 3, section 3). In a later chapter I recon struct semi-formally parts of the reduction of a (schematic) propositional
attitude psychological theory of cognitive representation to a (similarly schematic) connectionist-inspired counterpart (Bickl e 1998, chapter 5, section 3,
following Bickle 1993).
New wave psychoneural reductionism is thus the prediction that as
mature theories develop in psychology (TRs) and neuroscience (TBs), images
(IBs) of the former will be constructible within the models of latter. The
challenges that Hooker (1981) was unable to meet can be addressed when his
insights about scientific reduction are reconstructed and extended within an
alternative account of theory structure and intertheoretic relations. This leave s
only the "put up or shut up" challenge of demonstrating existing or potential
reductions of actual, genuinely cogn itive psychological theories to neurosci entific counterparts. I argued that the new wave reduction relation is already
obtaining, albeit sketchily, between cognitivist associative learning theories
and neurobiological account s of experience-driv en synaptic plasticity (Bickle
1998, chapter 5, following Bickle 1995b). This case appears to be headed for
a location on the intertheoretic reduction spectrum midway between the
"smooth" and "bumpy" endpoints, akin to the location of the equilibrium
thermodynamics-to-statistical mechanics and microphysics case . This predicted location warrants a "revisionary physicalism" about the reduced
psychological posits on the isomorphic ontological consequences spectrum.
The upshot of Bickle (1998) was that an IR reformulation of the
tradit ional philosophical mind-body problem built upon the "new wave"
account of intertheoretic reduction provides viable-and at present, the most
fully developed-resources for articulating and defending psychoneural
reductionism. The account is heir to a tradition running from U.T. Place and
IJ.c. Smart through Ernest Nagel and Paul Feye rabend to Kenneth Schaffner,
Clifford Hooker, and Paul and Patricia Churchland. In the best tradition of
scientifically informed philosophy, it weaves together both philosophy of
science and current empirical science, and applies the resulting mosaic to
tradit ional dilemmas of perennial philosophy (like the mind-body problem).
29
5 WWSD (WHAT WOULD SOCRATES DO?)

5.1 Problems for New Wave Reductionism
Of course, not all philosophers agreed with my assessment. Ronald
Endicott (2001) succinctly expresses a common criticism, writing:
[O]ne should distinguish between an account of scientific
reduction versus a defense of reductive physicalism, which,
more briefly, is the difference between "reduction" and
"reductionism." The latter may entail the former .. . But the
former does not entail the latter, seeing that an account of
scientific reduction might shed light on those cases where the
world shows itself in a simple and uniform way even if,
contrary to reductive physicalism, the world does not always
show itself in a simple and uniform physical way. Thus I
applaud Bickle's work on reduction. But I take issue with his
reductionism. (2001, 378; author's emphases)
One of Endicott's arguments is especially pertinent. He argues that my
adaptation of Hooker's "token reduction" relation into a structuralist
framework depends entirely on an implausible eliminativism about functional
kinds. Maurice Schouten and H. Looren de Jong (1999) also chide me for
ignoring functionally characterized psychological concepts, especially ones
employed in my favorite case from actual psychoneural science, the
"memory-long term potentiation (LTP) link." They claim that this case is not
an "accomplished reduction," as I claim that it is in my answer to the "put up
or shut up" challenge. Instead, they insist, there remain some legitimate, not
purely heuristic questions that can be answered only with genuinely functional
concepts. Lower-level "combinatorial" explanations cannot address these
(1999, 255-256). They write:
The completed account of the molecular alphabet will only
specify causal roles or dispositions, not real functions ..
.When we adopt an etiological view of functions, as we have
claimed one must, we cannot but conclude that a finalized
specification of the dispositions of LTP is not enough to build
a bridge to the functionally identified property it is supposed
to explain, say, spatial learn ing. ... Invoking functions , to find
out why a structure or mechanism exists, intrinsically refers
30

to the evolutionary reason s of higher-l evel organization.
(1999 ,256)
The beat goes on. Robert Richardson (1999), drawing on themes

developed by William Wimsatt (1974), insists that the rationale for pursuing
research into lower level mechanisms is explanatory adequacy, not ontological parsimony. Scientists descend to lower levels only when a higher-level
account fails to explain phenomena within its domain . This purported
historical fact about scientific practice raises a problem for seeking
ontological consequences directly from an intertheoretic reduction relation.
Richardson points out further that my detailed example of a psychoneural
reduction from actual current science nowhere addresses the claimed
dependency or sensitivity of human-level cognition to semantic content, and
so he worries that the cellular models I focus on might thereby lack sufficient
resources to handle human cognition. He writes: "W ithout structured
representations and without an understanding of the plasticity in human
learning, such a reductionism will not go very far in converting the convicted
cognitivist" (1999 , p. 306). Barbara Hannan (2000) is even harsher on this
point:
How is a scientist supposed to capture the fact that an internal
state has a representational content without using language,
including sentences containing terms that refer to objects,
states, and events outside the brain ? Bickle's offhand
reference to a neurophysiological account of content is
disingenuous; we possess no such account, and he ought to
know we don 't (2000,54)1 7
Finally, Achim Stephan wonders what justification I can give for why
we "s hould look through the glass of theory reduction to come to grips with
the old ontological mind -body problem" (2001, 281). Following Joseph
Levine (1983) and David Chalmers (1996), both Stephan (2001) and Ansgar
Beckerman (2001) see only one way to forge this connection: through an
account of reductive explanation that first defines mental concepts in
functional terms and then looks to intertheoretic reduction to provide the
physical mechanisms that realize those funct ional roles . The problem with this
justification is the seeming impossibility of defining some mentali stic
concepts functionally (even cognitive concepts, according to Stephan). So my
IR reformulation lacks appropriate justification, and according to Beckerman
(2001) involves a mistaken notion of what physicalism amounts to. "
Notice that each of these criticisms questions, in one way or another,
my application of new wave intertheoretic reduction to psychoneural
31
examples from science. One reply we might expect from a philosopher (of
science) would be to tinker with the general theory of reduction so that its
application would avoid these problems. Another would be to seek out other
examples from the psychological and brain sciences that better fit the general
theory. (These typical responses explain the joke that serves as the title to this
section.) But I'm going to take a different tack. As I've already noted, a
ruthless reductionism underlies both theory and experimental practice in the
cellular and molecular core of current mainstream neuroscience. So instead of
massaging superficial descriptions of results from these disciplines so as to
make them fit-to "apply"-a general concept of scientific reduction, I am
going to delve into the details of current research and theory. Once I show
how psychological theories are actually being explained in current cellular
and molecular neuroscience, independent of any pre-theoretic or "philosophical" account of explanation, we can investigate the potential of this
reductionism-in-practice for behavior and cognition generally. We can even
examine its consequences for "hot" current topics in the philosophy of mind.
As we are about to see, this move both implies and is based upon a drastic
alternative to currently popular views about the nature and role of philosophy,
including the view that motivated new wave reductionism.
5.2 New Wave Metascience

This book is thus an exercise in "bottom-up" philosophy of science, a
project in what I will call new wave metascience. "Bottom up" refers to my
letting a sense of reduction emerge from the detailed investigations drawn
from recent scientific practice, instead of "imposing" a general account of
scientific reduction onto them from science in general, or the "top down."
Since in current cellular and molecular neuroscience, the devil is in the
experimental and methodological details, and since these details are virtually
unknown among philosophers and cognitive scientists, I will emphasize them
in this book. Only then will we see how current neuroscience is "linking mind
to molecules," and so only then can we speculate responsibly on the
explanatory potential and scope of these resources for behavior and cognition
generally.
One feature of my new wave metascience deserves explicit
discussion. It eschews all traditional concern with ontology and "metaphysics," even with the secondary, derivative sense I advocated in my (1998)
book . There my guiding intuition was that science is the best method we have
for "doing ontology," for constructing informationally rich representations of
a "real world" assumed to exist independently of our representing it. I
assumed that neuroscience is doing just fine by ignoring "philosophical"
32
objections to psychoneural reduction and proceeding instead in the fashion I

tried to capture within new wave reductionism. I was urging that we accept
science's strategy tout court. Trust scientific practice and let ontological chips
fall where they may! However, I've since come to see even that project as
"too metaphysical." Even though scientists talk a "realistic"-sounding
language, we should not interpret this talk as addressing questions "external
to" the practices and concerns of a given scientific endeavor. The job of new
wave metascience is simply to illuminate concepts like reduction as these
imbue actual scientific practice. To what end? Not to achieve some better way
of addressing reformulated "external" questions about the existence and
nature of "theory-independent ontology." Rather, because a reasonable
explanatory goal is to understand practices "internal" to important current
scientific endeavors and the scope of their potential application and
development. The tasks of this book are part of a meta science of
contemporary psychology and neurobiology, not a part of some "ontology of
mind."
I borrow the terms "internal" and "external" deliberately from
Rudolph Carnap ([1950], 1956). Philosophers might remember that Carnap
there distinguishes two types of existence questions: "first, questions of the
existence of certain entities . . . within the [linguistic] framework; we call
them internal questions; and second, questions concerning the existence or
reality of the system of entities as a whole, called external questions" ([1950]
1956, p. 206 ; author's emphases)." Using his favorite example of the "thing
language," Carnap continues: "Once we have accepted the thing language
with its framework for things, we can raise and answer internal questions,
e.g ., 'Is there a white piece of paper on my desk?,' 'Did King Arthur actually
live?,' 'Are unicorns and centaurs real or merely imaginary?' and the like.
These questions are to be answered by empirical investigations" (207).
External questions, on the other hand, have a more "problematic character":
"From these questions we must distinguish the external questions of the
reality of the thing world itself, e.g., 'Does this table really exist?" (ibid.).
According to Carnap, only philosophers ask external questions, as contrasted
with "the man in the street" and scientists. The usual methods of investigation, logical analysis and empirical inquiry, do not provide complete
answers to them. For these questions possess a different character than
existence questions asked within a linguistic framework. He continues: "If
someone decides to accept the thing language... this must not be interpreted as
if it meant his acceptance of a belief in the reality of the thing world; there is
no such belief or assertion or assumption, because it is not a theoretical
question. To accept the thing world means nothing more than to accept a
certain form of language" (207-208). Such a decision typically is influenced
by genuinely "theoretical knowledge." What are the purposes for which the
33
framework is to be used? How efficient is the framework for achieving these

purposes? How fruitful? These questions are addressable by the usual
methods of empirical inquiry . But it would be incorrect to infer that positive
answers to them "support the reality of things." Instead, we should say that
these features of the framework's use "made it advisable to use the physical
thing language" (208)
These remarks lead Carnap to a general account of external existence
questions. He writes: "Those who raise the question of the reality of the thing
world itself have perhaps in mind not a theoretical question as their
formulation seems to suggest, but rather a practical question, a matter of
practical decision concerning the structure of our language" (207). This is not
a special feature of external questions about physical things. He draws the
same conclusion about external questions pertaining to numbers, propositions,
thing-properties, spatio-temporal coordinate systems for physics, and the
abstract objects used by semanticists in the formal analysis of linguistic
meaning.
Carnap notes some profound methodological consequences of this
account, especially for those seeking to undermine or "reduce" some
framework to another. Consider his remarks to "nominalists" who seek to
eliminate the "abstract entities" (properties, relations, propositions) commonly
employed in formal semantics. Such critics must do more than simply raise
problems for the "ontology" of such entities. That move only conflates
internal and external questions . Instead, nominalists must "show that it is
possible to construct a semantical method which avoids all references to
abstract entities and achieves by simpler means essentially the same results as
other methods" (221). Or consider the options Carnap makes available to
skeptics of the thing language. Casting aspersions upon "the ontology of
things" likewise conflates internal and external questions. Instead, skeptics
must either "restrict [them]selves to a language of sense-data and other
"phenomenal" entities, or construct an alternative to the customary thing
language with another structure, or, finally , ... refrain from speaking" (207).
There is no short cut around actually constructing the alternative (linguistic)
framework
1 advocate a similar lesson for psychoneural reductionists. Physicalist
reductionists need actually to provide psychoneural reductions . Paraphrasing
Carnap, they must show that it is possible to construct a neuroscientific theory
that avoids all references to psychological entities and achieves by simpler
means the same explanatory scope as psychology. Luckily, contemporary
cellular and molecular neuroscientists are doing the constructing. But these
scientific developments must be set within the appropriate context. There is
no shortcut for the psychoneural reductionist around this wholly empiricalcum-metascientific task. The only factual questions are "internal" ones, about
34
the way actual neuroscientific practices are proceeding and about these practices ' foreseeable explanatory potential.
Philosophers will wonder whether any kind of "internal/external"
distinction can be drawn. Didn't Willard Quine trash Carnap's distinction
soon after the ([1950] 1956) paper was published, by undermining any
absolute distinction between the analytic and synthetic (roughly, between
sentences "true by virtue of the meanings of their terms" and others "true by
virtue of the way the world is")? Quine thought that he undermined Carnap's
distinction. In the paper where he famously dismantled an absolute
analytic/synthetic distinction, Quine writes: "Carnap has recognized that he is
able to preserve a double standard for ontological questions and scientific
hypotheses only by assuming an absolute distinction between the analytic and
the synthetic; and I need not say again that this is a distinction which I reject"
([1949] 1953,45-46). Two years later, Quine reiterated this point: "If there is
no proper distinction between analytic and synthetic, then no basis at all
remains for the contrast which Carnap urges between ontological statements
and empirical statements of existence. Ontological questions then end up on a
par with questions of natural science" ([1951] 1966, 134). Is any approach
motivated by Carnap's distinction thereby dead in the water, in philosophy's
post-Quinean times?
No. For there is at least one argument that Carnap offered in his
([1950] 1956) essay that nowhere depends upon some absolute
analytic/synthetic distinction. Commenting on the classic philosophical debate
between "Platonists" and "nominalists" about the existence of numbers, he
writes : "I cannot think of any possible evidence that would be regarded as
relevant by both philosophers, and therefore, if actually found, would decide
the controversy or at least make one of the opposite theses more probable than
the other" (219) . From this he concludes: 'Therefore I feel compelled to
regard the external question as a pseudo-question, until both parties to the
controversy offer a common interpretation of the question as a cognitive
question; this would involve an indication of possible evidence regarded as
relevant by both sides" (ibid.). Call this the "fruitless question" argument.
Without any agreement about relevant evidence by disputants, a question isn't
worth serious pursuit. I contend that the class of "fruitless questions" matches
up closely with Carnap's original class of "external questions"-including
philosophy's traditional mind-body question and even "phy sicalist" answers
to it. Nothing in Quine's famous arguments against an absolute analytic/
synthetic distinction counts against this Carnap-inspired distinction between
(pragmatically) fruitful and (pragmatically) fruitless questions.
There is even some textual evidence that Carnap did not acknowledge
the "recognition" Quine attributed to him (see again the first quote from
Quine above). Throughout the entire ([1950] 1956) essay , Carnap only
3S
mentions the "analytic/synthetic" distinction once, in a discussion of distinct

types of internal questions: those in frameworks for logic and mathematics as
compared to those in frameworks for empirical science. Nowhere in his extensive discussions of external questions does he ever suggest that the analytic/
synthetic distinction is involved. Furthermore, in a footnote he adds: "With
respect to the basic attitude to take in choosing a language form (an
"ontology," in Quine's terminology, which seems to me misleading), there
appears now to be agreement between us: "the obvious council is tolerance
and an experimental spirit?" (footnote 5). This hardly sounds like the
"recognition" Quine claimed! Of course, this does not show that an absolute
analytic/synthetic distinction isn't lurking in the background of some of
Carnap's arguments. But it is very difficult to find it lurking anywhere in or
around the "fruitless questions" argument I drew out of that text and
expounded one paragraph above .
So I contend that we can maintain a Carnap-inspired distinction
between pragmatically fruitful and fruitless questions even in these "Quinean"
days, and that the class of pragmatically fruitless questions by and large is
coextensive with Carnap's original class of "external questions." Roughly, a
question is fruitful only if those advocating different answers can come to at
least minimal agreement about relevant evidence toward deciding it. This
distinction was not the centerpiece of Carnap's ([1950] 1956) essay, and so
new wave metascience isn't merely old-fashioned logical positivism rediscovered. New wave metascience doesn't treat traditional ontological questions
as "cognitively meaningless," just pointless to pursue as serious intellectual
work because they amount to nothing more than disagreements over bare
intuitions, with no agreed-upon evidence for resolving them. Clearly this
distinction justifies my exclusive focus on practices and developments within
interdisciplinary cognitive and brain sciences and my eschewing "ontological" considerations, even ones "secondary to and dependent upon"
questions about intertheoretic relations, practices and attitudes within the
relevant sciences. Relevant evidence is straightforward for questions "internal
to" scientific practices: what experiments are being pursued, which results are
being published in the best journals, which researchers (asking which
questions) are getting funded , who is winning the most prestigious awards'r"
The ontological questions, however, even when reformulated using resources
from philosophy of science and the relevant sciences, remain as "fruitless" as
ever.
Am I being realistic? Can we really ignore the traditional mind-body
problem? The honest must admit that the problem still exerts a pull. Can we
assuage this? Again, we can learn from Carnap. Early in his corpus, he
divided his "philosophical" project into "pos itive" and "negative" parts,
writing:
36
The positive result is worked out in the domain of empirical

science; the various concepts of the various branches of
science are clarified; their formal-logical and epistemological
connections are made explicit. In the domain of metaphysics,
including all philosophy of value and normative theory,
logical analysis yields the negative result that the alleged
statements in this domain are entirely meaningless. ([1932]
1958, 60-61)
Being jealous guardians of their historical turf, philosophers focused their
critical attention almost exclusively on Carnap's negative project. Thereby
they missed an important lesson buried in his positive project, and continue to
do so up to the present day.
In the Introduction to his classic essay/monograph, The Unity of
Science (1934), Carnap remarks that the "Physicalism" developed thereinthat "all statements in science can be translated into physical language"-is
"allied to that of Materialism." But "the agreement extends only as far as the
logical components of Materialism: the metaphysical components, concerned
with the question of whether the essence of the world is material or spiritual,
are completely excluded from our consideration" (1934, 28-29). We must
reject Carnap's tools for carrying out his positive project: his quaint accounts
of "translation" and "logical analysis" and his willingness to massage actual
scientific practice to generate "rational reconstructions." The replacements I
am urging indicate another way that new wave metascience is more than just
warmed-over Carnap, despite shared anti-metaphysical motivations." But I
now agree wholeheartedly with the intuition guiding this quote. The "physicalism" of new wave metascience is not an "ontological hypothesis." It is
instead a description of reductive practices and accomplished results "internal
to" contemporary cellular and molecular neuroscience. It is an argument for
the explanatory potential of these practices. It is an application, mostly
debunking, of this "reductionism-in -practice" to some trendy assertions in
current philosophy of mind. And it is an account of the entire multi-leveled
discipline of neuroscience from the perspective of its current cellular and
molecular core.
There is another sense in which new wave metascience implies
changes to methodology in philosophy. Even philosophers sympathetic to
science and its potential impact on philosophical issues tend to reserve a task
for philosophy that is independent in at least some sense from scientific
details. The status of "reduction" seems clearly to have something to do with
"explanation." The reducing theory must at least explain all the data that the
reduced theory explains. So doesn't any defense of reductionism require an
37
account of explanation? And isn't that a task for philosophy-sensitive to

scientific practices and results , but "over and above" them ? If one isn't armed
with such an account, isn't the choice of investigating cellular and molecular
neuroscience, as opposed to cognitive neuroscience or even cognitive science
in general, arbitrary?
New wave metascience says no. Scientists tend to do just fine with a
rough -and-ready understanding of what counts as an "explanation" and what
distinguishes a "good" one from a "poor" one. At the very least, the practices
accepted by practitioners of succes sful scientific disciplines are enough to
separate successful from unsuccessful programs and to self-correct errors and
mistake s. (It is commonplace to hear philosophers of science remark on
science's "self-corrective" nature . One gues ses that scientists didn't acquire
this capacity by reading philosophers of science.) Starting a philosophical
analysis with a detailed example of a scientific explanation is part of "taking
science at face value." There is nothing wrong with trying to make these
practices more explicit, to abstract general principles out of specific instances.
That is a philosophical project. But the guiding intuition of new wave
metascience is that this project should be pursued from the bottom up, that is,
from a detailed investigation of real examples of current research and an
attempt to extend these results and practices to issues traditionally reserved
for (and by) philosophers. However, there is no "magic argument" for this
feature of new wave metascience. The only procedure is to start constructing
from current scientific examples and then to challenge philosophers to
articulate what seems to be missing; and, if an answer to the first question is
offered, to ask why those missing features are important. The rest of this book
is a first step toward carrying out this procedure with regard to the "ruthless
reductionism" endemic in current mainstream (cellular and molecular)
neuroscience. Judge this methodological feature of new wave metascience,
that of trusting what scientists consider successful explanations, only after the
construction has been presented ."
Some philosophers might worry that new wave metascience,
especially its Camap-inspired "pragmatism," concedes too much to "antirealists" and "mysterians." Can 't one mobilize pragmatic considerations
directly to defend a Reali stic interpretation about the results of our succes sful
sciences? And what about philosophy itself? What is the content of my
purported "positive" project ? Another important philosopher from the midzo" century, Hans Reichenbach-whose writings are now as equally
neglected as Carnap's-had some pithy remarks that address both of these
que stions. In his more popular book, The Rise of Scientific Philosophy (1957),
Reichenbach notes the misplaced confidence that some philosophers place on
science, especially philo sophers of science, as contrasted with the scientist
who from his own experience knows better:
38
It is a strange matter of fact that those who watch and admire

scientific research from the outside frequently have more
confidence in its results than the men who cooperate in its
progress. The scientist knows about the difficulties which he
had to eliminate before he could establish his theories. He is
aware of the good luck which helped him discover theories
that fit the given observations and which make later
observations fit his theories. He realizes that discrepancies
and new difficulties may arise at any moment, and he will
never claim to have found the ultimate truth . Like the disciple
who is more fanatical than the prophet, the philosopher of
science is in danger of investing more confidence in scientific
results than is warranted by their origin in observation and
generalization. (1957, 43).
The "scientific realists" that came to dominate late-20 th century philosophy of
science stand guilty of Reichenbach's charge. I suppose that if one is going to
do metaphysics, then scientific realism beats the alternatives. But there is the
option of eschewing metaphysics for investigations internal to particular
practices, where the latter are chosen to be worthy of investigation on
pragmatic grounds.
This still leaves the nagging question about the content of a "positive"
philosophical project aimed at mainstream (cellular and molecular)
neuroscience. I've already said that providing the first step toward this is the
project of this entire book, so small wonder that I can't articulate a convincing
short answer to it in a few sentences at the outset. But a remark from Reichenbach is instructive about where the materials for this positive project are to be
found. Commenting on the origins of the "new scientific philosophy" in 19th
century science, Reichenbach writes :
Just as the new philosophy originated as a by-product of
scientific research, the men who made it were hardly
philosophers in the professional sense . They were mathematicians, physicists, biologists, or psychologists. Their
philosophy resulted from the attempt to find solutions to
problems encountered in scientific research, problems which
defied the technical means thus far employed and called for a
re-examination of the foundations and the goals of
knowledge. Rarely was such philosophy detailed or explicit,
nor was it extended beyond the confines of the fields of their
maker's particular interests. Instead, the philosophy of these
39
men lurks in the prefaces and introductions of their books and

in occa sional remarks inserted in otherwise purely technical
expositions. (1957 , 123)
Regarding Reichenbach's last point , recall the quotes from the introductory
chapters of the last two editions of Kandel, Jessell, and Schwartz's text that I
cited early in this chapter to illustrate the "ruthless reductionism" endemic in
current mainstream neuroscience. The job of Reichenbach's "scientific
philosopher" is to "collect these results" of scientific research, add to them
with scientific investigations (if motivated to do so), and "present them with
all their interconnections." The result is a "synopsis of scientific answers to
philosophical questions" (1957,119), or at least to the residue of the questions
that are fruitful to address. Scientific philosophy as it will be practiced in this
book-new wave metascience-is heir to a brilliant tradition spanning most
of the 19th and zo" centuries, until the end of the latter, when "speculative
metaphysi cs" infected even philosophy of science and philosophy of particular sciences with "external," "pragmatically fruitless " debates that tum on
nothing more than clashing intuitions.
My strong statements about an alternative to philosophy of
neuroscience as currently practiced, and more generally to the broader current
philosophical project of "naturalizing the mind," coupled with references to
half-decade-old (and more) writings by out-of-vogue positivists, won't
convince philosophy's orthodoxy. That isn't my purpose in this chapter. I'm
here only staking out a position, making explicit the bigger picture that
motivates my interest in state-of-the-art cellular and molecular neuroscience,
the mainstream of the current discipline. The only arguments that should
convince anyone are the ones mobilizing resources from these sciences and
showing how much they can accomplish. Even if you don 't share my
philosophical motivations or program, or if you are neutral on these issues
about philosophical method (as I suspect most neuroscientist readers are), you
might still find it interesting to see how much can be said in response to some
influential philosophical issues by building up from the "ruthlessly reductive"
basis provided by current mainstream neuroscience. If you count yourself as a
philosopher of neuroscience, you at least owe a look at the work that
constitutes the bulk of your discipline's current research. And if you are an
anti-reductionist about mind for any reason, how can you know the limits of
"ruthless reductionism" if you don't know what that approach is actually
doing and achieving in current science?
I close this introductory chapter with one more nod to a forerunner.
Obviously, a work like this one owes much to Patricia Churchland's
groundbreaking Neurophilosophy (1986). Before that book carved out a niche,
no one would have thought to combine philosophy and neuroscience in the
40
way I will here. But I think I can do Churchland one better, and not just by
updating her neuroscience by a decade-and-one-half. In the Gene ral
Introduction to her book, she writes:
Philosophers who are expecting to find in the introduction to
neuroscience a point-by-point guide of just what facts in
neuroscience are relevant to just what traditional philosophical problems will be disappointed. I have made some
occasional efforts in that direction, but in the main my eye is
on the overarching question of the nature of a unified,
integrating theory of how, at all its levels of description, the
brain works . (1986 , 7)
I too am interested in that overarching question (as a piece of new wave
metascience), but I also think I can give a "point-by-point guide " of which
results from recent cellular and molecular neuroscience are relevant to which
current philosophical problems. Details and applications over the next three
chapters will confirm or disconfirm this.
NOTES
1
My discus sion in the next two sections draws on Bickle (2000) .
2 Carl Craver has pointed out to me that stressing this observation might be useful in breaking
down some of the communication barriers that still exist between neuroscientists and
philosophers. Current philo sophers as a group are not a horde of substance dualists-as the
introductory chapters in many neuroscience textbooks suggest or imply!
See Churchland, P.M. (1987, 58-59) for some simple examples.
4 In my (1998) I took folk psychology to be the theory with which our common sense ontology
of mind is affiliated. A number of commentators took me to task for this, including Bontley
(2000), Hannan (2000) , Richardson (1999), and Stephan (2001) . 1 accept their criticisms
(although in Bickle 2001 1 try to explain why I took this view). Nothing hinges on this change
for the advantages of the " Intertheoretic Reduction reformulation" of the mind-body problem
about to be stressed.
1 treat this point in detail in my first book, including providing exampl es of "lin guistic "
arguments still in vogue in some philosophical circles today that are irrelevant to theory
reduction issues in science . See Bickle (1998 , Chapter 2, section 3, especially pages 44-45) .
Hooker (1981) provides a nice introduction to these details. I capture some of these detail s
within a quasi-formal account of the intertheoretic reduction relation (Bickle 1998, chapters 2
and 3).
This is the same example 1 employed, independently, in Bickle (1995b) and (1998 , chapter 5).
My usc of this case to illustrate an actual psyehoneural reduction goes back to my doctoral
dissertation (Bickle 1989).
7
41
With regard to both of these first two features, Schaffner is implying that not only is Nagel 's
account of intertheoretic reduction incapable of handling psychology-to-neuroscience theory
relations, but that the very account of "theory" he presupposed is wrong for these cases, also.
Thanks to John Symons for calling this point to my attention .
8
For this reason , I've changed the symbol Hooker (1981) uses to denote the "analog structure."
Unfortunately, he chooses the symbol T R*.
10 In Bickle (1998, 30) , 1 provide a diagram of these isomorphic spectra with these historical
reductions located on both.
11 In Bickle (l992b) and (1998, chapters 5 and 6) 1 offer empirical evidence for a future
revisionary reduction of psychological to neuroscientific theories . There 1 argue that while a
synaptic weight-vector based account of cognitive content drawn from cognitive and
computational neuroscience eschews the propositional contents of belief-desire psychology, it
nevertheless preserves the coarse-grained functional (cause-and-effect) profile and the
intentionality that the latter ascribes to cognitive states, especially to states near the sensory and
behavioral (motor) peripheries.
12 Key papers are reprinted in Putnam (l975a). In Bickle (1999) I review key themes in the
literature that Putnam initiated, but see my (1998 , chapter 4) for a more detailed, technical
discussion.
13
I'll describe this argument more completely in Chapter Three, section 4.
14 1 introduced these terms in Bickle (l992a) and expounded them further in Bickle (1998,
chapter 4).
I explicate this account further in Bickle (l992a) and apply it to a psychoneural example in
Bickle (1998 , chapter 4).
15
16 Trent Jerde points out to me that a great majority of PET and tMRI studies use voxel sizes in
the 3 X 3 X5 mm range ; voxel sizes of 1 mm or less almost always refer to anatomical images .
And even this much grosser "standard working resolution" for functional images requires
statistical techniques that "smooth the data" or require the presence of "contiguous active
voxels" to assert a claim about significant activation. 1don't disagree . There is much guff about
functional neuroimaging in both the popular press and philosophical discussions. The study I'll
report in detail in Chapter Three below uses both the much less impressive voxel sizes he cites
and a variety of statistical "smoothing" techniques. But Scott Holland also informs me that
recent high-field magnets (e.g., 7 Tesla) , now approved only for research on laboratory
animals, can get far smaller junctional resolution than even 1 mm.
17 "Offhand reference"? 1 spent twenty pages (in my 1998, chapter 5) explaining the
psychological and neuroscientific details of associative learning, hierarchically structured
memory , and experience-driven synaptic plasticity, as these theories stood in the early 1990s. I
then spent most all of chapter 6 drawing consequences from these details for the realismeliminativism debate about propositionally structured content.
18 These are not the only challenges to new wave reductionism in print. Others will arise later in
this book.
19 Unless otherwise noted , quotes from Carnap in the remainder of this chapter are from his
([ 1950] 1956). 1 indicate them only by page number.
42
Notice that many of my questions go beyond what Carnap considered "internal questions."
Many will notice Kuhnian themes in my list. This is another sense in which new wave
metascience isn't merely logical positivism revived. The distinction I am stressing is Carnapinspired, not Carnap 's .
20
21 The philosophy of science (minus the ontolo gy) in Bickle (1998 , chapters 2 and 3) can be reinterpreted in light of new wave metascience as providing alternatives to Carnap 's quaint
notions and translational strategy .
Thanks to Carl Craver and Dingmar van Eck for insisting that I elaborate on this important
methodological difference between new wave metascience and more mainstream philosophy o f
science.
22
CHAPTER TWO
REDUCTION-IN-PRACTICE IN CURRENT
MAINSTREAM NEUROSCIENCE
Most philosophers of mind and many cogrutive psychologists still

doubt that "genuinely cognitive" psychological theories will reduce to
neurobiological count erparts. As I emphasized in Chapter One, this attitude
contrasts starkly with the reductive aspiration s of "mainstream" cellular and
molecular neuro scientists. My first substantive task in this book is to
characterize "reduction-in-practice," as it is generating experiments, results,
and explan ations in current mainstream neuroscience. For reasons sketched in
the previous chapter, I'll ignore at first the concerns typically assumed by
philosophers to occupy center stage in discussions of psychoneural reduction,
namely philosophical anti-reductionist arguments and the probl ems of
cha racterizing a concept of intertheoretic reduction for science generally .
Instead, I'll begin by presenting a detailed example , recent discoveri es about
the molecular mechani sms of long-term potentiation (LTP), an important type
of experience-driven synaptic plasticity, and the behavioral data these
mechanisms explain. This story is an accomplished neurobiological reduction
of psychology's "memory consolidation switch" that mediates the conversion
of short-term to long-term memories. These recent scientific details show the
nature of redu ction at work and succeeding in current cellular and molecular
neuroscience. Later in this chapter, as a piece of new wave metascience, I'll
generalize from this example, hopefully to provide a template for additional
psychoneural reductions.
There is a second reason why the detail s of this example are
important for philosophers and cognitive scientists. There is a widespread
misconception about the content of "neuroscientific explanations" of behavior. Many still think of these explanations on the model of what neuroscience
provided three decades ago, namely, circuit diagrams of anatomical
connections between neural region s known to subserve specifi c functions,
with some molecules (e.g., neurotransmitters) indicated in a few spots .
Indeed , there is a widespread misconception that "we don 't know much about
how the brain works." This misconception ignore s completely the "molecularbiological wave" that began sweeping through neuroscience more than two
decades ago. I gave sociological data in support of this trend in Chapter One
and there suggested that more than sociological justifications for it could be
44
offered. Now I want to make good on the latter assertion by presenting an

example of a detailed neuroscientific investigation, showing how this
approach has begun generating novel behavioral experiments, results, and
explanations. This example is intended to demonstrate clearly the character of
mainstream neuroscientific investigations in the present and the foreseeable
future. Finally, it will show that we do know a lot about "how the brain
works ," at least its fundamental components, and how we can manipulate
these components to produce specific, measurable behavioral effects. Even if
philosophers, cognitive scientists, and cognitive neuroscientists find reasons
to reject the "ruthless reductionism" I urge from this current example, they
will at least see some details about actual practices in mainstream
neuroscience circa the beginning of the 21st century.
1 A PROPOSED "PSYCHONEURAL LINK"

Much has been written over the past twenty years about a "learning
and memory-long term potentiation (LTP) link" And much of this scientific
and theoretical literature has been misunderstood. Not even its most strident
proponents think that LTP is the cellular/molecular mechanism for all "forms"
or "types" of memory (although their writings sometimes give this mistaken
impression .) About the "simple proposition" that "LTP is a substrate of
memory," neurobiologist Gary Lynch has written recently:
Substituting the definite article in this proposition converts it
to a form that is manifestly not true (it is easily shown that
forms of memory exist that do not involve LTP), and yet it is
widely employed, perhaps unwittingly, in LTP/behavior
arguments. These difficulties provide ample reason to be
pessimistic about the still-popular search for an experimentum
cruces that "proves" or "disproves" the role of LTP in
memory. (2000, 139).
Lynch has been among the most forceful and influential proponents of a
"learning and memory-LTP link" for more than two decades.
Memory-like features of LTP have been explored vigorously since its
discovery in the early 1970s. These include its
selectivity: It only affects transmission efficacy at specific synapses

on a given neuron.
cooperativity: Larger effects and more stable induction occur when
more afferent fibers (input lines to the neuron) are stimulated.
REDUCTION-IN-PRACTICE
45
multiple forms: Different molecular mechanisms underlie its variable

stability over time, remini scent of the short -term/long-term memory
distinction .
cumulative nature: Successive episodes of high-frequency stimulation
in the same afferent fibers produce increa sing amounts of synaptic
potentiation.
regional distribution: LTP has been documented in synapses in
mammalian cortex, hippocampus, other "limbic" structures, and
spinal cord (there called "wind up") . All of these areas have been
associ ated behaviorally with important types of long-term memory .'
Attempts to correlate the temporal dimensions of these synaptic

features with those from behavioral studies of memory have been prominent
since LTP's discovery. The advent and dominance of molecular techniques
and investigations in mainstream neuroscience over the past two decades have
influenced LTP research strongly. Work on the molecular mechanisms ofLTP
and attempts to relate these to measurable memory-guided behavior are at
present among the hottest and best-funded research areas in all of science.
Some decry this fact, but no one denies it. Since I want a detailed example
that reflects accurately the aims, goals, and aspirations of current cellular and
molecular neuroscience, this research program is a clear choice.
I should confess at the outset to finding this program and its
mechanisms to be among the most beautiful research in contemporary
science. I am inspired by it and what I see as its explanatory veracity and
potential. Accuracy and aesthetics aside, however, my purpose in describing it
is to present a paradigmatic example of an accomplished psychoneural
reduction, and hence a template for additional reductions. Like it or not, this
example illustrates the force of reductionism-including its ruthlessness-at
work in current cellular/molecular neuroscience.
However, I am discussing a limited proposal : that memory consolidation has been reduced ("linked") to the molecular mechanisms of LTP. This
is the extent to which a reduction is on offer , at least in the serious scientific
literature, and we'll even see in this chapter and the next a number of
distinctions that have been drawn between types of memory systems .
Scientists virtually never assert a global "memory is LTP" hypothesis in
primary research publications. Incidentally, the limited scope of this proposal
reveals errors in one of Schouten and de Jong's (1999) criticisms of my
previous appeal to this example as an accomplished new wave psychon eural
reduction. They write : "The functional characterization of associative,
Hebbian learning in terms of weight changes of synaptic connections has
driven LTP research by staging certain feature s of learning and memory. The
molecular account would have to provide the neces sary and sufficient
46
conditions for these feature s if a reductive explanation was to be achieved"

(1999', 249). They next point out that LTP is found in a variety of brain
systems, including some that apparently have nothing to do with learning and
memory . They cite approvingly the list given in McEachern and Shaw's
review (1996, section 9.1) and the lesson urged there: "LTP ... serves
functions other than, or in addition to, memory" (1996 , p. 80, quoted in
Schouten and de long 1999, p. 250) . So the "sufficiency" condition fails .
They also cite Saucier and Cain 's (1995) study , where a potent and specific
antagonist (blocker) of NMDA (N-methyl-D-aspartate) receptors, NPCI7742
(2RAR,5S-2-amino-4,5-( 1,2-cyclo hexyl)-7 -phosphonoheptano acid) completely blocked LTP induction in hippocampal dentate gyrus neurons.
Nevertheless, rats treated with this receptor antagonist that had been made
familiar with water maze task demands by non-spatial pre-training displayed
normal spatial learning in the water maze as compared to pharmacologically
untreated controls (Saucier and Cain 1995, Figures 2 and 3) .2 So LTP isn't
necessary even for spatial learning and memory. Schouten and de long
conclude, contra to my answer to the "put up or shut up challenge," that a
memory-to-LTP reduction isn't yet accomplished, as these concepts now
stand in psychology and neuroscience.
Two mistakes infect their criticism. First, the serious scientific
hypothesis links only one process of memory-consolidation-to LTP.
Second, no "necessity and sufficiency" requirements attach to reduction
projects in cellular and molecular neuroscience. No serious LTP researcher is
unaware of the results Schouten and de long cite . In fact, we'll see in the next
chapter that the molecular mechani sms of LTP are not even unique to
neurons , much less to neurons exclusively involved in memory consolidation!
And yet the "memory consolidation-molecular mechanisms of LTP link"
continues to thrive . Clearly , some other sense of reduction is at work than the
one Schouten and de long (1999) find lurking in some neuroscientists'
writings. It is this other sense that I want to bring to the surface by looking in
some detail at the actual neuroscientific accomplishments and assertions, as
they have been presented in the primary experimental literature (as opposed to
review papers, critical survey s, and the like),3
2 TWO PSYCHOLOGICAL FEATURES OF MEMORY

CONSOLIDATION
Con sider first some feature of memory con solidation revealed by
experimental psychology. These features are the explanatory targets of the
reductionistic search for molecular mechanisms. Sinc e the seminal research of
Ebbinghaus, MUller, and Pilzecker more than a century ago, and then
47
elaborated by William James in his classic Principles of Psychology (1890though James employed his own terminology), psychologists have distinguished short-term from long-term memory. Short-term memory is transient,
lasting anywhere from the immediate present ("iconic memory") up to several
minutes ("working memory") to perhaps an hour or more (with rehearsal),
and is very susceptible to interference by distraction . Long-term memory is
stable, lasting for weeks, months, years, sometimes even decades , resistant to
distraction, and (typ ically) is induced only with stimulus repetition (which can
include extended rehearsal)."
Two features of the "conversion" of short-term memory into longterm memory have been prominent in experimental studies for more than one
century. Ebbinghaus first showed that
stimulus repetition is necessary (in most instances) to induce stable,

stronger, longer-lasting memories.
He presented subjects with novel letter strings, composed of

"syllables" of one con sonant followed by one vowel followed by another
consonant.' To minimize existing "associations" that subjects might have with
particular strings, which could artifactually influence their retention and
recall, he constructed his sequences randomly and discarded strings that
formed actual (German) words. He thus constructed a novel " language" out of
over 2300 syllables, from which for a given experimental run he chose 7 to 36
to construct lists of varying lengths. Subjects heard members of the lists
spoken at a rate of 150 per minute during training and then had to recall the
syllables that occurred on the lists at various times after training. Ebbinghaus
varied the number of times a given list was repeated during a training session,
from anywhere between 8 and 64 times. He discovered a nearly linear
relationship between the number of repetitions of the list during learning and
retention tested on the following day .
Second, as MUlier and Pilzecker also first demonstrated experimentally more than one century ago ,
the conversion of information from short-term memory to long -term

memory can be disrupted by retrograde interference, by distractions
introduced after the initial items have been learned and sto red in
short-term memory.
They coined the phrase consolidation period to refer to the time

needed for the short-term " memory trace" to achieve stable long-term form.
Experimental psychologists ignored this in itial work on retrograde interference for more than a half-century. Clinicians, on the other hand, were quite
48
familiar with this effect, knowing that a head injury often era sed recall of
events that pre ceded the blow. Retrograde interference was then the subject of
two very influential studies in the late 1940s. Consider one of these, still cited
in the consolidation literature today .
Duncan (1949) had noticed during a prior study of electro-convulsive
cerebral shock on "new and old learning" that the time the shock is delivered
relative to the time of training had a significant influence on its effect. To
study this "retroactive inhibition," he used a conditioned aversion procedure
that rats learn easily. Rats were introduced into a two-compartment box with
an open doorway between the compartments and initially allowed to explore
both compartments for two minutes. The "test" compartment had an electric
floor grid. The "safe" compartment did not have a floor grid , but was brightly
illuminated. Because of their light-phobic nature, rats quickly retreat back into
the test compartment during this initial two-minute exploration. At the end of
the two minutes, the electric floor grid was activated, delivering a mildly
aversive foot shock. Rat s quickly ran into the safe compartment for the ten
seconds that the grid was charged. They were removed from the box
immediately.
Rats were then divided into one of eight experimental groups and a
control group. Members of each exp erimental group received a cerebral
electroshock, delivered through ear electrodes, that was sufficient to induce
grand mal convulsions on each application. Each experimental group received
the cerebral shock at a specified time after being removed from the training
box : 20 seconds, 40 seconds, 60 seconds, 4 minutes, 15 minutes, 1 hour, 4
hours , or 14 hours . (These specific times were chosen to make for easy
graphical representation of the data on a logarithmic scale.) Control animals
underwent the training but received no subsequent cerebral shocks. After this
initial training trial, each rat was returned to the test compartment once a day
for 17 days. Ten seconds after the rat was placed in the test compartment, the
electric grid was charged for ten seconds and rats received the foot shock if
they had not already run into the safe compartment. The rat was then removed
from the box and received the appropriate cerebral shock, depending on
which group it belonged. On the eighteenth day, no subsequent cerebral
shocks follow ed the test trial. Rats in the various groups were measured on
the average numb er of "anticipatory runs" they made into the safe
compartment during the ten seconds between being placed in the test
compartment and grid charging. Significance was measured by comparison to
the control averages."
Duncan's results spoke clearly in favor of a "consolidation" theory of
the effects of this retrograde interference, namely that "newly learned material
undergoes a period of consolidation or preservation. Early in this period a
cerebral electroshock may practically wipe out the effect of learning. The
49
material rapidly becomes more resistant to such disruption" (1949 , 44).

Compared to controls, only animals in the 20 second, 40 second, 60 second, 4
minute, and 15 minute groups showed statistically significant reduction in the
average number of anticipatory runs into the safe compartment prior to the
shock . (See Figure 2.1.) The 1 hour, 4 hour, and 14 hour groups were no
different than the control group even when averages were broken down into
three-day segments. Furthermore, Duncan found a steep and progressive
relation between the time after training that the cerebral shock (retrograde
interference) was delivered and the effect on learning and memory. The effect
lessened rapidly and progressively as more time elapsed.
A control experiment ruled out the possibility that this retrograde
interference was simply the result of the cerebral shock being a second
aversive CS-US pairing. By 1949, experimental psychologists knew that
training an animal on a second conditioning task soon after training on a first
inhibits learning of the first association. This possibility confounded the
interpretation that the retrograde cerebral electroshock interfered specifically
with neural consolidation mechanisms. Duncan (1949) trained rats on the
same conditioned aversion task described above, but instead of experimental
animals receiving cerebral electroshocks at some time subsequent to each
training episode, they received retrograde hind leg shocks (of the same
intensity as the cerebral electroshocks in the original study) . If the original
effect was simply one of inhibition based on a second CS-US pairing, then
experimental groups in the control study should demonstrate the same pattern
of behavioral inhibition as did the analogous groups in the original study.
They didn't. Only the 20-second group in the control study showed any
statistically significant inhibition. Every other experimental group of fifteen
minutes or less performed similarly to controls, who received no retrograde
leg shocks. And even the 20-second retrograde leg shock group was
significantly less inhibited in learning and remembering the conditioned
aversion task than their counterpart retrograde cerebral shock group.
Evidence from this control experiment supported the claim that the
retrograde cerebral shocks were inhibiting learning and memory by acting
directly on neural consolidation processes. Duncan concluded:
A consolidation period follows the completion of each trial in
the avoidance box. This period is less than 1 hr. and very
probably is not significantly longer than 15 min . The disruptive effect of an electro-convulsive shock, when applied
during the consolidation interval, produces a loss of retention,
which appears as slower learning. When the shock is
interpolated sooner after each trial , less time is allowed for
consolidation and there is greater interference with retention.
PHILO SOPHY AND NEUROSCIENCE
50
1
14
1 hr
C/)
controls
:::>
a:
>a: 1
le::(
0-
IZ
e::(
Statistically
significant up
to 15 minutes
ze::(
w
~
0
0
!J
LOG TIME
Figure 2. / . Average number of anticipatory runs into the "safe" box across all eigh teen trials as
a function of the trial-electro -convulsive shock time interva l expressed in logarithmic units.
Each point on the curve represen ts a different experimental grou p (indicated by the amoun t of
time separating the trial and electro -convulsive shock for each trial for each member of that
group) . Statistically significant differen ces compared to control were found up throug h the 15minute interval group. Graph created by Marica Bernstein. (See Duncan 1949, Figure I. )
When the shoc k is applied after consolidation has ceased, no

retent ion loss is found. (1949, 43)
Although Duncan ' s study occ urred during the zenith of meth odological beha viorism in ex perimental psychology, his results fit naturally with
a more recent "cognitivist" interpretation. Notice that he himself spea ks of
"retenti on, which app ea rs as slower learn ing" in the above passage, which
51
suggests a cognitivist interpretation. This point is important because a reduction of Duncan's results to underlying molecular mechanisms would thus be a
reduction of a "genuinely cognitive" psychological process, according to the
currently acknowledged "mark" of this distinction. The two key features of
consolidation-the need for repetition/rehearsal and its susceptibility to
retrograde interference-are naturally thought of in terms of memory
("mental") representations and operations involving their contents. Rehearsal
amounts to "being presented" or "calling to mind" repeatedly the representational content of the stimulus presentation now available in short-term
memory. Retrograde interference disrupts the needed maintenance of the
memory representation. Such accounts fit with the widely accepted "mark of
the cognitive," as requiring explanations that advert to operations over the
contents of cognitive representations. "Autonomists" about the cognitivepsychological from the neural should thus urge that explaining the memory
consolidation switch is a job for experimental psychology, not for neuroscience-and certainly not for the branch of neuroscience that deals with the
biochemistry of receptor proteins, intracellular second messenger signaling
pathways, neuronal gene expression, and the like .?
Unfortunately (for psychological autonomists, at least) subsequent
psychological research on memory consolidation quickly hit explanatory
limits. A few features of consolidation proved addressable at the psychological level: its variable time courses for different memory items; the nature
and time courses of effective versus ineffective retrograde interference; and
the variable amount of repetition/rehearsal required to consolidate different
items . Most glaringly, experimental psychologists were never able to explain
mechanistically the consolidation process or switch. Neuroscientists, on the
other hand, quickly made impressive progress. Specific pharmacological
manipulations dating back nearly forty years have produced animals (mammals included) with intact learning and short-term memory capacities but
profoundly deficient long-term memory. Such manipulations seem to disrupt
selectively the consolidation switch. Davis and Squire's (1984) influential
review paper has now been in print nearly twenty years, yet it still describes
how fruitful neuropharmacological manipulations were in elucidating the
initial mechanisms of the consolidation switch. Over the past decade, in
keeping with biotechnology's impact, experimenters have carried out these
manipulations using genetic knockout and transgenic rats and mice . These
technologies resolve nagging methodological worries that plague pharmacological experiments and the proper interpretation of their results." The
outcome of these neuroscientific investigations are current models of the
molecular mechanisms of LTP and the measurable behavioral effects on
learning, short-term memory, and long-term memory by manipulating them
directly.
52
3 LTP IS DISCOVERED
3.1 From Hebb's neuropsychological speculations, 1949, to
Norway, 1973
The current approach to learning and memory in mainstream
neuroscience follows a lead first developed explicitly by psychologist Donald
Hebb. (I say "explicitly" because the great Spanish neuroanatomist, Santiago
Ramon y Cajal , suggested this lead more than one-half century ea rlier in an
address to the British Royal Society.") In his classic book, The Organi zation
of Behavior (1949) , Hebb recommended that we think of learning and
memory in terms of synaptic strength and plasticity: the changeable effect that
a given neuron exerts on electrochemical membrane potentials in neurons
with which it shares an active synapse. He begins with a "bald assumption
about the structural changes that makes lasting memory possible ... in brief, ..
.that a growth process accompanying synaptic activity make s the synapse
more readily traversed. ... An intimate relationship is postulated between
reverberatory activity and structural changes at the synap se" (1949 , 60). On
purely theoretical grounds, Hebb postulated a "dual trace " synaptic mechanism for memory: first a "t ransient , unstable reverberatory trace ," followed
by a time-dependent, reinforcing "more permanent structural change" (1949,
62). As we will see, Hebb's postulation was remarkably prescient. After
briefly discussing the then -current neurobiological concept of "synaptic knob
growth," Hebb insists that "the details of these histological speculations are
not important except to show ... that the mechani sm of learning discussed in
this chapter is not wholly out of touch with what is known about the neural
cell" (1949, 65) . But in his final analysis, Hebb was honest: "As neurophysiology, this and the preceding chapter go beyond the bounds of useful
speculation. They make too many steps of inference without exp erimental
check" (1949 , 79). Of course, that was more than one-half century ago . Since
Hebb 's speculations, LTP has emerged as a promising candidate for
experience-driven synaptic plasticity. LTP's story begins explicitly in
Norway, 1973, although activity-driven synaptic efficacy had been reported
anecdotally for at least one prior decade in the experimental electrophysiological literature. Three researchers in Per Andersen 's laboratoryTimothy Bli ss, Terje Lemo, and Anthony Gardner-Medw in-artificially
stimulated fibers in the perfo rant path of both anesthet ized and unanesthetized
rabbits (reported in Bliss and Lerno 1973 and Blis s and Gardner-Medwin
1973, respectively). This fiber bundle is a collection ofaxons that synapse on
53
granule cells in the dentate region of the hippocampus. Available

electrophysiological techniques at that time enabled them to measure
the amplitude of the population excitatory postsynaptic potential

(EPSP) in granule cell dendrites that synapse with perforant path
axons, indicating depolarization in granule cells generated by
electrical stimulation of the perforant path;
the amplitude of the population spike in granule cells, measured in the
cell body layer of the dentate region, indicating discharge induced in
granule cells by perforant stimulation;
the latency of the population spike measured as time from perforant
path stimulation until initial peak discharge.
These descriptions may be going beyond some readers ' comprehension. And the "real neuroscience" example I'rn describing in this chapter
depends on the details of these and more recent experimental studies. So for
the "neurophysiologically challenged," let's begin with some basic cellular
neuroscience.
3.2 Some basic cellular neuroscience 10

Neurons-nerve cells-eonstitute one of two tissues in nervous
systems. I I They possess a characteristic lipid bi-Iayer cell membrane that can
propagate an electric charge down its length. This is due to two biophysical
features. First, there is an unequal distribution of various ions inside and
outside the cell membrane, most notably, positively charged sodium (Na +),
potassium (K+), and calcium (Ca 2+), and negatively charged chloride (Cl) and
various amino acids. Second, the activity of selectively permeable channel
proteins permits some of these ions to flow across the cell membrane. Since
the negatively-charged amino acids remain inside the cell (being too large to
cross through open channels), and in the "resting" membrane state there are
far more Na+ ions in the extracellular space, a neuron's membrane has a
negative resting charge or potential, typically around -70 millivolts (mV)
(although this value has some variation across different neuron types). This
potential is maintained by
1. the combined equilibrium potentials of the four key ions, and
2. active transport mechanisms.
The first is the membrane electric potential, for each type of ion, at
which it has an equal probability of flowing across the membrane in either
54
direction. The second are proteins that transport ions across the cell
membrane against their diffusion gradients or electrostatic pressure. "Active"
refers to the need for energy to drive this transport, derived from internal
cellular sources. Sodium-potassium pumps are the most prominent of these in
neurons. They actively transport Na+ out of and K+ into the neuron, despite
greater concentration of Na+ outside the cell (i.e. , against its diffusion
gradient), the negative resting potential inside the cell (i.e ., against electrostatic pressure-"opposite charges attract, identical charges repel ") , and the
greater concentration of K+ ions inside the cell (i.e., against its diffusion
gradient).
Where we start to describe neuron physiology is somewhat arbitrary.
We'll start at the axon hillock, with the generation of the action potential (or
"spike") (Figure 2.2). In a typical neuron, this structure is located where the
single axon emerges from the soma (cell body) . The soma houses the cell
nucleus, including the genes and gene expression machinery (more on this
later) and the protein production machinery (outside the nucleus, in the
cytoplasm). The axon is the neuron's "output line," through which it
communicates with other neurons, muscle fibers , or other biological tissues.
The axon hillock is dense with voltage-gated ion channels, especially ones
selectively permeable to Na". Voltage-gated ion channels, proteins floating in
the lipid bi-Iayer cell membrane, can change from closed to open with
changes to the electric potential in nearby membrane patches. An ion channel
is a configured protein. At resting membrane potential, its three-dimensional
configuration prohibits the flow of ions across the cell membrane (although
some leakage occurs regularly, in both directions: hence one need for Na+-K+
pumps, mentioned just above, to maintain the membrane's resting potential).
But when the membrane potential in surrounding patches changes, an electric
field is created that alters the configuration of these channels, opening them to
ion influx or efflux. Channels selective for Na+ are particularly important.
When opened, Na+ rushes into the neuron along its diffusion gradient and due
to electrostatic pressure, since there is more Na+ outside the cell and the inside
is negatively charged relative to the outside. The net result is a sudden
positive change in localized membrane potential.
If this potential exceeds the threshold of excitation, around -60mV in
typical neurons, voltage-gated Na+ channels open and the membrane potential
in a local axon patch quickly "spikes" to around +50mV. 12 (See Figure 2.3.)
About halfway through this electric "spike," voltage-gated K+ channels (with
a higher voltage threshold than Na+ channels) begin opening. With the
membrane potential now far above K+'s equilibrium potential and a higher
concentration by the forces of diffusion and electrostatic pressure. Within 1
REDUCTION- IN-PRACTICE
__--,1---
55
Cell body
Figure 2.2. A schematic neuron , indicating various subcellular components. Figure created by
Marica Bernstein.
millisecond (msec) (111000 of a second) after reaching their threshold of

excitation, Na+ channels begin to close . They undergo a brief refractory
period (about 1 msec) , during which no more Na+ enters the cell. K+ channels
remain open , and K+ efflux returns the local membrane potential back toward
resting value (-70mV). By the time these channels close, the membrane
potential has "overshot" its resting potential and become hyperpolarized due
to the accumulation of K+ outside of the membrane. Resting membrane
potential is quickly re-established by the diffusion of K+ throughout the
extracellular fluid and the active transport of Na+ ions out of and K+ ions into
the cell (the principal role of the Na+-K+ pumps) . The entire process takes less
than 3 msecs , after which that localized patch of axonal membrane is ready to
"spike" again.
The special feature ofaxons is the conductance of action potentials
down their entire length. The first basic "law" of axonal conductance is the
all-or-none law . An action potential either occurs down an entire axon or it
does not. Once generated in a localized patch, an action potential transmits
down the axon to its end . An action potential remains the same size (same
peak voltage value) during conduction. This is different from decremented
conduction in passive electric cables , in which signal size decreases with
distance from generating source owing to leakage and resistance. This
56
+50
>E
---co
~
Q)
(5
n,
Q)
.0
E
Q)
::2:
-50
Time (msecs)
Figure 2.3. A schematic neuron action potent ial, graphed as a function of membrane potential
to time. See text for a discussion of the action potential' s bioph ysical mechani sms. Figure
created by Marica Bernstein.
difference is due to the voltage-gated Na+ channels that generate action

potential depolarization. A localized depolarization creates an electric field
that drives nearby axonal membrane patches over their thre sholds of
excitation, opening voltage-gated Na+ channel s and generating an action
potential there. This occurs all the way down the length of the axon
membrane.
Conductance of action potentials is facilitated by myelination, a
feature of many (but not all) axons. Two types of glial cells-oligodendrocytes in the central nervous system and Schwarm cells in the peripheral
nervous system-wrap tightly around axon s, leaving no extracellular fluid in
between. Myelin ation leaves only small patch es of axon membrane, at the
nodes of Ranvier, in contact with extracellular fluid (see Figure 2.2 above).
These are the only places on myelinated axons where Na+ influxes through
REDUCTION-IN- PRACTICE
57
open voltage-gated Na+channel s. Myelinated axons passively conduct electric

current from node to node, exceeding the threshold of exc itation to re-trigger
an action potential at each. This so-call ed salutatory conductance has two
advantages over continuous action potential conductance in unmyelinated
neurons. First, it is less energy intensive. Active transport of Na+ and K+ to
reestablish the action potential in each local patch requ ires a significant
expenditure of a neuron's metabolic resources (measured up to 40 % in some
recent studies). Since Na+ enters everywhere along an unmyelinated axon,
these pumps must also be located all along the axon's length. In a myelinated
axon, however, these pumps need only be located around the nodes of
Ranvier. The second advantage is speed. Passive conductance depends on a
cable's diameter. The thicker the cable, the faster is its passive conductance
velocity. However, myelin serves as an insulator, which also increases
conductance velocity. The fastest myelin ated axon conducts action potentials
at almost four times the velocity of the much larger squid giant axon , despite
the latter 's being 25 times the diameter of the former.
Action potentials in motor neurons that synapse directly on muscle
fibers control the intensity of muscle contractions. Action potentials in
sensory neurons reflect stimulus intensity. It is not a far stretch from these
facts to the general idea , prominent in current neuroscience, that action
potentials are the currency of neuronal information exchange. But if action
potentials in individual neurons are all-or-none and constant in size , how can
they represent variable parameters? The current idea is that variable
information is represented by variable action potenti al (or "firing") rates. A
higher firing rate-more action potentials per time unit-in a motor neuron
produces a stronger muscle fiber contraction (by a mechanism we'll review
momentarily). A stronger stimulus produces a higher action potential rate in
selected sensory neurons. This has been dubbed the "rate law ," which
complements the all-or-none "law" (e.g. , Carlson 1996,40-41).
So far we've only sketched half of the story of neuron physiology,
namely neural conductance. We turn now to the process of neural transmission. We left off with the ali-or-nothing act ion potential reaching the end
of the axon. What happens next? Axons typically branch into numerous terminals (see Figu re 2.2 above). There the arriving action potential changes the
configuration of voltage-dependent Ca 2+ channels, opening them. Ca 2+
influxes into the axon terminal by the forces of diffusion and electrostatic
pressure. There is a far higher concentration of Ca 2+ in the extracellular fluid,
and despite the temporary depolarization of the membrane produced by the
act ion potential , Ca 2+ 's equilibrium potential is far highe r than the membrane
voltage at the peak of the action potential. Inside the terminal , Ca 2+ binds with
various proteins to bind the membranes of vesicles containing the neurotransmitter substance to the terminal membrane, producing a fusion pore through
58
both that permits exocytosis of the transmitter substance. Released transmitter

molecules diffuse passively into the synaptic cleft, the tiny space separating
the pres ynaptic axon terminal from its postsynaptic target (see Figure 2.2
above) . The postsynaptic cell, be it a neuron , a muscle fiber, or any other type
of cell that synapses with neurons, contains receptors, proteins with a
configuration that permits the binding of the transmitter molecule. Depending
on the nature of the postsynaptic receptor, transmitter binding initiates a
configurational change to the receptor protein that either initiates or inhibits
electrochemical activity in the postsynaptic cell. If the postsynaptic target is a
muscle fiber, transmitter binding either contracts the fiber or inhibits contraction by initiating well-understood biochemical activity. If the target is another
neuron, transmitter binding either depolarizes or hyperpolarizes the membrane
in the vicinity of the bound receptor.
Neurons contain two types of postsynaptic receptors. The simplest is
the ionotropic or "fast" receptor. Transmitter binding alters the configuration
of the receptor protein, opening a selective ion channel directly through it.
Excitatory ionotropic receptors are selective for either Na+ or Ca 2+ . Ion influx
(via forces of diffusion and electrostatic pressure) produces a small depolarization in the nearby membrane (far less than the threshold of excitation for
generating an action potential). This occurrence is dubbed an excitatory
postsynaptic potential (or current), or EPSP (EPSC). Inhibitory ionotropic
receptors are selective for Cr. At resting membrane potential, open cr
channels have little effect, since the force of diffusion is offset somewhat by
electrostatic pressure. There is more cr outside the cell relative to inside, but
resting membrane potential is very close to Clis equilibrium potential. When
the membrane is depolarized, however, the forces of diffusion and
electrostatic pressure drive cr influx, hyperpolarizing the local membrane
potential. This is an inhibitory postsynaptic potential (current), or IPSP
(IPSC).
The other type of receptor, and by far the more prominent, is the
metabotropic or "slow" receptor. Transmitter binding here does not directly
open an ion channel. Instead, the reconfigured receptor protein activates
intracellular G proteins. One subunit of an activated G protein (the a subunit)
breaks away and binds to a distant ion channel, opening it to generate an
EPSP or IPSP. Or, in an even more complicated scenario (which we'll see in
detail later in this chapter), the a subunit binds to an enzyme protein,
changing the latter's configuration to produce a second messenger: a protein
product that translocates to other parts of the cell to initiate biochemical
reactions there, including gene expression in the neuron's nucleus.
Following transmitter binding to postsynaptic receptor proteins, a
number of events occur around and in the presynaptic terminal. Active
transport mechanisms can reuptake transmitter molecules back into the
59
terminal. This can happen either to the unaltered transmitter molecule or

following its enzymatic degradation into inactive metabolites. Reuptake is a
principal mechanism for terminating neurotransmitter actions , clearing the
cleft for transmitter release driven by the next action potential. At high
concentrations, transmitter molecules can diffuse through the extracellular
fluid to bind to autoreceptors, protein receptors on the presynaptic neuron
(not necessarily on the terminal). As a general rule, autoreceptors are part of
negative feedback loops that slow or shut down transmitter synthesis or
release by the presynaptic neuron. Glial cells also actively take up transmitter
molecules from the cleft. Terminals from modulatory neurons can synapse
onto primary presynaptic terminals (so-called axo-axonal synapses), releasing
transmitters that bind to receptors on the presynaptic neuron to affect
transmitter release (often through direct actions on action potential
conductance or voltage-gated Ca 2+ ions).13
On the postsynaptic side, EPSPs and IPSPs occur throughout the
dendrites and soma. These local potentials interact along the membrane as
they travel in both directions from their sites of origins. Interacting depolarizations and hyperpolarizations tend to cancel each other out, but similar
charges summate both spatially and temporally. These are the processes of
neural integration. When these interactions sum to the threshold of excitation
at the axon hillock, an action potential is generated and propagated down the
axon's length. Unlike the action potential, propagation down dendrites and
soma is decremental. But owing to integration, the rate at which an axon
fires-and hence represents information or controls muscle contractions-is
determined by excitatory and inhibitory effects produced in its dendrites and
soma .
Even the simplest spinal reflex involves thousands of interacting
neurons-sensory, intraspinal, and motor. Thousands more are involved in
even the simplest cortical inhibition of a spinal reflex. These numbers
increase dramatically as we move to more complicated sensations, cognition,
and behaviors. But every neuron involved is operating along the basic
principles sketched in this subsection. These are, simply, the facts of the
matter.
Let's pause here for a reductionist epiphany-in part to set the stage
for what is coming, but also to remind readers why we are wad ing through
neurobiological detail. If action potential rate is the currency of neural
causation and information exchange, then the only wayan event can elicit
neural change is by affecting the processes that underlie action potential
generation in individual neurons. That is where the rubber meets the road.
Even those who champion "distributed" processing and "population coding
patterns" must realize that neural populations are composed of individual
neurons firing action potentials at variable rates. To affect, e.g., a population
60
spiking frequency, an event must affect the individual neurons comprising the
population. To do that, it must affect the processes governing action potential
generation in individual neurons. In other words, it must affect the opening
and closing of voltage-gated Na+ and K+ channels in the axonal membrane
and the activity of Na+-K+ pumps that re-establish membrane potential,
readying the axon for the next spike . Events that aren't "transducible" to that
level of biochemistry and biophysics cannot affect neuronal activity. There
are other "coding strategies" that are used in neuronal population studies
besides frequency of action potentials." But a population value of any sort is
a function of that value in the individual neurons comprising it. And those
neurons are nothing more than organized bags of molecular mechanisms for
getting ions across selectively permeable membranes. As we will see as this
chapter progresses, even pharmacological agents and the gene expression
involved in synapse restructuring and formation , cell death , and so on , exert
their effects on neurons at this level.
So you think, e.g., that poverty causes criminal behavior? Well,
criminal behavior is (at the very least in part) a matter of orchestrated muscle
fiber contractions. These contractions result from the release of specific
molecules (acetylcholine, Ach) by motor neurons onto the endplates of
individual skeletal muscle fibers comprising the affected muscle; and this
release is controlled by differential spiking rates in individual motor neurons
synapsing on specific fibers. A lot of cellular and molecular pathways affect
motor neuron firing , many of which interact with other bodily systems. But all
these features ultimately must affect the membrane proteins whose configurations at any given instant determine whether action potentials will be
generated, and hence whether molecular transmitter substances will be
released into neuromuscular junctions. I repeat the slogan I aired in the
previous paragraph: for all their molecular biological and biochemical
complexities, neurons are at bottom bags of processes that facilitate or hinder
ions crossing selectively permeable membranes. So if, e.g., poverty causally
influences behavior, it must be "transducible" down to this level of
biochemical mechanism. This is causal-mechanistic "reductionism," minus
philosophy of science's bells and whistles . Given what we know now about
how neurons work (and my sketch so far has been at the most cursory level of
cellular neuroscientific detail) , if you deny this , you really are a causal dualist
about behavior. "Not that there's anything wrong with that," to quote a
repeated punch line from a famous Seinfeld episode, but theorists should own
up to their commitments. If you reject a dualism of causal properties, for
whatever reason , you cannot deny this implication.
61
3.3 Back to Norway, 1973 15

Armed with this background (and its reductionistic implication), we
return to Per Andersen's lab in Oslo, 1973. Bliss and Lerno (1973) inserted
stimulating electrodes in the vicinity ofaxons forming the perforant path in
anesthetized rabbits and recording electrodes into either the dendrite layer or
the cell body layer of the hippocampus dentate gyrus . Perforant path axons
synapse directly onto dendrites of dentate gyrus neurons. Conditioning trains
delivered through the stimulating electrode consisted of either 10-20 electrical
pulses/second for 10-15 seconds or 100 pulses/second delivered for 3-4
seconds. In 15 of their 18 rabbits, Bliss and Lomo (1973) found potentiated
responses in dentate gyrus neurons to a single stimulating pulse delivered
after a conditioning train , as compared to responses to the same pulse
delivered prior to the conditioning train. This potentiation was measured as
increased amplitudes in both the population EPSP measured in the dentate
dendrite layer and the population spike measured in the dentate cell body
layer, and decreased latency to population spike in the latter. These
potentiations were measured from 30 minutes up to 10 hours after
conditioning stimulus trains. All parameters were potentiated in nearly 30 % of
all experimental trials. Reduction in population spike latency was the most
common result, potentiating in nearly 60% of all experimental trials. Amplitude of population EPSP increased in over 40% of all experimental trials, and
amplitude of population spike potentiated in 40 %.16
Bliss and Gardner-Medwin (1973) obtained similar results in a
follow-up study using chronically implanted stimulating and recording
electrodes in unanesthetized, alert, active rabbits. After single trains of
conditioning stimulation at 15 pulses/second delivered via the stimulating
electrode to a region of the perforant path , they measured long-lasting
potentiation to single stimulating pulses on all three parameters on 41 % of all
experimental trials . Measurable potentiation lasted from 1 hour to 3 days in
these preparations, and increasing the number of conditioning trains induced
long-lasting facilitation for increasing durations. By demonstrating potentiation in unanesthetized animals, Bliss and Gardner-Medwin (1973) showed
that the effect did not depend on neurons being in a depressed state.
Bliss and Lomo (1973) were quick to note potential implications for
memory research. They remind us that "the perforant path is one of the main
extrinsic inputs to the hippocampal formation, a region of the brain which has
been much discussed in connection with learning and memory" (1973, 355).
Clearly they were aware of lesion and neuropsychological research that had
already linked the hippocampus to memory , as they cite two important review
papers in their bibliography (Douglas 1967 and Olds 1972). In conclusion,
they assert that "our experiments show that there exists at least one group of
62
synapses in the hippocampus whose efficiency is influenced by activity which

may have occurred several hours previously-a time scale long enough to be
potentially useful for information storage" (1973 , 355) . Similar results by
Bliss and Gardner-Medwin in awake, alert , mobile animals justified their
conclusion that "it is at least possible that [the then-unknown mechanisms of
this effect] could underlie some forms of plasticity under normal conditions in
the hippocampus" (1973, 373). The cellular investigation of learning and
memory now had a new focus.
Since this start, LTP has been the subject of much in vitro and in vivo
research. Besides the memory-like characteristics that Bliss, Lemo , and
Gardner-Medwin discovered, results have yielded quantifiable data about
LTP' s enhancement by repetition of conditioning trains (reminiscent

of the consolidation effects of stimulus repetition and rehearsal).
Pharmacological and physiological manipulations have also revealed that LTP

is both
selectively blocked by treatments that inhibit certain types of longterm memory (measured behaviorally), and
induced by physiological manipulations that augment types of Jongterm memory (measured behaviorally).
The experimental literature on LTP and its connection with behavioral work on memory, landmarked by excellent review papers, is a prime
example of first-rate science.
4 MOLECULAR MECHANISMS OF LPT: ONE CURRENT

MODEL
A recent account of LTP that links intracellular molecular
mechanisms to measurable behavior is especially exciting. Over the past two
decades, much cell-physiological research on LTP in mammals has shifted
from the perforant pathway to the hippocampal Schaffer collateral pathway.
The Schaffer collateral pathway is a bundle ofaxons from cells in the
hippocampal CA3 region that project excitatory synapses to the hippocampal
CAl region. The connection between the hippocampus and long-term
memory storage and access is now even more direct than it was in 1973.
Bilateral hippocampal ablation in nonhuman primates produces little deficit in
initial learning and short-term recall, but profound deficits on certain types of
63
long-term recall tasks. It has been proposed as an animal model of human

global amnesia, which also results from bilateral damage to hippocampus (and
some surrounding tissue in the medial temporal lobe) . Medial temporal lobe
amnesics, like their experimental animal counterparts, have intact learning
and short-term memory but profoundly deficient long-term memory for
"declarative" items (Squire 1987).17
4.1 Early phase LTP

It is now common to distinguish distinct stages of LTP, based on their
method of induction and temporal stability (Izquierdo and Medina 1997).
Early phase LTP (E-LTP) begins immediately after a single high-frequency
electric pulse train is delivered to Schaffer collateral fibers. Enhanced
responses to subsequent pulses persist from one to three hours in CAl
neurons containing potentiated synapses. E-LTP induction and maintenance
does not require gene expression or new protein synthesis. Nguyen et at.
(1994) elegantly demonstrated this using hippocampal slices (400 microns
thick) through the CAl region from young (5-week old) rats. First they
established baseline field EPSP response rates to Schaffer collateral
stimulation prior to LTP induction. Then they induced LTP with three high
frequency pulse trains through the stimulating electrode in both control and
experimental slices . Immediately after these pulse trains ended, experimental
slices were bathed with a solution containing a nonspecific gene transcription
inhibitor, actinomycin D (ACT D). ACT D is a polypeptide antibiotic that
infuses into DNA molecules, forming a stable drug-DNA complex. This
inhibits DNA-mediated RNA polymerase activity, blocking transcription of
messenger RNA, an early step in gene expression. IS ACT D subfusion had no
statistically significant effect on LTP for more than two hours nor on baseline
field EPSPs for greater than three hours (Nguyen et al. 1994, Figure l A, B).
This result is especially interesting because it has been known for some time
that short -term memory in behaving animals is also relatively impervious to
gene transcription and protein synthesis inhibitors (Davis and Squire, 1984).
E-LTP induction involves two types of ionotropic ("fast") postsynaptic receptors for glutamate, the principal excitatory neurotransmitter in the
mammalian central nervous system. Presynaptic terminals of Schaffer collateral fibers stimulated by the conditioning pulse train release increased
amounts of glutamate due to the enhanced frequency of action potentials.
Transmitter molecules bind to postsynaptic AMPA receptors (a-amino-3hydroxy-5-methyl-4-isoxazole proprionic acid) (Figure 2.4A). Glutamate
binding changes the configuration of the AMPA receptor protein , opening a
channel through it selective for Na+. Na+ rushes into the cell through this
64
opened gate via its diffusion gradient and electrostatic pressure, producing
enhanced depolarization (positive current) of membrane potential in the
vicinities of bound receptors/open Na+ channels (all by the biophysical
properties outlined in the previous section).
Enter next a second type of postsynaptic glutamate receptor, the
NMDA (N-methyl-D-aspartate) receptor (see Figure 2.4A) . At resting and
weakly depolarized membrane potentials, NMDA receptors are blocked to
glutamate's influence by voltage gated magnesium ions (Mg 2+) embedded
within the protein's configuration. Under conditions of sufficient membrane
depolarization, via activated AMPA receptors in the near vicinity, the Mg 2+
block pop s out, permitting glutamate binding. Bound glutamate changes the
protein's configuration to open a direct channel selective both for Na+ and
Ca 2+. Ca 2+ rushes into the cell via its diffu sion gradient and electrostatic
pressure. (Even though the membrane potential is positive due to nearby
AMPA receptor activity, Ca 2+,s equilibrium potential exceeds even highly
depolarized membrane potential.) At the same time , continued Na+ influx
through open AMPA receptors activates a family of intracellular enzymes, the
tyrosine kinases, inside the postsynaptic terminal. The se enzymes phosphorylate a subunit of the NMDA receptor protein, further enhancing the
channel 's Ca 2+ conductance (see Figure 2.4B). Phosphorylation is a process
by which a phosphate group (P0 4 ) gets attached to a protein. This changes the
protein' s three-dimensional configuration, leading to a variety of changes in
the protein's interactions within the cell. Phosphorylation (and de-phosphorylation) is a crucial step in many cell-biological processes; it will come up
often throughout the rest of this and the next chapter. In turn the increased
Ca 2+ levels in the postsynaptic cell set in motion a biochemical cascade
involving numerous intracellular enzymes and interactions (Soderling and
Derkach 2000). Ca 2+ bind s with calmodulin (CaM), increasing the intracellular level of the Ca 2+-CaM complex. This increase has two crucial effects.
First, it stimulates adenylyl cyclase molecules that convert adenosine
triphospate (ATP) into cyclic adenosine monophosphate (cAMP). Conversion
of ATP into cAMP is a principal source of energy that drives cellular
metabolic processes. But cells have also come to use the by-product of this
process, cAMP, as a "second messenger" intracellular signal. (Mo re on this
below.) Second, increased levels of Ca 2+-CaM complex stimulate the
autophosphorylation of Ca 2+/calmodulin -dependent protein kinase II
(CaMKII) into its active form. Phosphorylated CaMKII (P-CaMKII) in turn
interacts with the phosphorylated NMDA receptor, maintaining the open
receptor' s affinity for Ca 2+ influx (see Figure 2.4B).
From here, two separate chains of molecular events occur within the
postsynaptic terminal to maintain increased Na+ conductivity through bound
AMPA receptors. Fir st, the increased levels of cAMP bind to regulatory
65
REDUCTION-IN- PRACTICE
(b)
(a)
r.: Na
AMPA
receptor
- 10pS
EPSC
PSD
' :1: N~D:

"I receptor
AutophosphOl)l1ation
~Ca2'-CaM
~ ~ (basal)
<c::
PPI
Dendritic
spine
(e)
Weak stimulation
(E-LTP express ion)
. . :r.:- Na+\
. , ... .6....
12 rvt- ,
I ....
oes
...
...
PS'2.-~
~j ,~
EPSC ~
11
PKA
)'(.
cAftAPP'
Dendritic
spine
Figure 2.4 . Effects of mol ecul ar mechanisms of E-LTP. A. Postsynaptic conductance cap acity
through bound AMPA receptors under condition s of a weak stimulation to pres ynaptic afferent
and no potentiation. NMDA receptors remain unde r Mg 2+ blockade . B. Strong affe rent
stimul ation yields incre ased AM PA receptor bind ing, stron g membrane depo lari zation , release
of NMDA receptor blockade, glutamate bindin g to NMDA receptor. Ca2+ influx through open
NMDA receptors, and an intracellul ar biochemical cascade that increases conversion of ATP
into cAMP. C. Biochemical cascade yields AMPA receptors with a nearl y thre e-fold increase
in conductance capacity for Na+ The weak pres ynaptic stimul ation now yield s an enhanced
post synaptic response. The synapse rema ins potentiated due to these molecular mech ani sms for
rou ghly one to thre e hours. See text for full explanation . Reprinted from Trends ill
Neuro sciences, 23, T. Sod erlin g and V. Derkach , " Postsynaptic protein pho sphorylation and
LTP ," Pages 75 -80 , Cop yright (2000) with permission from Elsevi er Science.
66
subunits of protein kinase A (PKA) molecules (see Figure 2AB) . PKA

consists of two regulatory protein subunits and two catalytic subunits. When
bound to the regulatory subunits, the catalytic subunits remain relatively
inactive. But cAMP binding releases the PKA catalytic subunits, making them
active in cellular interactions. First they phosphorylate inhibitor 1 (II), which
in turn inhibits the activity of protein phosphotase 1 (PP 1). PP 1 dephosphorylates both CaMKII and the subunit of the NMDA receptor, returning them to
their inactive states . With this inhibitory enzyme inhibited, both CaMKII and
the NMDA receptor subunit remain in their active (phosphorylated) state.
Second, P-CaMKII in turn phosphorylates a subunit on the AMPA receptor,
nearly tripling its conductance capacity for Na+ (see Figure 2AC above). The
result is a temporary increase in the conductance capacity of AMPA
receptors. While in this potentiated state, a weak stimulation to the
presynaptic fiber-which in the basal state would produce only a weak
depolarizing current in the postsynaptic membrane-now produces by itself
an enhanced excitatory postsynaptic potential (EPSP) due to the amount of
Na+ that influxes through the potentiated AMPA receptors. This increased
EPSP in turn raises the probability of spatial and temporal current summation
at the axon hillock of the postsynaptic cell, and hence the probability that the
postsynaptic cell will fire an action potential in distal response to presynaptic
neurotransmitter release. This in turn raises the probability that its activity
will influence action potential rates in the neurons it synapses on ... all the
way to motor effectors that drive muscle contractions. (Bear in mind that LTP
is occurring in selected synapses all the way down this cortical pathway.)
Potentiated response in each affected synapse to weak presynaptic stimulation
and glutamate release sets off the entire NMDA receptor-coupled intracellular
biochemical cascade allover again, maintaining potentiation in each affected
synapse in the circuit. These molecular mechanisms of E-LTP persist from
one to three hours.
The molecular story of E-L TP grows even more interesting in light of
recent discoveries about retrograde neurotransmission, from post- back to
presynaptic terminals. The most prominent retrograde neurotransmitter of late
is nitric oxide (NO). The biochemical details of its activity are still being
unraveled, but the basic process and its influence on E-LTP are now
experimentally verified in cultured Schaffer collateral-CAl slices (Arancio et
at. 1996, Son et at. 1996) (Figure 2.5). Increased Ca 2+ influx through opened
postsynaptic NMDA receptors activates NO synthase, an enzyme that stimulates NO production from the amino acid l-arginine, Being a soluble gas, NO
readily diffuses out of the postsynaptic cell and into the presynaptic cell,
where it has effects only at active terminals. There it activates soluble
guanylyl cyclase and cyclic guanine monophospate-dependent protein kinase.
This presynaptic cascade is known to be involved in enhancing
REDU CTION-IN-PRACTICE
67
neurotransmitter (glutamate) release (Arancio et al. 2000). So the molecular

mechanisms of LTP not only make glutamate-binding postsynaptic AMPA
receptors better conductors of Na+ currents; they also generate a retrograde
signal that enhances presynaptic glutamate release. The combined result is an
even sharper increase in EPSPs at E-LTP affected synapses, persistent for up
to 3 hours. This temporal feature correlates nicely with experimental
psychological work on short-term memory, the labile, disruptable, transient
form of virtually every type of memory (Squire and Kandel 1999, chapter 6).
4.2 Late Phase LTP

One remarkable feature of LTP is its temporal duration. Affected
synapses can remain potentiated for hours to days. What molecular mechanisms "consolidate" and maintain LTP beyond the one-to-three hour duration
of E-LTP? Inducing late phase LTP (L-LTP) in the Schaffer collateral
pathway requires a series of conditioning pulse trains to the presynaptic
fibers. Notice right away that this is a phys iological laboratory analog of
stimulus repetition (or rehearsal), known from psychological studies dating
back over one century to be necessary for "consolidating" most short-term
memories into stable, durable, long-term memories. In affected synapses, this
increased presynaptic stimulation increases the rate and amount of glutamate
released by presynaptic terminals, and in turn the number of bound
postsynaptic AMPA receptors. The resulting increase in Na+ influx further
depolarizes the postsynaptic membrane, opening more voltage-gated NMDA
Ca 2+ channels and even more quickly raising the level of intracellular Ca 2+_
CaM complex. Up to this point, the molecular mechanisms of L-LTP are
identical to those of E-LTP; but next a new player arrives.
Modulatory interneurons become active, releasing a monoamine as
their neurotransmitter-dopamine (DA), in mammals-and synapsing directly
on CAl dendritic spines in close proximity to bound AMPA and NMDA
receptors (Figure 2.6). Their postsynaptic DA receptors are metabotropic. As
described in the previous section, these receptors do not open ion channels
directly, but instead engage the postsynaptic cell's metabolic machinery
through a second messenger. DA binding activates a G-protein, guanine
triphosphate (GTP), which couples to adenylyl cyclase to convert ATP into
cAMP (the latter is the "second messenger") (Bernabeu et al. 1997). As with
the molecular mechanisms of E-LTP , cAMP binds to the regulatory subunits
of PKA, rapidly increasing the level of free catalytic PKA subunits in the
postsynaptic cell. But this increase initiated by the second messenger
generates enough freed catalytic PKA subunits so quickly that the latter
68
translocate to the cell's nucleus and the "consolidation" events begin that
distinguish L-LTP from E-LTP.
Nitric Oxide
Synthase
Nit,*OXi~
(NO)
Ca 2 + Calmodulin
Cytoplasm
Figure 2.5. Mechanisms of postsynaptic nitric oxide (NO) production and retrograde
transmission, driven by the molecular mechani sms of E-LTP induction. NO, a gas, diffuses
across the cell membrane back into the presynaptic neuron, where it increase s glutamate release
rate through a G protein intracellular pathway . This increased glutamate release falls onto
already-potentiated AMP A receptors. Reprinted from Cell, 88, T. Abel et al., "Genetic
demonstration of a role for PKA in the late phase of LTP and in hippocampus-based long-term
memor y," pages 615-626, Copyright (1997), with permiss ion from Elsevier Science.
As with the induction of long-term memory in behaving animals (Davis and

Squire 1984), L-LTP induction requires new gene expression and protein
synthesis. Consider again the Nguyen et al. (1994) study discussed in the
previous subsection . Immediately after multiple pulse trains were deliv-ered
through the stimulating electrode, hippocampal slices were dunked into either
ACT D or a different transcriptional inhibitor, 5, 6-dichloro-l-~-D
ribofuranosyl benzimidazole (DRB) with a different mechanism of action on
RNA polymerase activity. Activity following single pulses returned levels
MAP
69
Nucleus
fll fI ~
W Q!I'
PKA
cAMP......... /Yf
......
R(AB)
Regu lators
Effectors
Adenylyl
Cyclase
Figure 2.6. Early steps in the molecular mechanisms inducting L-LTP. Increased amount of
cAMP molecules bind to the regulatory subunits of more PKA molecules, enabling freed
catalytic PKA subunits to translocate to the neuron's nucleus. See text for detailed discussion .
Reprinted from Cell, 88 , T. Abel et al., "Genetic demonstration of a role for PKA in the late
phase of LTP and in hippocampus-based long-term memory," pages 615-626, Cop yright
(1997), with permission from Elsevier Science .
within three hours, Control slices receiving the multiple pulse trains without
either gene transcriptional inhibitor remained potentiated well beyond this
time, Recall from the previous subsection that these nonspecific transcriptional inhibitors had no effect on E-LTP induction.
This experiment does not suggest which new genes and proteins are
necessary for L-LTP induction ; but one year earlier Frey et al. (1993) had
published an experimental report that took its departure from previously
70
published reports (mainly from Frey and his colleagues) that L-LTP was
blocked by dopamine receptor D 1 antagonists." Since the bound D 1 receptor
was known to stimulate postsynaptic adenylyl cyclase, yielding an increase in
cAMP and freed catalytic subunits of PKA, Frey et al. (1993) explored the
hypothesis that L-LTP induction depended on PKA activity. They inhibited
PKA activity in hippocampal Schaffer collateral-CAl slices using Rp-cyclic
adenosine 3', 5' -monophosphorothioate (Rp-cAMPS), which permeates the
cell membrane and competes with cAMP for binding sites on PKA regulatory
subunits. Once bound, however, Rp-cAMP does not release PKA catalytic
subunits. Applying Rp-cAMPS fifteen minutes before inducing LTP had no
significant effect on normal synaptic transmission for up to three hours and
only a small negative (statistically significant) effect on E-LTP induction and
maintenance 60 minutes after LTP-inducing stimulus trains . However, L-LTP
in treated slices was completely blocked, measured as both field EPSP and
population spike activities at three hours and longer after inducing stimuli
(Frey et al. 1993, Fig. 1). The time course ofL-LTP blockage by this selective
PKA inhibitor closely matched those obtained using the nonselective protein
synthesis inhibitors, suggesting that both results came from inhibiting the
same mechanism.
They next applied a membrane-permeable cAMP analog, Sp-cyclic
adenosine 3', 5' -monophosphorothioate (Sp-cAMPS), to hippocampal slices.
This agent activates PKA. Even in the absence of conditioning pulse trains,
this manipulation yielded an L-LTP time course similar to that induced in
control slices in the usual fashion (via multiple pulse conditioning trains to
Schaffer collateral fibers) (Frey et a1.1993, Figure 3). A standard biochemical
tetanization procedure further revealed that cAMP levels were elevated
significantly one minute later in slices that received the three-pulse
stimulation that induces L-LTP but not in slices that receive a single
stimulating pulse, which only induces E-LTP. Elevated levels were not found
in slices receiving the three-pulse stimulation along with SCH 23390, a D 1
receptor antagonist, or with DL-2-amino-5-phosphonovaleric acid (DL-APV),
an NMDA receptor antagonist. Ten minutes later, cAMP levels in slices
receiving the three-pulse stimulating current and no inhibitory treatment had
returned to normal.
Based on these results, Frey et al. (1993, 1663) conclude that L-LTP
requires protein synthesis,

seems to be maintained by the generation of cAMP and activation of
PKA , and
begins with PKA activity and new protein synthesis that starts within
the first hour of LTP induction, while E-LTP is still in progress.
71
Viewed in retrospect, these studies were crucial steps toward one current
model of the molecular mechanisms of L-LTP.
Infusing experimental techniques from molecular biology into
mainstream neuroscience has since produced evidence of the specific gene
expression and protein synthesis involved in L-LTP induction and maintenance. Catalytic PKA subunits, freed from their regulatory subunits by the
rise in cAMP molecules, translocate to the postsynaptic cell nucleus. There
they have two principal targets:
they phosphorylate a class of cAMP response element binding proteins (the CREB -l class), which in tum bind to cAMP response
elements (CRE subregions) on regulatory regions of a variety of
immediate early genes;
they phosphorylate mitogen-activated protein kinase (MAP K), which
in turn binds to a regulatory subunit on a second class of CREB
proteins (CREB-2).
The overall effect is a changing balance in the molecular triggers that

activate or inhibit new gene transcription and hence new protein synthesis.
This molecular balance is sensitive to the intracellular biochemical cascades
originating at synapses where LTP gets induced.
To understand these mechanisms we must make a brief foray (one
paragraph and illustration!) into contemporary molecular genetics . Unfortunately, the rudiments of molecular genetics have been as ignored by
philosophers of mind and many cognitive scientists as has been molecular
neuroscience. (This dual neglect is not surprising since molecular
neuroscience and molecular genetics fit together, hand in glove.) A gene can
be divided into two functional regions: the coding region, whose base pair
sequences code for DNA transcription into messenger RNAs (mRNA); and a
control region, (typically) upstream to the coding region, whose base pair
sequences in conjunction with regulatory ("response element binding")
proteins control the initiation of the gene's transcription (Figure 2.7 A).
Transcribed mRNA molecules in turn trans locate out of the cell nucleus and
onto ribosomes in the cell cytoplasm for translation into new proteins. A
gene 's control region is further divided into regulatory and promoter regions
(Figure 2.7A). An RNA polymerase enzyme binds to promoter proteins and
the DNA in the gene 's promoter region . Base pair sequences in the gene's
regulatory region form response elements, to which regulatory protein
molecules bind . Regulatory proteins come in two functional varieties:
response activators and repressors . When response activators bind to
response elements, this literally changes the shape of the DNA molecule,
causing it to fold over and loop so that the regulatory proteins bound to it also
72
contact and bind to the RNA polymerase (Figure 2.7B). This begins the
process, or "turns on," transcription of mRNA. When either the appropriate
activators are not bound to the ir response element or response repressors are
bound there , the necessary alteration to DNA shape does not occur and gene
transcription does not begin. Paired activators and repre ssors often compete
for response element binding sites .i"
I
s
RNA-Polymerase II
"i=+iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii_~ S'
---
3'
Response
element
binding site
Promotor region
Response
element binding
prolein (activator)
S'
COding region
B
/
RNA,POlymCraSC II
i-'__
3'.,.-u I~====B3'
S'
Response
element
binding site
Promotor region
COding region
Figure 2.7. Basic constituents of a gene. A. No mRNA transcription initiated . B. mRNA

transcription initiated by response clement binding proteins bound to response clement sites on
the gene's regulatory region. Sec text for discussion. Figure created by Marica Bernstein.
We return to the molecular mechanisms of L-LTP. Phosphorylated cAMP

response element binding protein (isoform) la (P-CREB-la) is a response
element binding activator for a variety of immediate early genes . When
bound to a cAMP response element (CRE) site on these genes ' control
regions, it initiates gene transcription?' In L-LTP (in CAl neurons), two types
of immediate early genes are affected. Immediate early gene activation is a
cell's earliest genetic response to a particular inducing stimulus. The protein
73
products of immediate early genes typically act as enhancer binding proteins

for gene expression further downstream, acting "immediately" to activate or
repress transcription of other genes (in the same manner illustrated in Figure
2.7 above). In LTP induction, the inducing stimulus is the multiple-pulse
conditioning train in afferent (input) fibers . Immediate early genes and their
response profiles are common to cells of all tissue types, but are especially
prominent (and well adap ted) in neurons to support the latter's function as
networked signaling units"
One immediate early gene containing CRE sites to which P-CREB-la
binds transcribes mRNA to synthesize the protein, ubiquitin carboxyl-terminal
hydrolase (Figure 2.8). This protein is part of an intracellular ubiquitin
proteasome complex that destroy s free regulatory subunits of PKA molecules.
Without the activity of this complex, PKA regulatory subunits quickly bind
back onto and re-inhibit the PKA catalytic subunits, halting the latter 's
synapse-potentiating effects (in both E- and L-LTP). But when these
regulatory subunits are destroyed by the ubiquit in proteasome complex, intracellular PKA remains in a persistently active state, and so continues to
translocate to the cell's nucleus and produce postsynaptic potentiation (Chain
et al. 1999). Expression of the immediate early gene ucli that transcribes
ubiquitin hydrolase is thus driven by the very molecular proce ss it in turn
maintains.
CCAAT enhance r
binding protein
Ubiquitin
hydrolase
Y:l Proteins for structural
changes at synapses
Figure 2.8. Specific gene target s of phosph orylated CR EB. ucli transcribes mRNA for ubiquit in
hydrolase, a key constituent in a proteasome that de stroys free PKA regulatory subunits,
keepin g the catalytic PKA subunits persistent ly active. CIEBP tran scribes CCA AT enh ancer
binding prote in, itself an act ivator for late-respon se genes tran scribin g mR NA for proteins that
perman entl y pote ntiate posts ynapti c structure and function . See text for detail s. Figure created
by Mar ica Bern stein.
74
A second immediate early gene activated by P-CREB-la codes for

production of another transcription factor protein, CCAAT enhancer binding
protein (C/EBP) (Figure 2.8). This protein is a response activator for a class of
late response genes that code for proteins inducing growth of new postsynaptic dendritic spines, and hence new synapse formation (Taubenfield et al.
2001). The final result of these molecular mechanisms is a lasting increase in
the number of synaptic sites between the presynaptic fiber and the postsynaptic neuron, a lasting structural change underwritten by new gene expression
and protein synthesis. The increased number of synapses raises the probability
of summated postsynaptic activity induced by presynaptic transmitter release.
This heightened probability is, of course, L-LTP. Only now, the gene-driven
structural changes persist for days or even longer, not just for a few hours.
These activational effects of P-CREB -la are supplemented by
phosphorylated mitogen activated protein kinase (MAP K), a second effect of
translocated catalytic PKA subunits into the cell nucleus (Cammarota et al.
2000). A second type of CREB protein, CREB-2, is a repressor of the
immediate early genes for which P-CREB-la is an activator. The two
molecules compete for CRE binding sites on the genes ' control regions.
CREB-2 lacks a binding site for catalytic PKA. But it has a binding site for PMAP K, which in turn phosphorylates CREB-2 to inhibit its repressor
activity. Shutting down CREB-2 further enhances the activation effects of PCREB-la. The interplay of this activator and repressor determines the ease or
difficulty with which E-LTP "consolidates" into L-LTP.
I end this section with one final scientific question. How can this
account of L-LTP's molecular mechanisms explain the synapse specificity of
LTP? If L-LTP ultimately is governed by events occurring in the postsynaptic
neuron's nucleus, how do the protein products "know" which synapses to
translocate to and potentiate, and which to avoid? Given that potentiation is
specific to active synapses, the idea of a stimulation-induced "mark" or "tag"
seems theoretically promising. Such a tag could enable active synapses to
capture and incorporate functionally the products of gene expression that have
been transported nonspecifically throughout the cell. Recent empirical studies
on invertebrate neurons have revealed components of gene translational
machinery, including mRNA and ribosomes, in distal dendrites. The
intriguing idea that active synapses synthesize "locally" their own protein
"tags" that interact with and "capture" gene expression protein products being
transported down from the soma has some recent experimental support
(Martin et al. 1997), and is currently a topic of intense investigation .
The conditioning pulse trains that generate L-LTP eventuate in
postsynaptic neurons whose gene expression, protein production , synaptic
structure and biochemistry-and so their capacity to generate action potentials
in response to afferent input-are lastingly altered. These same events are
75
occurring at specific synapses throughout the neuronal circuitry leading from

sensory receptors to motor effectors, namely, at the ones selectively activated
by the afferent stimuli or endogenous neural activity. In the behaving animal ,
the se molecular mechanisms eventuate in permanently changed motor
responses (behavior) to specific sensory inputs or endogenous activity. Our
next task is to see how recent neuroscience has built an explanation of
behavioral data in memory consolidation tasks out of this developed account
of the molecular mechanisms of E- and L-LTP.
5 BUT IS THIS REALLY MEMORY (CONSOLIDATION)?

I've imposed upon readers a fair amount of cellular and molecular
detail and more is coming. So now is a good spot to remind you about where
this discussion is headed and why the se details are important. Given the
induction similarities and temporal correlations between short-term memory,
long-term memory, E-LTP, and L-LTP, a tempting hypothesis is that
psychology's "consolidation switch" reduces to a three-part sequence
described within contemporary cellular and molecular neuroscience. Thi s
sequence starts with activity-induced enhancement of freed catalytic PKA
subunits at active synapses in specific neurons, leading to
1. increased binding of P-CREB-la to CRE regulatory subregions on

one immed iate early gene that transcribes a protein for maintaining
persistently active PKA, and another that transcribes a response
element binding activator that turn s on expression in late-response
gene s coding for proteins that generate the growth of new synaptic
sites;
2. inhibition of CREB -2, a transcription repressor for the same
immediate early genes ; and
3. the translation of protein products transcribed by these immediate
early and late-re sponse gene s.
What psychologists call "retrograde interference" turns out to be any process
initiated after initial stimuli that disrupts any of these cellular/molecular steps .
In Duncan's (1949) study (this chapter, section 2 above), electro-convulsive
shocks delivered fifteen minute s or less after the unconditioned stimulu s were
very disruptive of these steps, while retrograde hind leg shocks were not.
Tempting as this hypothesis is, the key question remains outstanding:
Can this potential reduction be verified experimentally? The most straightforward way to do this would be to manipulate these molecular mech anisms of
L-LTP directly to produce observable effects that spare short-t erm memory
76
but compromise long-term memory behavior. This result would be especially

compelling if it involved a form of long-term memory specific to "higher"
mammals, for which it is widely agreed that this form is "genuinely
cognitive." This procedure has become central to the explanatory claims of
these molecular mechanisms in recent mainstream neuroscience, and hence to
the proposed reduction of mind to molecules. We now shift to the additional
scientific background necessary to understand the experimental details that
have recently yielded exactly these results .
5.1 Declarative memory

Declarative memory is just such a form. It admits of the short-term
memory/long-term memory distinction and is specific to mammals. This term
is best known from the work of neuropsychologist Larry Squire and his
colleagues. Its origins lie in different patterns of learning and memory
preserved versus lost in human global amnesia. The principal neuropsychological symptom of global amnesia is impairment to remembering new
information, including acquisition, retention, and/or retrieval. Most human
global amnesia occurs as part of Korskoff s syndrome, due to long-term
alcohol abuse . But there are also selective cases in which the deficit results
from bilateral damage to the medial temporal lobes, a part of the limbic
system housing the hippocampus. Celebrated case H.M . is one of the latter.
H.M. underwent surgical bilateral ablation of the medial temporal lobe in
1957 to treat otherwise intractable epilepsy. The surgery produced a profound
memory impairment against a background of retained cognitive and intellectual capacities (Scoville and Milner 1957). H.M.' s anterograde amnesia for
events that have happened since the surgery is virtually complete, and his
retrograde amnesia stretches back for a few years prior to the surgery.
Interestingly, his memory for events in his distant past, well before the
surgery, remains intact, comparable to nonamnesics controlled for age. Up to
the present day, he forgets events soon after they end . For example, he is
unable to remember what he ate at his previous meal, he can't remember
having met individuals before that he sees regularly (e.g., doctors, nurses),
and he can't remember being told before that his parents have died.
Nevertheless, soon after his surgery, neuropsychologist Brenda Milner
showed that H.M . could acquire and perform a motor skill at a normal rate
(compared with nonamnesics), despite his inability to recall prior episodes of
training. The skill was pursuit-rotor, where subjects guide a mechanical
device to keep it on target with a dot rotating on a turntable at variable speeds.
Following .this initial demonstration of retained skill acquisition capacity,
77
neuropsychologists began to explore the extent of retained learning and

memory capacities in global amnesics .v'
Squire's distinction between declarative and nondeclarative memory
systems emerged directly from this research. Figure 2.9 illustrates his current
manner of carving up memory systems . The declarative system is compromised in global amnesics, particularly in medial temporal lobe amnesics.
Squire and his collaborators initially chose the term , "declarative," to suggest
the ability to declare one 's knowledge verbally. Thi s memory type or system
has often been characterized in terms of conscious recognition and recall. In a
fairly recent publication, for example, Squire introduces the term this way:
One kind of memory provides the basis for conscious
recollection of facts and events. This is the kind of memory
that is usually meant when the terms ' memory' and
'remembering' are used in ordinary language. Fact-and-event
memory refers to memory for words, scenes , faces , and
stories , and is assessed by conventional tests of recall and
recognition. This kind of memory was termed "declarative"
to signify that it can be brought to mind and that its contents
can be "declared." (1992, 232)
Clearly, this characterization picks out a type of memory that qualifies as
"genuinely cogniti ve." The problem is that under this description it is difficult
to apply to memory research using experimental animals. Connecting up the
human neuropsychological literature with experimental animal work requires
a broader characterization of declarative memory .
Memory
->.
Nondeclarative (Implicit)
Declarative (Explicit)
Facts
Events
Skills and
Habits
Simple
Classical
Conditioning
Priming
Nonassociative
Learning
Fig ure 2.9 . Squir e and colleague s' current division o f memory systems and subsystems. (See
Squire 1992. 233.)
78
The key to achieving this, according to Squire, is to adopt a "brain

systems framework" (1992, 240). No matter which of numerous popular
terms one favors to denote the memory type or system compromised in
human global amnesia-'declarative,' 'explicit,' 'relational,' 'configural'they all converge on tasks that require an intact hippocampus and surrounding
tissue in the medial temporal lobes . Squire writes:
If one considers the various biological and purely
psychological concepts that have been used, it is striking that
they sort themselves out in terms of ideas about what the
hippocampus does and does not do in the service of memory .
... The important point is that the terms explicit memory and
declarative memory, when one considers the properties that
have been associated with each, describe a biologically real
component of memory that depends on particular structures
and connections in the brain. (1992b, 205)
This approach forges a connection between human neuropsychological data
and experimental mammalian research . The "particular structures and connections," namely the hippocampus proper, entorhinal cortex, perirhinal cortex,
and perihippocampal gyrus, have homologs across the mammalian class.
Since declarative (or explicit) memory is coextensive with hippocampalrequiring memory , the term is applicable to memory research on humans,
other primates, and rodents (Squire 1992b).
Another result that holds across the mammalian class is that a second
prominent structure in the medial temporal lobe, the amygdala, is not
associated with declarative memory . Although H.M.'s surgical ablation
included his amygdala bilaterally, humans with more circumscribed lesions
that do not include it still display H.M.'s pattern of preserved and impaired
memory capacities. Controlled studies with surgically lesioned monkeys indicate that hippocampal lesions (and surrounding tissue) produce significant
deficits in delayed nonmatching to sample experiments, while amygdala
lesions alone do not; and amygdala lesions conjoined with hippocampal
lesions do not exacerbate this behaviorally measurable deficit (Squire and
Zola Morgan 1991). In delayed nonmatching to sample, subjects are presented
with an object and then, following a delay period, are presented with the
original object and a novel object. They must choose the novel object (the
"nonmatch to sample") to receive the task reward . Human global amnesics are
impaired on this task. Results using it with surgically lesioned monkeys were
crucial in exploring the extent of medial temporal lesions that interrupt
declarative memory and establishing the currently accepted primate model of
human global amnesia (Squire and Zola Morgan 1991).
79
Two procedures have been prominent in the rodent experimental

literature on declarative/relational/hippocampal-requiring memory. One is the
multiple-armed radial maze. This consists of a raised platform with a
compartment in the center and hollow arms radiating out (like spokes of a
wheel). Arms are baited with a small food reward at their ends away from the
center compartment. Rats can be trained to search efficiently for food
rewards; after around twenty trials, they never return during an individual trial
to a compartment they visited earlier in that trial. They can even' do so when
prevented from following a fixed sequence of arm visits (e.g ., when they are
blocked from visiting next the arm immediately to the left). Rats with lesions
to hippocampus, fornix , or entorhinal cortex, however, cannot learn to visit
the arms efficiently. Whether or not they re-enter an arm later in a trial that
they visited earlier in that trial falls to chance. On each trial they eventually
find all the food , but only after entering many of the arms multiple times
(Olton 1983). This is not due to an inability of the lesioned rats to distinguish
among the arms. They can learn whether a given arm contains food across
trials or never does. Olton and Papas (1979) demonstrated this result elegantly
using a seventeen-arm radial maze. Before each trial, eight arms were baited;
the other nine arms were never baited on any trial. Trained rats with fornix
lesions visited the baited arms randomly and inefficiently, re-entering arms
they had previously entered earlier in the trail. But they also learn at a rate
comparable to controls to avoid the arms that are never baited. Apparently
they cannot remember where they have visited recently, but they can
remember which locations regularly contain food and which don't.
A second common experimental procedure is the Morris water maze.
Rats and mice are hydrophobic; they don 't like being in water. The Morris
water maze takes advantage of this motivational fact. It is a pool filled with
opaque liquid (originally water mixed with powdered milk) with a single
submerged platform high enough for the rat or mouse to be perched mostly
out of the water. Visual stimuli are painted on the walls of the maze (or the
room containing it) so that the rodent can navigate a path to the platform. Rats
are put in the pool initially at a random location and swim until they
encounter the platform. They are then put back into the maze at a different
location. Rats quickly learn to navigate to the submerged platform and after a
few trials will swim directly to it, no matter where they are placed into the
pool. In his original study, Morris (1982) found that rats with neocortical
lesions performed similarly to normal controls, but rats with hippocampal
lesions swim randomly on each trial until by chance they come upon the
platform. This defici ency in memory for spatial locations is consonant with a
similar deficiency in human amnesics. It falls clearly within the declarative/
relational/hippocampus-requiring category . However, hippocampal-Iesioned
rats and mice can learn and remember other tasks in the water maze-
80
nonrelational, stimulus-response tasks. For example, if the platform is

elevated to jus t above water level , making it visually accessible, hippocampallesioned rats quickly learn to swim directly to it. Even with the submerged
platform, if hippocampal-Iesioned rats are always released at the same place
in the pool, they learn to swim in the same direction every time.
However, the popularity of radial arm and water maze tasks
sidetracked animal researchers into thinking that the rodent hippocampus was
involved primarily in spatial memory. Recently , Howard Eichenbaum and his
colleagues have showed that hippocampal-Iesioned rats also fail on a
nonspatial relational memory task. Bunsey and Eichenbaum (1996) trained
rats to dig for food rewards buried in jars under scented sand. The odors were
originally presented as paired associates . For example, on Training Set 1, the
rats learn that if odor A is presented as the sample, they are to dig in the jar
under the sand scented with odor B rather than odor Y. If odor X is the
sample, then they are to dig in the jar under the sand scented with odor Y
rather than odor B. After mastering that training set, the rats are presented
with Training Set 2. Here they learn to pair sample odor B (from Set 1) with
the choice of odor C instead of odor Z, and to pair odor Y (also from Set 1)
with the choice of odor Z instead of odor C. (See Figure 2.10) Hippocampal
lesioned rats acquire all paired associate combinations at a rate statistically
similar to control sham rats (who undergo a similar surgical procedure leading
up to the hippocampal lesions, but don 't receive the lesions). Next Bunsey
and Eichenbaum (1996) subjected the trained rats (hippoc ampal lesioned and
sham controls) to a Test for Transitivity. Either odor A or odor X (both from
Set 1) is presented as sample, followed by a choice between odors C and Z
(both from Set 2). If A is the sample odor, the food reward is under C; if X is
the sample, the reward is under Z (Figure 2.10). Sham rats performed far
above chance in this test requiring a transitive operation over the paired
associate odor memories; hippocampal lesioned rats, who learned the paired
associates with equal ease, performed slightly below chance. Bunsey and
Eichenbaum (1996) also tested the paired associate-trained rats on a Test for
Symmetry. Odors are the same as in Training Set 2 of the transitivity study,
but now samples are either odor C or Z and the choices are odors Band Y. If
C is the sample , the food reward is buried under odor B; If Z is the sample, it
is buried under Y (Figure 2.10). Here again, sham rats performed far above
chance, whereas hippocampal lesioned rats were virtually at chance. These
results are also consistent with reports of stimulus -stimulu s association
learning deficits in hippocampal lesioned monkeys (and surrounding tissu e)
and amnesic humans .
81
TRAINING SET 1: AB & XY
Sample
Choice
B VS Y
TEST FOR TRANSITIVITY: AC & XZ
SAMPLE
B VS Y
TRAINING SET 2: BC & YZ
Sample
Choice
C VS Z
CHOICE
C VS Z
C VS Z
TEST FOR SYMMETRY: CB & ZY
C VS Z
Sample
Choice
B VS Y
B VS Y
Figure 2.10 . Odor paired associates in Bunsey and Eichenbaum's transitivity and symmetry
memory tasks. (See Bunsey and Eichenbaum 1996,256.)
Clearly, the hippocampus mediates more than just spatial learning

even in rodents. Bunsey and Eichenbaum (1996) refer to this declarative/explicitlrelational/hippocampal-requiring memory feature as flexibility,
referring to the animal 's ability to use such memories to solve novel problems
other than the ones on which they were acquired. This is in contrast to the
inflexibility of nondeclarative memories, which are tied directly to the stimuli
or training regimen generating them. Cohen and Eichenbaum's (1993)
picturesque phrase, the "promiscuity" of declarative memory, is also intended
to display its accessibility by numerous and novel expressive routes . This
feature locates declarative (lexplicitl...) memory squarely within the realm of
the "genuinely cognitive." And once again, in keeping with Squire's "brain
systems framework" for memory, the common link underlying flexible,
promiscuous memories across mammalian species is their dependence on the
intact hippocampus (and surrounding tissue) .
5.2 Biotechnology solves a long-standing methodological

problem in LTP-memory research"
Now we can make more specific the question raised at the beginning of this
section . Can a direct experimental link be established between the
82
molecular mechanisms of E-LTP and L-LTP and the consolidation of

declarative short-term memories into long-term memories? The search for
experimental links between LTP and memory has been both exciting and
frustrating-and continually subject to methodological criticisms. Many
critics have focused on the pharmacological agents and procedures that up to
a few years ago were the best available laboratory techniques (see, e.g.,
McEachern and Shaw 1996). Two limitations on pharmacological manipulations are obvious. First, pharmacological agents are rarely completely
specific for a single type of protein target (receptor or enzyme). So even
though elaborate delivery techniques can restrict application of an agent to an
extremely local area, it can still be difficult to control for effects on molecules
and proteins that the experimenter is not targeting. Second, and specific to the
emerging story of the molecular mechanisms of LTP, as of the mid-1990s
pharmacological manipulations had not yet revealed the relation between LLTP and long-term memory in behaving animals. It was then that some
groups turned to targeted gene manipulations that, owing to advances in
biotechnology, can deliver maximally specific enhancement or disruption of
particular molecules in intracellular pathways. A particular gene directs the
production of a specific protein that helps to determine the structure or
function of each cell in which it is expressed. By the early 1990s, techniques
existed to manipulate any gene in the mouse genome.
Two of these techniques have dominated recent neuroscientific
research. One generates "knock-out" organisms; the other generates
"transgenics." Knock-outs show "what happens when we create deliberate
mutations in a particular gene in a living organism," while "transgenics illuminate the role of a protein product's function in intracellular processes"
(Weaver 1999, 127). Lodish et al. (2000) make an even grander claim about
the importance of these techniques, opening their chapter describing them by
asserting: "The goal of modern molecular cell biology is nothing short of
understanding the biochemical, cellular and organismal functions of all the
proteins encoded in the genome" (281).
Interestingly, they end that
discussion by describing "one important example . . . [that] comes from
studies on learning and memory" (2000, 286).
Ultimately, both knock-out and transgene techniques depend upon
inserting foreign DNA into an organism's genome. Both techniques start with
the same procedures. Since these technologies have been applied to many
different research questions, molecular cell biologists have developed many
technical variations. In this subsection we will describe the processes of
making knock-out and transgenic mice in only general terms. But we will
include enough detail to foster an appreciation for just how laborious (and
expensive!) the technologies are, and for how rigorous the bench science must
be.
REDUCTION-IN-PRACTlCE
83
If one intends to understand a gene's or protein product's biological

function, then not just any nucleotide sequence can be inserted into the
genome. The gene 's specific nucleotide sequence must first be known, the
sequence of adenine, thymine, cytosine, and guanine (the As, Ts, Cs, and Gs)
that carry the DNA's information. Thus the first step in both knock-out and
transgenic technologies is to identify and sequence the gene of interest (or its
protein product). There are now several technologies that facilitate this,
especially the creation of searchable "libraries" of organisms ' genomes.
Theoretically, this process is straightforward. One suspects that an expressed
protein is important in a particular biochemical pathway or process occurring
in a particular tissue at a particular time. One isolates the immediate product
of DNA transcription, messenger RNA (mRNA), from those tissues." The
initial product will be mRNA molecules corresponding to all the genes being
actively transcribed in those cells at that time. In order to sequence the gene of
interest, its mRNA must be separated from the rest. Relatively large
quantities of the initial product are needed for this separation (and for
subsequent steps). Additionally it is DNA, not RNA, that will be inserted back
into the genome, so the gene-specific mRNA sequence must be re-converted
into a DNA sequence. Increasing the number of molecules (amplification)
and converting mRNA into DNA is accomplished using the workhorse of
molecular cell biology, the polymerase chain reaction (PCR). PCR is an in
vitro biochemical reaction that exponentially increases the number of RNA or
DNA molecules. Because the enzymes that make RNA copies of DNA in
cells are incredibly faithful , they can be used in a special kind of PCR, reverse
transcriptase-PCR (RT -PCR) . This process both "back-tracks" mRNA into
DNA and generates thousands of copies of cDNA ("complementary DNA," a
DNA copy of an RNA molecule). From those cDNAs, the gene of interest is
isolated and sequenced using any of a number of standard biochemical
techniques. Most techniques filter the cDNA through an agarous gel that
separates molecules according to their molecular weights. Individual "bands"
on the gel, formed by deposits of many cDNA fragments of a given molecular
weight, can be physically cut out of the gel and re-amplified. But appreciate
that in order to isolate the right band, some features of the gene must already
be known.
Note that the preceding chain of events need not occur in precisely
this order, and that details depend on the specific objectives of the research.
For our purposes, the important point is that the product will now be purified
nucleotide sequences corresponding to the gene of interest. This cDNA can
now be altered-mutated-in a variety of ways that render the cDNA or its
protein product non-functional when inserted into a cell's genome. This is the
essence of a "knock-out" gene. An endogenous functional gene is replaced by
another that has been mutated in a specific way. However, as we saw in the
84
previous section, not all regions of a gene wind up translated into a protein
product. "Control" regions regulate transcription timing and rate (see Figure
2.7 above). If those regions are selectively mutated, then when the
"transgene" is inserted into the genome, the resulting protein is functional but
expressed in abnormal quantities.
Now the fun begins! The ultimate goal is to insert a DNA sequence
into double stranded DNA in the nucleus of a mammalian cell. Fortunately
for molecular biologists, nature has been inserting foreign DNA into native
DNA for millennia, outside the laboratory. Viruses, phages, plasmids and
even modified yeast chromosomes all have the capacity, when introduced into
a host cell, to insert their own small genomes (or portions of them) into those
of their hosts. Molecular biologists make use of plasmids as "vectors" to
transfer and insert foreign DNA into mouse genomes. Plasmids are small
circular pieces of bacterial DNA that replicate independently of the bacterial
cell's chromosome. They also contain "restriction enzyme recognition sites"
within their own nucleotide sequences. Restriction enzymes recognize
relatively short but very specific DNA nucleotide sequences and break the
phosphodiester backbone of DNA between them. They literally cut up DNA.
There are many restriction enzymes found naturally in bacterial cells . There
they function to remove foreign , potentially damaging DNA from bacterial
genomes. By selecting the appropriate plasmid and restriction enzymes,
molecular biologists cut the plasmid DNA and insert the mutated gene of
interest directly into it. Other enzymes seal the break and we now have a
circular piece of DNA, one portion containing the altered gene of interest. By
asexual reproduction, these plasmids can replicate virtually indefinitely,
generating many "clones" of the original gene .
Two additional details require mention. First, some biochemical
tinkering with the gene is necessary so that the ends of its sequence are
complimentary to the ends of the broken plasmid DNA. At this stage, the gene
of interest is a single stranded DNA molecule. Single strands form double
strands in virtue of complimentary base pairing: A-T, C-G . The sequence of
one end of the gene must be engineered such that it is complimentary to one
of the plasmid ends that results when the circular plasmid is cut by the
restriction enzyme. In principle this is not difficult, since we know the restriction enzyme recognition site. Second, often the unaltered gene is inserted and
the mutation is induced after the gene is in the plasmid. This method is useful
when the mutation involves removing longer nucleotide sequences that might
correspond to functional domains of the protein.
Amplifying, identifying, sequencing, and cloning are virtually identical for both knock-out and transgenic techniques, although subtle differences
will depend on specific research objectives. At this point, two questions
become critical:
REDU CTION-IN- PRACTICE
8S
1. Did the mutated gene of interest insert into the mouse' s genome?
2. If so, where in the genome is the insertion located?
Knock-out and transgenic techniques differ in their answers to question (2).

When creating the plasmid vector carrying the altered gene of interest, steps
are taken to answer these questions. We will first focu s on knock-out mice
and return later to the construction of transgenics.
Remember that in a "knock-out" the functional wild-type gene is
replaced by a deliberate mutation . Cultured mammalian cells are naturally
permeable to DNA and there are several ways to increase this permeability.
Thus the plasmid vector can be inserted into embryonic stem (ES) cells from
brown mice. (As we will see below, coat color is useful to determine whether
the mutated gene has been inserted properly.) Under the right conditions,
some of those plasmids will translocate into the ES cell nucleus . Because ES
cells are actively dividing, the enzyme machinery that promotes incorporation
of the gene of interest into the genome is present and active. Gene insertion
takes place via recombination, the physical exch ange of DNA between two
chromosomes, which is occurring naturally in reproducing cells. "Homologous recombination" is the actual "crossing over " of the two sister
chromosomes during meiosis I, the stage of the cell cycle during which
homologous chromosomes are closely aligned. During crossing over, a gene
from the maternal chromosome is exchanged for the same gene on the
paternal chromosome. If that recombination occurs in germ line cell s, the
resulting gametes will have genomes fundamentally different from both the
parental genomes. That recombination occurs in sexually reproducing organisms explains in large part the huge amount of variation in populations; glibly ,
it explains why siblings can have such different combinations of their parents'
features. The event that leads to insertion of foreign DNA is a recombination
event.
Three things can happen to the foreign DNA in plasmid vectors
entering the ES cell nucleus :
1. Nothing. No recombination takes place and no DNA is inserted.
2. Random recombination takes place in all or many chromosomes. The
gene of interest is inserted, but only along with portions of the
plasmid sequence and not at its proper location.
Or
3. homologous recombination occurs. The mutated gene of interest
cros ses over with a chromosome where the wild type gene resides.
The wild-type is replaced by the gene of interest.
But how can we tell which of these three possibilities occurred?
86
We said earlier that steps are taken when constructing the plasmid
vector to answer this question. Typically, two other genes are inserted into the
vector along with the gene of interest. The first gene permits ES cells
containing no insertions to be separated from those containing some
insertions . The classic technique is to insert a gene that confers resistance to
an antibiotic, usually neomysin. (In fact, disrupting the functional gene (i.e.,
mutating the gene of interest) is often accomplished by interrupting the gene 's
sequence with the antibody resistance gene.) If all cell s are then grown on a
medium containing the antibiotic, those with no resistance (no insertion) will
die . The second gene permits separation of those ES cells undergoing nonspecific recombination from those undergoing homologous recombination.
For example, the gene for thymadine kinase, tk, can be inserted into the vector
outside the gene of interest. If non-specific recombination occurs , tk will be
inserted in addition to the gene of interest. (Remember, if homologous
recombination occurs , only the gene of interest is inserted.) By treating the
remaining cells with a drug that is lethal to cells containing tk, we eliminate
all but those in which the mutated gene of interest has replaced its wild-type
homologue.
Finally! We now have brown mouse ES cells that contain the
disrupted gene. However, they are heterozygotes. The gene on one chromosome is disrupted, but the other remains wild-type. The next step toward
generating mice that are homologous for the disrupted gene is to inject these
ES cells into bla ck mice embryos. The embryos are transplanted into black
surrogate mothers. The resulting offspring are referred to as "chimera."
Because the embryos contain cells from both a brown and a black mouse, the
offspring have a brown and black coat. Assuming that some of the original ES
cells became germ-line cells , basic Mendelian genetics dictates the remaining
steps. The mature male chimera is mated to a black female . Brown is the
dominant coat color , and some of the male's gametes will have derived from
original black cells, some from brown ES cells (with the disrupted gene) .
Black offspring-double recessives-obviously do not have the gene of
intere st. But only some of the brown offspring do. (Remember that the
original ES cells were heterozygous for the gene of interest.) Simple genetic
analy sis reveals which of the brown mice carry the gene of interest, but even
these mice will be heterozygous, as one copy of the wild-type came from the
black mother to which the chimera was mated. A final round of crossing
heterozygotes will generate a proport ion of offspring who are true "knockouts ." These animals have two copies of the disrupted gene of interest.
Researchers constructing the first knock-outs confronted an
immediate and difficult challenge. Consider the general research objective: to
understand a gene product's function in a particular tissue at a particular time.
But the first knock-outs were in essence genetic lesions, and familiar
87
methodological objections to lesion studies applied to them. A single gene

(and its protein product) can function in many pathways or processes in
different tissues at different times over the course of an animal 's development. To conclude that the disrupted gene causes some malady that a knockout suffers can fail to acknowledge the gene's "downstream" effects.
Furthermore, disrupting some genes during development is lethal. Genetic
engineers solved these problems by developing "conditional" knock-outs.
Using more sophisticated versions of the plasmid vector techniques, molecular biologists now construct vectors inserting mutated genes that can be
turned on or off in only specific tissues and/or at precise times. Typically the
trigger is a pharmacological agent, but in the next chapter we'll see another
mechanism employed (a promoter turned on by heat shock) .
We turn next to transgenics. The fundamental difference between
knock-outs and transgenics is the specificity of the gene insertion. In knockouts, a wild-type gene is replaced by a mutated, nonfunctional version through
homologous recombination. This specific replacement occurs at a very low
frequency. The frequency of random DNA insertions is much higher. In fact,
many copies of a transgene can be inserted into a single cell's genome.
Organisms whose cells contain these many insertions-up to 100 copies per
cell-are referred to as "transgenics." Here the gene of interest contained
within the plasmid vector is injected into the two nuclei of a fertilized egg
(one from each parent) before they fuse . Because development is at the singlecell stage and the transgene insertion is non-specific, there is no need to select
for specific insertion (as there was with knock-outs). Once fused, these
fertilized eggs are injected into a pseudopregnant female (which has received
hormone treatments to make her receptive to the eggs) . The eggs divide and
differentiate normally. 10-30% of the offspring will contain the inserted
foreign DNA (the transgene) in all cells of all tissues, including germ tissues .
Each cell will contain equal numbers of the gene insertion. Non-specific
insertions can disrupt normal development and function, but typically 10-20%
of these mice will breed normally. Using breeding schemes similar to knockout technology, mice can be crossed and backcrossed to produce pure transgenic lines. In transgenics, the protein product of the inserted gene is
overexpressed compared to wild-type controls. This expression is in addition
to whatever products are expressed by other genes on the chromosomes on
which the trangene was inserted.
One cannot overemphasize the impact of biotechnology across all of
contemporary biology. The last quarter-century has revolutionized the discipline. This impact has resonated through neuroscience. As Karl Herrup has
remarked, "rapid and complementary advances in the fields of molecular
biology and experimental embryology have combined to offer neuroscience
researchers unprecedented power to manipulate the mammalian genome"
88
(Zigmond et al. 1999, 42l)-and with it, the protein products that drive
neuronal activity and plastic ity. This subsection is only intended to give a
sense of the techniques that have come to dominate mainstream neuroscientific research. In the next subsection, we'll see in specific detail how
these techniques have been employed to engineer animals that tie the
molecular mechanisms of L-LTP sketched above directly and experimentally
to memory consolidation in behaving animals.
5.3 An experimental link between molecules and behavior:

PKA, CREB, and declarative long-term memory
consolidation
Armed with biotechnology's ability to engineer knock-out and
transgenic mice, the long-sought experimental link between the purported
molecular mechanisms of L-LTP and declarative memory con solidation
appears to have been forged . First , Rusiko Bourtchouladze and Alcino Silva's
group at Cold Spring Harbor generated mice in which the gene expressing
CREB-l was partially knocked out (Bourtchouladze et al., 1994). Then Eric
Kandel's group at Columbia University generated transgenic mice partially
expressing a cloned gene transcribing a protein that keeps catalytic subunits of
PKA bound to regulatory subunits (Abel et al., 1997). Using an ingenious
behavioral paradigm that simultaneously trains mice on a hippocampalrequiring (hence declarative) and a non-hippocampal-requiring (hence nondeclarative) task, both groups demonstrated intact E-LTP in hippocampal
neurons and short-term memory on the declarative task but deficit L-LTP and
long-term memory in the engineered mutants . Given the details of these
studies, especially the careful controls both groups employed, it is difficult to
resist the conclusion that "the PKA pathway is critically important for the
consolidation of short-term memo ry into protein synthesis dependent longterm memo ry, perhaps becau se PKA induces the transcription of genes
encoding proteins required for long-lasting synaptic potentiation" (Abel et al.
1997.615; my emphases).
The CREB" mouse mutant results from standard homologous
recombination in embryonic stem (ES) cells (described in the previous
subsection). The targeting vector consisted of cloned mouse CREB genomic
DNA and a promot erles s neomycin-resistant (Neo) gene. The Neo gene is
inserted into an exon common to all known CREB mRNA isoforms. Chimeric
males are mated with wild-type female s to produce ES cell-derived offspring.
Heterozygotes for the gene target were mated to generate homozygou s CREB "
mutant mice. Standard molecular biological analyses revealed no CREB
89
expression in CREB - mutants. Nevertheless, mutants surviving to adulthood

displayed a normal phenotype and behavioral repertoire, no histological or
morphological defects, and no impairments in growth or development
(Hummler et al. 1994).26
Informing the background of Bourtcholadze et al.' s (1994) studies
with the CREB - mutants were
results with nonspecific protein and RNA inhibitors that effectively

blocked long-term memory while leaving short-term memory intact;
results with cultured neurons from the invertebrate Aplysia (a sea
slug) indicating that CREB antagonists bound to CRE sites selectively
blocked long-term synaptic facilitation while leaving short-term
facilitation intact (reviewed in the next chapter);
results with Drosophila (fruit fly) mutants indicating a role for each
step in the cAMP-PKA-CREB pathway in odor conditioning longterm memory (also reviewed in the next chapter).
Could CREB be implicated behaviorally in mammalian memory consolidation? CREB- mutant mice seemed a promising experimental preparation.
Bourtcholadze et al. (1994) report that their CREB- mutants show no
gross hippocampal anatomical defects or abnormalities when coronal brain
sections are matched with wild-type controls. They also report that the
mutants matched controls' performance on tests for foot shock sensitivity. In
one-trial-per-day training for fifteen days in the Morris water maze, CREB mutants were significantly impaired on time to find the submerged platform,
percent time in the correct quadrant during a probe task after the fifteenth
training day when the platform was removed (the quadrant of the pool where
the platform had been on every previous training session), and number of
times crossing the previous platform location during the probe task.."
However, most declarative memory tasks, including the Morris water maze,
take several days for rodents to learn . Their temporal dimensions make it
difficult to track and compare short-term memory with long-term memory,
and hence the mechanisms of memory consolidation, experimentally in
behaving animals. What is needed to test the specific effects of the CREB knockout on mammalian declarative memory consolidation is a hippocampaldependent task that mice can learn quickly and retain for a period that exceeds
short-term memory .
Bourtcholadze et al. (1994) came up with such a task. They subjected
CREB- mutant and wild -type control mice to a novel environment for two
minutes; this "context" was a conditioning box with metal and Plexiglas sides,
equipped with a removable electric floor grid. After the two minutes, a tone
was sounded through a wall speaker for thirty seconds; this conditioned
90
stimulus (CS) was at 85dB, well above the background noise in the chamber.
After the tone ended, a foot shock was delivered for two seconds through the
electric floor grid; this unconditioned stimulus (US) was .75 milliamperes, far
above the sensitivity threshold for mice. Hence simultaneously and on a
single trial, mice were learning an aversive contextual conditioning tasknovel environment paired with foot shock US-and a fear conditioning tasktone CS paired with foot shock US . It has long been established that
contextual conditioning in mammals requires an intact hippocampus, while
fear conditioning does not. In these studies, the behavioral measure is
percentage of time spent freezing for five minutes after being put back into
the conditioning chamber (contextual conditioning) or hearing the tone in a
different context (fear conditioning). Freezing is a stereotypical rodent fear
response in which the animal crouches and ceases all externally observable
movements (except breathing).
CREB- mutants and wild-type controls displayed statistically identical
freezing responses immediately after US presentation. This indicates that
immediate learning was not compromised by the lack of CREB in either
hippocampus or amygdala. CREB- knockouts were also statistically similar to
controls in freezing to both the context and the CS (tested separately) 30
minutes after initial training. Both groups froze on average around 40 % of the
five minutes to both the novel environment and the tone. This indicated that
both declarative and nondeclarative short-term memory remained intact in the
CREB - mutants. However, at both one hour and twenty-four hours after
training, the CREB - mutants averaged only about 10% freezing time to the
contextual cue (novel environment), while the wild-type controls still
averaged around 40 %. A similar result obtained for the fear conditioning
experiment. At two hours and twenty-four hours after training, CREBmutants showed equally less freezing than controls. The key to these results is
that the loss of CREB function disrupts long-term memory for both contextual
(declarative) and cued (nondeclarative) conditioning without affecting shortterm memory. This is directly in keeping with the model of L-LTP induction
sketched above and the hypothesis that these are the molecular mechanisms of
memory consolidation in behaving animals.
Bourtcholadze et at. (1994) strengthened this argument by also
studying the physiological responses of CREB - mutant hippocampal slices .
Slices were prepared, kept alive, and measured for baseline response activity
by standard methods sketched in the previous section. They induced LTP by
stimulating Schaffer collateral fibers with a train of 100 electric pulses at 100
Hz and recorded field EPSPs from CAl neurons. Wild-type control slices
showed L-LTP response two hours past the inducing pulse train , while CREBmutant slices had already decayed to baseline levels ninety minutes past LTP
induction . Although initial response (E-LTP) was similar in CREB- mutants
91
and wild-type controls, statistical differences in potentiated field EPSPs began

appearing only ten minutes after the inducing stimulus. This is during the time
that E-LTP processes are still occurring, directly in keeping with the current
model of L-LTP induction sketched in the previous section. These are
powerful experimental results, warranting Bourtcholadze et al.'s conclusion
that "the behavioral and electrophysiological findings reported in this
manuscript identify CREB as an important component of memory consol idation in mammals" (1994, 66; my emphasis).
Nevertheless, questions still remained. Bourtcholadze et al. (1994)
did find a fine-grained deficit in E-LTP induct ion in the CREB- knockouts
following a single stimulus train. Also, since CREB is a ubiquitous
transcriptional factor that is activated by second messenger pathways other
than cAMP-PKA, the above results don't define a conclusive role in memory
consolidation for the earlier steps in the L-LTP pathway. Bouercholadze et al.
note these limitations explicitly when they conclude that "although in the
CREB- mutants, both LTP and memory decay within the first hour , it is
unclear whether the decay in LTP is related to the decay in memory or
whether these two phenomena are unrelated consequences of the CREB mutation" (1994, 64-65). Interestingly, engineered mice with ES knockouts of
individual PKA subunit isoforms did not yield compelling data linking PKA
to either LTP or long-term memory (Qi et al. 1996). There also remained the
numerous methodological problems with research using knockouts beyond
developmental stages. Neither Hurnmler et al. (1994) nor Bourtcholadze et al.
(1994) found developmental abnormalities in the CREB- knockouts, though of
course this does not mean that subtle ones weren 't present. There is also the
complete nonspecificity of the CREB- deletion; it was equally expressed in
amygdala as well as hippocampus, and so could not be used to focus on the
exclusive mech anisms of declarative long-term memory.
For all these reasons, Kandel 's group adopted a transgenic approach
(Abel et al. 1997). Transgenic mice were engineered by the pronuclear
injection method (described in the subsection above) with a calciumcalmodulin kinase IIa (CaMKlla)-R(AB) transg ene. The CaMKIla promoter,
containing upstream control regions and the transcriptional initiation site,
drives expression only postnatally and limits expression to forebrain sites,
including hippocampus, neocortex, amygdala, and corpus striatum. The
cloned R(AB) DNA (cR(AB)) code s for an inhibitory form of the Ria subunit
of the PKA molecule. PKA molecules composed of this inhibitory subunit
carry mutations in cAMP binding sites and act as a dominant inhibitor of both
types of catalytic subunits. The se PKA molecules thus resist cAMP-driven
release of catalytic PKA subunits and block further steps in the cAMP-PKACREB pathway. Founder animals receiving the pronuclear injection bred
92
successfully and standard biochemical analyses reveal ed that the offspring

generations (R(AB)-l, R(AB)- 2) received and expressed the transgene .
Abel et al. (1994) used a standard in situ hybridi zation technique to
reveal the regions of R(AB) transgene expression throughout the brain. This
technique utilizes the "recipe" for the (mutated) protein coded as a sequence
of nucleotide bases that make up the transcribed mRNA. Since this sequence
is known for R(AB) (after all, the transgene had been cloned), a piece of
radioactive mRNA containing the complementary nucleotide sequence could
be synthes ized. Slices of brain tissue from variou s regions are exposed to the
radioactive RNA , which binds to the molecule s of transcribed mRNA from
the expressed transgene. The slices are then mounted on slides and exposed to
a solution containing a radioactive antibody for the protein. The slides are
then rinsed, leaving molecules of radioactive RNA in cells that contain the
complementary mRNA. These slides are developed using standard photographic emulsion. The radioactivity exposes the emul sion, and regions of the
brain where the transgene was expressed appe ar as bright images on photographic plate s. The last part of this process is called autoradiography and is
standard fare in molecular genetics research. These techniques, using a
radioactive oligonucleotide that binds to base pair sequences present in an
untran slated leader within the transgene, revealed R(AB) expression through out the forebrain of the engineered mice, especially in the hippocampus and
neocortex (Abel et al. 1997, Figure lA, B, C) .
The PKA transgenics were next tested for PKA activity in the
hippocampus and for L-LTP in Schaffer collateral-CAl slices. Both in the
presence and absence of cAMP , PKA activity in transgenic hippocampus was
reduced by more than 25%, compared to littermate wild-type control (Abel et
al. 1997, Figure 2). L-LTP in hippocampal slices, measured in the standard
way via field EPSPs to Schaffer collateral stimulation for up to three hours
following pulse-train induction stimulation, revealed that activity fell to prestimulation baseline in R(AB) transgenics in less than two hours . L-LTP in
slices from wild-type littermate controls remained heightened and constant for
the entire duration of the study (three hours after inducing stimulation) (Abel
et al. 1997, Figure 4). Intact transgenics were deficient in the Morris water
maze, using measures and probe trials similar to Bourtcholadze et al.' s (1994)
study. These results are consistent with PKA 's role in the cAMP-PKA-CREB
pathway in L-LTP induction and maintenance, as well as L-LTP' s
hypoth esized role in declarative (hippocampal-requiring) long-term memory.
But it was on the contextual-cued conditioning task borrowed from
Bourtcholadze et al. (1994) that the most convincing data with the R(AB)
transgenics emerged. Abel et al. (1997 ) found that the transgenic s were not
deficient , compared to littermate wild-type controls, on either immediate
learning or short-term memory on the contextual conditioning task. All groups
93
spent the same amount of time freezing immediately after the initial foot
shock and when placed back into the novel environment (training box) one
hour later. However, when replaced back into the context twenty-four hours
later, transgenics froze for only about one-third of the time that wild -type
littermate controls did, indicating significant reduction in declarative
(hippocampal-requiring) long-term memory. In the CS fear conditioning
(nondeclarative) task, transgenics displayed normal short-term memo ry when
tested to the tone CS 1 hour after initial training (compared to wild -type
littermates), and displ ayed normal long-term memory to the tone when tested
twenty-four hours later. That is, R(AB) transgenics spent the same percentage
of time freezing after tone presentations as did wild-type littermate controls.
Finally, transgenic performance statistically matched that of wild-type mice
treated with anisomycin, a potent nonspecific prote in synthesis inhibitor,
delivered by injection either thirty minutes before or immediately after context and cued training (Abel et al. 1997, Figure 6). This match strongly
suggests that the same molecular mechanisms are affected by the specific
transgenic manipulations and the nonspecific pharmacological intervention.
One key to Abel et al.'s (1997) results is the dissociation between
consolidation switches for declarative and nondeclarative long-term memory,
the first being compromised while the second was preserved in the R(AB)
transgenics. Presumably this result was due to a lower level of transgene
expression in the amygdala. Not only does this further the case for the
hypothesis that the cAMP-PKA-CREB molecular pathway is the consolidation switch for declarative long-term memory in behaving mammals, but it
also removes possible confounds that the behavioral deficits are due to visual,
motivational, or motor coordination difficulties. R(AB) transgenics were able
to learn and retain the CS fear conditioned response, whose visual ,
motivational, and motor demands are similar to the contextual conditioning
task .
Armed with these experimental results , and embedded within the
model of L-LTP we sketched in the previous section, Abel et al. (1997) draw
a bold conclusion:
Our experiments define a role for PKA in L-LTP and longterm memory, and they provide a framework for a molecular
understanding of the consolidation of long-term explicit
memo ry in mice . ... The consolidation period is a critical
period during which genes are induced that encode proteins
essential for stable long-term memory. The long-term
memory deficits in R(AB) transgenic mice demonstrate that
PKA plays a role in the hippocampus in initiating the
molecular events leading to the consolidation of short-term
94

changes in neuronal activity into long-term memory. (1997,
623-624; my emphases)
The 2000 Nobel Prize committee agreed. In combination with his earlier (and
continuing) groundbreaking work on the molecular basis of memory using the
sea slug Aplysia (discussed in the next chapter), Eric Kandel was awarded a
share of the 2000 Nobel Prize for Physiology or Medicine. In his presentation
speech at the ceremony, Urban Ungerstedt remarked:
I am convinced that you and I will remember this Nobel
ceremony for many years . This is because of the dopamine
which Arvid Carlsson discovered, enabling the brain to react
to what we see and hear; the second
messengers that Paul
Greengard described, carrying the signals into the nerve cell;
and the memory functions that Eric Kandel found to be due to
changes in the very form and function of the synapses. .. Eric
Kandel's work has shown us how these transmitters, through
second transmitters and protein phosphorylation, create shortand long-term memory, forming the very basis for our ability
to exist and interact meaningfully in our world. (http ://www.
nobel.se/medicinellaureates/2000/presentation-speech.html)
My mentioning this award is not a bald appeal to authority. I offer it
only to indicate the importance that the scientific community has placed on
Kandel's and his colleagues work . I should also point out that this work is not
without controversy in the neuroscience community. Controversy remains
about whether the account I described provides the correct molecular
mechanisms of LTP, above and beyond the more familiar controversy about
LTP's being the reductive link for memory consolidation (Lynch 2000). For
example, Gary Lynch and his colleagues have stressed the importance of
activity-dependent cell adhesion molecules in breaking down and then reconstructing the dendritic spine in an entirely local fashion . Their data are
intriguing (e.g., Bahr et al . 1997, Staubli, Chun, and Lynch 1998). Nevertheless, no matter how such controversies pan out, all serious alternatives now on
the table about the molecular mechanisms of LTP exhibit the same
reductionism about the "LTP-memory con solidation switch" (McGaugh,
2000). "Ruthless reductionism" is a cornerstone of this parade case from
recent molecular neuroscience.
I close this section with a final remark, before we leave the scientific
details for a philosophical interlude. I remarked at the beginning of this
chapter that many philosophers of mind and cognitive scientists bemoan "how
little we know about how the brain works ." To hear these folks talk, it is as
95
though we're still flummoxed by the brain's complexity and warring anxiously for some novel experimental breakthroughs or theoretical insights. I
hope that by reaching this point in this book, readers now reject this common
misunderstanding. Current "ruthlessly reductive" neuroscience knows a lot
about "how the brain works," at least about its basic constituents and how
they interact; and it possesses tools of discovery that we can confidently
expect to increase our knowledge. That this knowledge and these tools have
been ignored by higher level theorists about mind-even by some who fancy
themselves "philosophers of neuroscience"-is an interesting datum for "sociologists of philosophy." Some might find it difficult in the abstract to see how
mind reduces to molecules; but one will never get over that intellectual hurdle
or show conclusively why such reductions can't obtain by remaining ignorant
of the best existing scientific attempts to do exactly this.
6 THE NATURE OF "PSYCHONEURAL REDUCTION"

AT WORK IN CURRENT MAINSTREAM (CELLULAR
AND MOLECULAR) NEUROSCIENCE
Perhaps this case doesn 't meet everyone's pre-theoretical notion of
what "reduction" amounts to or requires. Philosophers might ask: Where are
the cross-theoretic bridge laws? Where are the explanatory generalizations or
laws that compose the two theories? Where is the logical derivation of
reduced from reducing theory? I've already proposed that we set aside
preconceived notions and consider what the relation amounts to in this
detailed example of an actual scientific reduction-in-practice. To generalize
from this example, I will employ the metascientific resource of explanatory
posits becoming "structured through a reduction" to lower level sciences. In
the "structuralist" tradition in the philosophy of science, C.U. Moulines
(1984) has sugge sted such a resource, and in Bickle (2002) I develop it further
by applying some quasi-formal machinery to illuminate an abbreviated
version of the memory consolidation-molecular mechanisms of LTP link.
Here I will ignore the formalism and elaborate on the conceptual detail s.
In the previous sections we've heard much of cross-theory psychological-to-cellularlmolecular "links." Can we cash that metaphor? We can say
at least this much: entities, properties, and events postulated in psychological
explanations (as in higher level theories generally) typically get characterized
functionally, purely in terms of their causes and effects, with little or no
explicit concern for the underlying neurobiological (lower level) events and
processes producing this functional profile. Memory consolidation is a
process that typically requires multiple stimulus presentations or rehearsals
96
and can be disrupted by certain events that occur during a short temporal
interval after presentations (its causes). It transforms a labile , easily disrupted
short-term memory into a stable, durable long-term memory with related
memorial or cognitive content (its effects). In this fashion, the psychological
concept resembles that of "gene expression" in Mendelian through transmission genetics. The latter process is there characterized in terms of trait
ratios in various breeding combinations across parent and offspring populations using theoretical posits-'allele,' 'dominance'-defined in their terms.
In current molecular genetics, however, these purely functional concepts get
linked to elaborate sequences of molecular and biochemical transcriptional,
translational, and recombinatory pathways. This linkage is analogous to the
way psychology's "memory consolidation switch" posit gets linked to the
intracellular molecular processes of E- to L-LTP, occurring in hundreds of
thousands of selective neurons to alter synaptic efficacy and ultimately
behavior. But these examples only set up a problem for metascientific
analysis, and philosophers of science have not carried this idea of posits
"structured through reduction" beyond a few illustrations. I propose to go
further, using the details of the example provided in this chapter.
It can be useful to characterize scientific theories in terms of models
and intended empirical applications. A theory's models are the systems, both
real world and mathematical, that share the structure characterized by the
theory's fundamental assumptions and explanatory generalizations. Its
intended empirical applications are all those real-world systems to which at
any given time the theory is thought to apply (by the appropriate scientific
community), the systems thought to possess the structure of the theory's full
contingent of fundamental assumptions and generalizations. At any given
time , some of a theory 's intended empirical applications will have been
shown to be actual models, i.e., to have the structure characterizing the
theory's models. Other intended empirical applications will not yet have been
shown to be models. Some of the latter will be in the process of empirical
test. 28
Using this scheme, a theory's models can be usefully characterized as
consisting of three components:
"empirical" or "base sets," whose elements are grouped into the

fundamental categories by which the theory "carves up" the world ;
"auxiliary sets," the mathematical or other abstract spaces the theory
employs in its explanations; and
fundamental relations and functions, defined as operations over
elements of the other two components, characterizing the theory's
basic explanatory properties and relations.
97
All these components, but especially the last type, combine to specify the
theory's basic assumptions and generalizations. The structure of the latter
characterizes the structure of the theory's models.
A simple example helps to illustrate these abstract descriptions.
Consider the theory of classical collision mechanics. Each of its models are
composed of two empirical base sets (a set of particles and a set of time
instances), one auxiliary set (the set of real numbers), and two fundamental
relations (the mass relation, which assigns positive real numbers to elements
of the particle set; and the velocity relation, which assigns ordered triples of
real numbers to particles at time instances). These elements combine into the
theory's one law, the conservation of momentum before and after a particle
collision (namely, the sum for all particles p of the mass of p times the
velocity of p at time instance t] equals the sum for all particles p of the mass
of p times the velocity of p at time instance t2, where t] is prior to and t2 is
after a particle collision). Any system, real world or mathematical, that
possesses this structure is a model of classical collision mechanics. Its
intended empirical application is to every real world system composed of
particles with mass and velocity and time instances. (Theories in the special
sciences have less broad sets of intended empirical applications.)
From this perspective on the structure of theories, reduction is partly a
mapping of specific models and intended empirical applications across the
two theories that meets a variety of limiting conditions (not any old mapping
constitutes a reduction relation). One of these limiting conditions is the
requirement of mappings of each empirical base set of the reduced theory into
a base set, a fundamental relation, or combinations of the two of the reducing,
guided by the two theories' shared intended empirical applications.i" In some
theory reductions, empirical base sets of models of the reduced theory get
mapped onto identical base sets of reduction-related models of the reducing
(in the extensional, set-theoretic sense of identity-the mapped sets contain
the same elements). For example, the set of planets in the intended empirical
application of Kepler's astronomy is identical to the set of planets in the
reduction-related intended empirical application of Newton's celestial
mechanics. But typically, in interesting scientific reductions, some (if not all)
of the empirical base sets of models of the reduced theory get mapped onto
nonidentical sets making up the reduction-related models of the reducing.
Examples from the history of science abound (Moulines 1984). In the
reduction of rigid body to classical mechanics, the empirical base set of rigid
bodies in models of the former gets mapped to that of Newtonian particles in
reduction-related models of the latter. But elements of the former set are not
elements of the latter. Neither set identity of inclusion holds across this link.
No rigid body is itself an element of any set of Newtonian particles,
98
A second example is even more analogous to our case study, where a

single empirical base set in models of a reduced theory gets linked to a
sequence and combination of both base sets and fundamental relations in the
reduction-related models of the reducing theory. This is the reduction of
Mendelian to molecular genetics." Elements of the base set of genes,
characterized solely (i.e., purely functionally) by Mendel's laws of Segregation, Independent Assortment, and Dominance, are not members of any
empirical base sets in models of molecular biology . The latter are sets of
organic molecules and the theory's fundamental relations specify their
interactions to generate DNA transcription, messenger RNA translation,
protein synthesis, and phenotype production. The two ways that the reduced
and reducing theories "carve up" the world of genetic inheritance and
expression differ, as expressed in the different elements composing their
models' empirical base sets, fundamental relations, and explanatory generalizations. Nevertheless, the base sets and fundamental relations making up
models of these two theories get related as a condition on the global theory
reduction relation across models of the two theories, and the two theories'
intended empirical applications (the real-world biological systems expected to
have the structure of models of the two theories) are virtually identical.
In cases like this, where an empirical base set in models of the
reduced theory get linked to sequences and combinations of base sets and
fundamental relations in reduction-related models of the reducing, we also get
an account of what it is for "amorphous" basic entities of the reduced theory
to become "structured through reduction ." An entity or process, characterized
entirely by way of the fundamental relations and generalizations of the
reduced theory, comes to be related in a domain eliminating way to sequences
of entities and processes characterized by the relations and generalizations of
the reducing theory. From the perspective of their reducing theories, there are
no rigid bodies, separate space and time, or Mendelian genes. Elements of
these empirical base sets aren't elements of the empirical base sets or
fundamental relations and functions by which their reducing theories (respectively, classical particle mechanics, special relativity theory, and molecular
genetics) "carve up the world." However, the roles that elements of these
empirical base sets play in the relations and generalizations that partly define
models of the reduced theories bear interesting structural similarities to the
roles played by sequences and combinations of empirical base sets and
fundamental relations in reduction-related models and empirical applications
of the reducing theory . In particular, entities characterized on the reduced
theory primarily by their functional (input-output) features get linked to
complex structures (sequences and combinations of entities and processes)
whose dynamics and interactions specifiable entirely within the reducing
theory
99
1. apply to roughly the same intended set of real-world systems, and

2. provide causal mechanisms that explain the former's functional
(input-output) profile.
This, in rough outline, is the metascientific concept of a theoretical posit
becoming "structured through reduction ."
We can abstract this metascientific resource out of the deta iled
memory consolidation-molecular mechanisms of LTP reduction. Models of a
psychological theory of memory , by their empirical base sets, fundamental
relations, and generalizations characterized in terms of these , posit an
entity/process, the consolidation switch. But they characterize this posit only
in terms of the time course and amount of repetition needed to convert a given
type of memory item from short-term memory to long-term memory and the
behavioral efficacy of different types of retrograde interference. In other
words , psychology characterizes this entity/process in purely functional
fashion, in terms of the events causing it and the events it in turn causes, with
no concern for the specific underlying neural mechanisms generating this
functional profile. (Notice also that psychology does the same for short-term
memory and long-term memory, characterizing both primarily in terms of
duration and distractability of recall after initial presentation .) Cellular and
molecular neuroscience then developed an account of the mechanisms of
LTP, and behavioral neuroscience verified that these mechanisms underlie
memory consolidation selectively in behaving animals (using, e.g., the Kandel
lab's R(AB) transgenic mice). The consolidation switch empirical base set in
models of psychology got mapped to sequences and combinations of
empirical base sets and fundamental relations in reduction-related models and
intended empirical applications of molecular neuroscience, in particular to
those involved in the transition of E-LTP into L-LTP and the maintenance of
L-LTP. The empirical base sets of the latter include intracellular and neural
transmission molecules: adenylyl cyclase, cAMP, PKA, CREB enhancers and
repressors, DNA, RNA polyrnerases, ubiquitin hydrolase, CCAAT enhancer
binding protein, glutamate, dendritic spine cytoskeleton components, AMPA
receptors, NMDA receptors, and so on. Out of these components , the
fundamental relations of cellular/molecular neuro science build membranes,
plastic neurons, neural circuitries from sensory input to motor effectors,
neuromuscular junctions, and the specific conn ections with the muscles and
skeletal systems (themselves built up by fundamental relations out of their
molecular elements). Molecular neurobiology 's fundamental relations and
generalizations also specify these elements' subcellular phy siological and
extracellular interactions: phosphorylation, selective molecul ar bindings, gene
transcription and translation, protein synth esis , chan ged receptor affinity for
100
ion conductance, changing protein configuration, ions conducting through

selective protein channels, and neuromuscular interactions leading to
combinations of muscle contractions pulling against a calcium frame (the
skeleton).
The mapped posits of the two theories play structurally similar roles
across reduction-related models and intended empirical applications, namely ,
mammals behaving in observable, quantifiably measurable ways in tasks
involving stimulus expo sures earlie r in time . However, elements of
psychology's "consolidation switch" empirical base set are not elements of
the sequences and combinations of molecular neuroscience's empirical base
sets and fundamental relations across reduction-related models and intended
empirical applications. Cellular and molecular neuroscience "carves up the
world" of behavioral memory data in a fundamentally different way than
psychology, even though the intended empirical applications of the two
theories are virtually identical. The detailed scientific example of reductionin-practice described in this chapter gives us a beautiful example of a
psychological posit-the consolidation switch-"structured through reduction" to a molecular process-the mechanisms of L-LTP-occurring in the
neurons comprising the appropriate anatomical circuits. Such is the metascientific nature of the reductionism both endemic to and successful within
current mainstream neuroscience ."
A useful visual metaphor of this abstract cross-theory link is two nets
composed of strings tied onto rings (Figure 2.11). The rings are analogous to a
theory's posits (empirical base sets and fundamental relations) while the
connecting strings are analogous to the causal connections its generalizations
postulate to hold between them . The rings of the reduced theory have
significantly greater diameter than the rings of the reducing, and the lengths of
strings between the reduced theory's rings are considerably longer. This
reflects the much more coarse-grained , purely functional way that the reduced
theory "carves up the world" (the shared intended empirical applications)
compared to the reducing. Notice that the specific ring sizes and string lengths
making up the reduced theory do not belong to the reducing theory. In the
latter, there are no rings of those specific diameters or string segments with
those specific lengths and connectivity patterns . But if we lay the reduced net
on top of the reducing-the analog of relating an intended empirical
application of the reduced to one of the reducing theory , as part of the theory
reduction relation-we see which collection of reducing rings and strings
(posits and causal connections between them) gets located underneath the
reduced rings and string s. Applied to our case study, we see how (and not just
that) the emerging cellular/molecular story arranges its constituents together
into a sequential and combinatorial structure abstractly similar to
psychology 's coarse-grained functional posits. In our example, psychology's
101
consolidation switch-one type of big ring-is an important functional

approximation of the cellular/molecular mechanisms that signal the switch
from E-LTP to L-LTP and maintain L-LTP selectively in many synapses
throughout the neuronal network. Visual metaphors are crude and have their
obvious limitations, but this one captures some important aspects of the nature
of psychoneural reductionism-in-practice in our detailed example from
current mainstream neuroscience.Y
Figure 2.11 . Visual metapho r for the structure of intertheo retic (scientific ) reduction. Large
rings represent posits of the reduced theory; the thicker strings connecting them represent
posited causal relations by that theory's explanatory generalizations. Small rings represent
posit s of the reducing theory ; thinner strings connecting them represent posited causal relations
by that theory' s explanatory generalizatio ns. See text for detailed discuss ion. Figure created by
Marica Bernstein.
If you are still not satisfied with the example of redu ctionism-inpractice presented in this chapter or the metascientific account I abstracted out
of it, you might examine your reasons. Perhaps some prior epistemology you
are applying to science leads you to a view about "what reduction has to be."
Perhaps you are influenced by the nature of reduction in other disciplines
(physics or mathematics). Perhaps you are guided to a theory of reduction
based on prior ontological considerations about "what physicalism requi res."
Perhaps the resources you employ to theorize about science aren't neutral, but
influence you to understand the reduction relation in a particular fashion, e.g.,
perhaps you still conceive of theory structure in terms of the syntax of
predicate logic and so still think about intertheoretic relations in terms of firstorder deductive validity. Or perhaps you're motivated by an external
normative consideration; perhaps you want a ju stification , external to the
results and accepted explanations of current science, e.g., of why this research
is where funding should be going or what level of research neuroscientists
102
should be pursuing.f Regardless of what might motivate dissatisfaction with

my calling the scientific results reported in this chapter a "reduction," the
facts revealed here are these. Psychoneural reductionism is alive and thriving
in current cellular and molecular neuroscience, as revealed in the attitudes of
its practitioners and their choices of experimental manipulations and investigations. And the specific nature of the reductionism encapsulated within these
practices and results is usefully illustrated by the "structuring" of psychology 's purely functional posits into specific sequences and combinations of
cellular and molecular entities, processes, and causal interactions. That is
reductionism-in-practice in current mainstream neuroscience and its metascience. And if it doesn't meet your borrowed ideal of "what reductionism has
to be," perhaps first philosophy and a priori normative epistemology really IS
your cup of tea. The target of this final comment is not just the "American
Philosophical Association" crowd, but also many "naturalistic" philosophers
of mind-despite the latters' denial of caring about such "traditional
philosophical" concerns. With regard to declarative memory consolidation,
"ruthlessly reductive" neuroscience is already down to the "molecular
pathways" level.
NOTES
1 It might seem strange to talk about the spinal cord and memory , but this has recently received
some attention in the study of chronic peripheral pain . In fact, the same molecular mechanisms
involved in cortical and hippocampal long-term potentiation underlie the experience-driven
synaptic plasticity hypothesized to explain features of chronic pain. See Sufka (2001) for a
clear exposition of this hypothesis .
these scientific details elude you now, fear not. Later in this chapter I'll explain in detail the
molecu lar mechanisms of LTP and experimental protocols and results in studies of this sort.
For novices, I'll even include a brief "primer" on basic cellular neuroscience .
21f
3 Schout en and de long inform me that this argument in their (1999 ) was intended to reveal the
cross-categorizations that obtain between psychological and neurobiological concepts,
rendering any sort of "bottom-up" explanatory approach untenable . If so, then their argument
rests upon issues I will take on in Chapter Three, namely, question s about methodology within
neuroscience dominated by the search for cellular and molecular mechani sms and the issue of
multipl e realization.
Many readers will be familiar with well-known exceptions to the stimulus repetition feature .
One-trial learn ing that remains stable for long periods has been studied for more than four
decades. Typically, these are species-specific and evolutionarily prominent learning processes,
e.g., conditioned taste aversions in rats . See, e.g., Garcia and Koelling (1966) (although most
reputable learning theory textbooks will include a description of this phenomenon) . For ease of
exposition, I won't be qualifying the "stimulus repetition " condition on consolidation in the
discussion to follow . Fans of one trial learning can make the appropriate mental adjustments to
my assertions, e.g., that stimulus repetition and rehearsal typically improves long-term memory
4
103
recall and performance. Than ks to Carl Craver for pointing out the need to qualify some
remarks I make in the discussion below .
5
This study is summarized nicely in Squir e and Kandel 1999, 130-132.
Duncan also measured the l atency of time spent in the test compartment after rats were placed
there initially, but got little useful information from this behavioral measure and didn 't includ e
it in his publi shed results .
This point is not the only thing that "cognitivists" and "autonomists" could and do say. We'll
return to this point in earnest in the first two section s of Chapter Three . This discussion also
foreshadows that about declarative memory, later in this chapte r.
I'll explain some of these biote chnol ogical manipulations later in this chapte r.
9 Cajal' s original publication is in French . Squire and Kandel (1999 , 35-36) give a brief report
of Caja l' s speculat ions .
10 Material presen ted in this subsection is abbreviated standard neurob iology textbook fare.
Any reputable recent textbook will prov ide additional details . For those interested in a state-ofthe art account of how neurons work, I recommend Levitan and Kaczmarek (200 1).
11 The other type is neuroglia, a kind of connective tissue. The many roles of glial cell s, and
neuron -glial interactions, are targets of much current research. Much of this is beyond the scope
of our concerns.
12 The "spike" metaphor describes the shape of the event recorded on an oscillo scope , as
pictured in Figure 2.3.
13 Obviously, the "presynaptic terminal " of the primary synapse is postsynapt ic to the
modulat ory neuron . More recently discovered complexities render the original "pre -" and
"postsynaptic" terminology confus ing.
14 Thanks to Carl Craver and Ken Sufka for reminding me of this point. Sufka poin ts out in
particular the new evidence of "pattern coding" in axons above and beyond rate/frequency
coding of their action potentials (e.g., in Reichling and Levine 1999). I have not studied this
literature carefull y; nevertheless it is difficult for me to see how it escapes the general poi nt I' m
urging in this paragraph (and the next). At bottom, however, this is an empirical issue . If
someone can show me specifically where or how this argument is wrong in light of some new,
empirically confirmed coding discovery, I'll give up this version.
15 This subsection is not intended to be a comprehensive early history of LTP . Philosopher of

neuro science Carl Craver has a forthcoming manuscript (2003) that explores this history in
excellen t detail. Part of his history is based on interviews with key participants.
16 Notice that this is amplitud e increase in population spike, not in individual spikes. (Recall the
all-or -none law of the individual action potential.) This means that the single stimulating pulse
to the perforant path caused more dent ate neuron s to reach threshold of excitation after the
stimulus conditioning train had been delivered , compared with before.
17 I'll say more about the hippocampus and decla rative memory in the next section, with
referen ces to the recent literatu re.
18 The molecul ar genetic details are not cruci al here , but will become so later in this section. I
will then provide a brief prime r on the molecular mechani sms of gene transcription .
104
19 A receptor an/agonist blocks the transmitter's effect , typically by competing with transmitter
molecule s for receptor binding sites but not initiating the transmitter's activity when bound . On
the other hand , a receptor agonist increases the efficiency of transmitter molecules at the
receptor site.
This much molecular genetics is sufficient for our purposes in this section, but I'll present
more details in section 5.2 of this chapter when I turn to biotechnology's contribution to the
neurobiology of memory consolidation. For readers interested in pursuing this fascinating area
more thoroughly, Lewin (1999) is a state-of-the-art textbook for current molecular genetics, but
any reputable introduction to biology text will contain a discus sion of molecular genetics in
plenty of detail for the average non-scientist. Consult your local Biology department.
20
CREB proteins only have their transcriptional effects when phosphorylated. Recall from
above that catalytic PKA subunits have translocated to the neuron's nucleus . These translocated
subunits phosphorylate CREB proteins .
21
Clayton (2000) is a useful recent review , although his "operational definition " of 'i mmediate
early gene ' does not square exactly with its usage by other molecular neuroscientists.
22
A detailed discussion of celebrated case H.M. with references to the primary literature can be
found in any reputable neuropsychology textbook, e.g., Kolb and Whishaw (1996) , 357-360.
23
24
Subsection 5.2 is co-authored by Marica Bernstein .
Recall from the brie f description in subsection 4.2 of this chapter the basic dogma of
molecular genetics: DNA -7 (transcription)-7 RNA -7(translation)-7protein.
25
This raises an interesting question, especially given CREB 's ubiquitous occurrence in all
types of biological tissues. Why aren 't the knockouts more deficient? Hummler et al . (1994)
argue that CREB operates in tandem with two other cAMP-driven transcriptional activators,
cAMP response element modulation protein (CREM) and activating transcription factor I
(ATF I) , which can compensate for each other. CREB - mutants do in fact ovcrcxpress CREM
in all tissue types .
26
27 More intensive training-three blocks of four trials per day over three training daysovercame these statistical differences between CREB - mutants and controls. Bourtcholadze et
al . (1994) suggest that this is due to CREM compensation for the CREB deficiency (mentioned
in the previous footnote).
This description constitutes an informal account of the structuralist program 's concept of a
theory-element. Most structurali sts find that set theory and category theory provide useful
mathem atical resources to characterize these concepts formally. See Balzer et al., (1987) ,
Balzer and Moulines (1996) , and Bickle (1998) for detailed presentations, including models of
intertheoretic reduction (Balzer et al. 1987, chapter 5, Bickle 1998, chapter 3) and an
application to earlier developments in the reduction of psychological theorie s of memory to
neurobiological counterparts (Bickle 1998, chapter 6).
28
The relations making up this condition are called "ontological reduction links" by Moulines
1984. 1adopt this terminology in Bickle 1998, chapter 3, and 2002) .
29
That this is a case of reduction is very controversial in philo sophy of biology . However , the
molecular geneticists 1 know have no qualms about calling and treating it as such . Perhaps a
reduction-in-practice of recent work in that discipline would be a useful contribution to
philo sophy of science .
30
31
105
To see these features developed in quasi-formal fashion , see Bickle (2002 ).
I've used a "net" metaphor, across changing concerns and philosophy of science
backgrounds, to express ideas about theory reduction in psycholo gy and neuroscience since one
of my earliest papers (Bickle 1992b). See elaborations in Bickle ( 1998, chapter six).
32
33
Thanks to Carl Craver for stressing this "normati vity" worry.
CHAPTER THREE
MENTAL CAUSATION, COGNITIVE
NEUROSCIENCE, AND MULTIPLE
REALIZATION
We next trace implications of the detailed example from Chapter Two

for two prominent issues in current philosophy of mind and one increasingly
prominent area of current neuroscience. Reduction is central to all three. The
first philosophical issue-the problem of mental causation-questions
whether or how mental properties can exert causal effects on behavior. The
second philosophical issue-the multiple realization argument-is widely
thought to be one of two decisive arguments against the reduction of mind to
brain or psychology to neuroscience.' We saw in the previous chapter that
"reductionism" is alive and well in current cellular and molecular
neuroscience. Now we'll see whether that reduction-in-practice and its results
carry helpful implications for issues that have attracted serious philosophical
attention. The scientific issue concerns the status of cognitive neuroscience
vis-a-vis the discipline's cellular and molecular core. How do studies involving, e.g., neuron population dynamics or specific activations across neural
regions, relate to ones exemplified by our detailed example from the previous
chapter?
1 THE PROBLEM OF MENTAL CAUSATION

"Mental causation" is an ideal problem for modern-day philosophers
of mind and cognitive science. That our mental states are causally efficacious
for our behavior is central to our shared self-conception. Ruth answered "B"
to question #3 on the personality assessment because she believes that she is
kindly. Alonzo signaled the waiter because he wants another martini. The
simplest reading of these common assertions is that the mental properties of
our cognitive states-e.g., the contents of Ruth's belief and Alonzo's desirewere part of the nexus that generated the action (including the limb
movements). Had Ruth's or Alonzo's causally efficacious belief or desire
contents been different, different behaviors would have resulted; perhaps arm
movements with the pencil that result in Ruth 's drawing a circle around
108
answer "C ," and ones that result in Alonzo's brining the martini glass up to
his open mouth. Yet this common feature of our self-conception raises
delicious puzzles. For example, are mental contents in their capacity as (or
"qua") mental contents (and not, say, in their capacity as the brain states that
realize them) causally efficacious, or are content properties like the color of
the benzodiazepine tablet with regard to the pill's tranquilizing effects? The
soprano hits a high C note while uttering the English word "Shatter!" The
glass shatters. Clearly the semantic properties of her utterance had no causal
effects on the glass's shattering. That was entirely an effect of the event's
acoustic properties on the microstructure of the glass 's molecules. Had the
event's semantic properties differed while the acoustic properties remained
constant, the glass still would have shattered. Might mental properties be like
that: causally inefficacious, with all the causal work on behavior being done
by the event's physical (e.g., neuronal) properties? That suggestion is contrary
to our self-conception of being the possessors of causally efficacious mental
states. But how are mental properties different in a way that permits us to
locate them justifiably in the causal fray? A little reflection on this puzzle and
philosophers are off and running. The mental causation literature contains
theories of "supervenient causation," "causal-explanatory exclusion," "downward causation," "ceteris paribus laws," "causal compatibilism," and other
philosophical exotica. And from such humble origins!'
To my lights , Terence Horgan (2001) has recently provided the best
formulation of the metaphysical and epistemological intuitions driving both
sides of the mental causation debate. He presents them as an inconsistent
quintet of claims, each plausible individually but inconsistent when
conjoined?
1. Physics is causally closed. Each physical event is determined (to
whatever extent it is determined) completely by purely physical
events.
2. Mental properties are causal properties
3. Mental properties are not identical to physical causal properties,
because the former are multiply realizable on the latter."
4. Mental properties are real and instantiated in humans .
5. If physics is causally closed , then all causal properties are physical
causal properties.
(By 'physical property' Horgan means the kind of property postulated in
fundamental physics.) Any four of these intuitively plausible claims is
consistent, but those four conjointly entail the falsity of the other. The
"problem of mental causation" becomes which claim to reject. "Causal
emergentists" deny claim 1, "epiphenomenalists" deny claim 2, "identity
CAUSATION, COGNITIVE NEUROSCIENCE, REALIZATION
109
physicalists" deny claim 3, "elirninativists'' deny claim 4, and "causal

compatibilists" deny claim 5. Each view "solves" the problem of mental
causation, but at the expense of an intuitively plausible metaphysical or
epistemological claim .s
Some physicalist philosophers-reductive physicalists, mind-brain
identity theorists, "central state" materialists, "brain state theorists"-are initially attracted to Jaegwon Kim's "causal-explanatory exclusion" arguments
(1993, 1996, 1998). Using Horgan's (2001) schema, Kim's arguments defend
claim 5. Physicalists so moved will thus opt for rejecting either claim 3 or 4.
Horgan (2001) gives an insightful presentation of Kim's reasoning. First,
since philosophical orthodoxy insists at least that all mental causal properties
"supervene upon" or "are realized by" physical causal properties, the ultimate
causes of any mental property's becoming instantiated are themselves
physical properties. In particular, they will be whichever physical properties
cause that mental property's physical realizers to occur. Second, if mental
properties are both causal properties and distinct from physical properties,
then either
A)
some events depend causally (at least in part) on the prior

occurrence of some mental property(ies),
B)
some events are causally overdetermined by individually

sufficient physical and mental properties.
or
If (A), then the effect-states can't be physical or else the causal closure of
physics is violated. There would then be physical events for which physical
causes would not be sufficient. But the effect-states can 't be mental, either,
without violating the demand of physical supervenience or realization of
mental properties. There would then be mental properties for which physics
does not provide a complete supervenient base or realization. So (A) is
incoherent (for a "minimal" physicalist). Given the causal closure of physics,
mental properties can only be causal properties via the physical properties that
realize them . But that rules out (B) since there couldn't then be causal overdetermination in any real sense . There would then be no independent causal
route leading from mental cause to effect. Kim's upshot is that the physical
properties "do all the causal work." Physical causal explanations of behavior
"screen off' or "exclude" mental causal explanations.
Unfortunately for physicalists, and despite its seeming reasonableness, Kim's causal-explanatory exclusion argument has been attacked
relentlessly by the philosophy of mind orthodoxy. Tyler Burge (1993) and
110
Lynne Rudder Baker (1993) have urged that our commitment to mental
causation, as reflected in our ordinary mentalistic explanatory practices, is far
stronger than our commitment to abstract physicalist metaphysical principles.
We should focus on the former and learn to love expl icit mental causation. At
least since his (1993) essay , Terence Horgan has argued that treating causal
claims as reflecting "counterfactual dependencies" renders mental causal
claims seemingly obviously true, despite any physicalist metaphysical
scruples we might hold. We all readily agree that had Alonzo not desired a
beer (counterfactually), then ceteris paribu s he wouldn 't be reaching right
now for the Schlitz can in his fridge. Numerous philosophers have urged "the
generalization problem" on Kim. Doesn't causal-explanatory exclusion do
away with all causation "higher than" basic physics , including the neurobiological, molecular biological, and biochemical ; and isn't a metaphysics that
eschews that much simply reduced to absurdity? Physicalists of course have
some retorts. Kim himself offers a number in chapter 3 of his (1998) book.
But the debate quickly takes on the "fruitless clash of intuitions" flavor I
vowed to avoid in Chapter One. I for one am not an enthusiast for wading into
this argument here.
So I propose a different tack. I want to appeal to our detailed
scientific example from Chapter Two and remind readers of the two distinct
theoretical/explanatory levels at work: memory consolidation from experimental psychology and the molecular mechanisms of LTP from cellular and
molecular neuroscience, both applied to behavioral data like that from the
Kandel lab. I contend that when we fix our gaze on aspects of scientific
practice in this actual recent example, we see that psychological explanations
lose their initial status as causally-mechanistically explanatory vis-a-vis an
accomplished (and not just an anticipated) cellular/molecular explanation. All
attempts by philosophers to "save" mental causation presuppose that
psychological explanations remain (causally) explanatory, at least in certain
contexts, even in light of accomplished "low level" neurobiological
explanations of the same behavioral data . I'll argue instead that within
scientific practice, psychological explanations become otiose when the type of
cellular/molecular explanation encapsul ated in the detailed example and now
dominant in mainstream neuro science is achieved. There is no need to evoke
psychological causal explanations, and in fact scientists stop evoking and
developing them , once real neurobiological explanations are on offer (but not
merely "on promise"). Philosophers who deny this point are usually guided by
outdated accounts of real neuroscientific practice. Contra Kim, lower level
explanations need not "exclude" higher-level accounts in any deep epistemological or metaphysical sense. But the former do render the latter pointless,
along with any furth er search fo r empirically improved successors at the
same level-except for some residual, purely heuristic tasks. I aim to show
111
that accomplished lower-level mechanistic explanations absolve us of the

need in science to talk causally or investigate further at higher levels, at least
in any robust "autonomous" sense.
2 LETTING SCIENTIFIC PRACTICE BE OUR GUIDE

To articulate and defend these claims, I propose that we let scientific
practice be our guide. As cellular/molecular accounts of memory consolidation emerge, like the Kandel group's, what becomes of psychological-level
research on the topic? I'm asking here about the empirical search for causal
explanations of behavioral data. The answer is, it disappears. Circa 2002,
there is no "psychology of memory consolidation ." No one is searching for
empirical refinements or supplements of, e.g., Duncan's (1949) causalexplanatory resources (Chapter Two, section 2 above). Psychological research
into the mechanisms of memory consolidation hit an explanatory wall
throughout the middle of the last century and neurobiology (including
molecular genetics) quickly took over."
We must realize how surprising this shift should be for anyone who
advocates the "autonomy" of psychology from neuroscience. Neuroscientists,
and even many experimental psychologists, may be surprised to learn just
how popular "psychological autonomy" is among ph ilosophers of mind. It is
the orthodox view at present, due to its connection with the dominant
"nonreductive physicalist" solution to the mind-body problem. Memory
consolidation would seem like a natural explanadum for the psychological
autonomist. We saw in the previous chapter (end of section 2) how its basic
empirical features-the typical need for stimulus rehearsal and the impact of
retrograde interference-can be given natural "cognitive" interpretations.
Rehearsal amounts to "calling to mind " repeatedly the representational
content of stimulus presentations available in short-term memory. Retrograde
interference disrupts the required maintenance of the contents of memory
representations. Such accounts fit with the widely accepted "mark of the
cognitive," as requiring explanations that advert to operations over the
contents of cognitive representations. Furthermore, these principal features of
memory consolidation are operative especially in mammalian "declarative
memory," the "flexible" form compromised in human amnesia and its best
current primate models (Chapter Two, section 5.1 above). In memory con solidation, especially as it occurs in the higher forms of memory prevalent in
mammals, we appear to have clear examples of mental causation. The content
properties of the short-term memories are part of the causal processes,
primarily rehearsal, that generate long-term memories with specific, related
contents. Investigating the nature of the "consolidation switch" seems surely a
112
task for psychology, not neuroscience, and especially not low-level cellular
and molecular neuroscience that invest igates receptor biochemistry, membrane ionic conductance, and intra-neuron signaling pathways. Yet as we saw
in the previous chapter, in great detail, the scientific facts are otherwise .
The lesson I drew in the previous chapter from the detailed scientific
example pertained to the nature of real reductionism in real neuroscience.
Now I draw a second lesson . In light of this existing cellular/molecular
explanation and these experimental results, it seems silly to count
psychology's "explanation" of consolidation as "causally explanatory,"
"mechanistic," or a viable part of any current scientific investigation still
worth pu rsuing. The explanation, "Kurt remembered that telephone number
today that I relayed to him yesterday because he rehearsed it mentally fifteen
times without retrograde interference for thirty minutes after he heard it,"
pales in comparison to one that appeals to activity-dependent molecular
(including molecular genetic) mechanisms occurring in millions of selective
neurons during the "consolidation phase ." In any case , that is the lesson from
scientific practice. Recall from Chapter Two that what psychologists called
"stimulus rehearsal" (and "repetition") amounts to increased activity and
glutamate release in selective presynaptic axons, enhancing the amount of
freed PKA catalytic subunits in the postsynaptic cell , permitting these
subunits to translocate to the neuron's membrane and exert their molecular
genetic effects that lead to long-term structural changes that lastingly
potentiate the neuron's response to similar inputs later. What psychology
called "retrograde interfe rence" gets explained as proces ses initiated after
initial stimulus presentations that interfere with any of the cellular/molecular
steps generating L-LTP. The Kandel lab's ingenious behavioral data, using
PKA regulatory subunit transgenic mice as the key experimental group,
establishes that this molecular pathway is a mechanism for long-term memory
behavior in mammal s.
When considered in isolation from available cellular/molecular
neuroscientific explanations, or when consid ered prior to (or during) their
development, psychological explanations appear genuinely causal. Let's
develop the example I suggested in the previous paragraph in more detail.
Kurt has just executed a serie s of limb movements that resulted in the
successive depression of raised buttons on a telephone's vertical faceplate
numbered "3" "6" "5" "4" "9" "0" "9". How might the "psychology of
memory consolidation" explain this behavior?
He rehearsed the verbal sequence fifteen times after I spoke it to him

yesterday and no effective retrograde interference occurred during the
thirty minutes that followed.
113
On its face, this sounds perfectly reasonable and perfectly causally

mechanistic. Kurt's rehearsals combined with no retrograde interference
caused the number sequence to be consolidated into a long-term memory
accessible today . However, once we have an experimentally verified account?
of the underlying neuronal/molecular mechanisms tied directly to controlled
behavioral data, we can juxtapose the two "explanations" and assess their
claims to revealing the event's causal-mechanistic nexus. Compare the above
with:
Following the auditory stimulus, activity in frontal neurons led to

dopamine release onto hippocampal neurons that had also been
activated by anatomical pathways from stimulated auditory receptors.
These combined activations in thousands of selected neurons led to
increased levels of freed PKA catalytic subunits in them through
intracellular adenylyl cyclase-cAMP-PKA interactions, enabling
some of these molecular subunits eventually to translocate to affected
neurons' nuclei. There these molecules phosphorylated CREB proteins that in turn bound to CRE sites on various genes' control
regions, turning on transcription, translation, and ultimately protein
production. These new proteins kept the PKA catalytic subunits free
from regulatory subunits and so in a persistently active state, and
activated expression of late-response genes that transcribed proteins
required to lastingly enhance synaptic efficacy . This process was also
occurring in synapses of selected neurons downstream from the
hippocampus, leading ultimately to specific motor effectors. Twentyfour hours later, inputs similar to the activation that originally induced
these selective changes generated the sequence of action potentials in
the motor neurons to orchestrate sequences of muscle fiber contractions and relaxations, pulling against Kurt's skeletal frame, to produce
the dynamical pattern of limb movements that depressed those numbered buttons in that order on the telephone's vertical faceplate .
Of course, as the details articulated in the previous chapter reveal, even this
explanation is condensed and greatly oversimplified in light of currently
available knowledge.
Juxtaposed next to each other, one need not possess a detailed
philosophical theory of causal explanation to judge how explanatorily
impotent-how empty-the psychological causal story has become. The
cellular/molecular neurobiological account explains many key causal
processes that the psychological account is either completely blind to or
leaves as input-output black boxes. In other words, it explains events that the
psychological account leaves unexplained. When we have neurobiological
114
cau sal explanations actually in place, psychological causal explanations are

rendered otioise . No wonder there is no currently active "psychology of
memory consolidation ." There is no longer anything to contribute to the
available causal account by working at that level. Thi s conclusion is not a
matter of " intuitions" about what con stitutes a better explanation. It is rather a
report of scientific prac tice regarding this example over the past thirty years.
Th is is not to say that current science eschews all mentalistic or
psychological causal explanations! (On this point my new wave metascience
of current cellular and molecular neuroscience differs from what philosophers
typically understand by "eliminativism.") First, many psychological-level
explanations retain their status as "causally-mechanistically explanatory" at
the present time because lower level cellular and molecular accounts are only
beginning to take formativ e shape. The latter are still "on promise," not yet
"on offer ." Impressive as it is, the "LTP is memory consolidation" accomplishment remains a parade case for current reductionistic neuroscience. And
even after cellular and molecular accounts develop beyond formative stages,
psychological causal explanations still play important heuristic roles in
generating and test ing neurobiological hypotheses. At the very least, they
suggest where to look for the key cellular and molecular mechanisms and how
best to construct behavioral paradigms that will generate useful tests of
hypothesized cellular and molecular mechanisms.' We saw an excellent
example of this second crucial heuri stic role in the detailed scientific example
of the previous chapter. Kandel and his colleagues constructed their crucial
behavioral test-the simultaneously acqui red one -trial declarative and
nondeclarative learning and memory tasks-based upon well-known
psychological investigations. These crucial heuristic roles of psychological
concepts and explanations would seem to imply that there is something
correct about them. Indeed. Their "correctness" is captured in the "rings and
strings" visual metaphor in Figure 2.11 above. They are approximations of the
best causal-mechanistic story we can tell at present.
There are also numerous contexts, both everyday and scientific,
where the explanatory power of the cellular/molecular neuroscientific causal
account isn't important and so doesn 't require elaboration. Contexts do
determine the answers we give and accept to "how" and "why" questions .
However, it is contrary to scientific practice to say that contexts set
parameters on what count as suffi cient causal explanations (as, e.g., Horgan
200 1 says). Rather, contexts determine whether a causal-mechanistic explanation , given the current state of scientific knowledge, is genuinely causally
mechanistic. In scientific contexts, typically only one is treated as such at any
given time, namely, the one genuinely available at the lowest level (and not
merely "under development" or "available in principle") . This is the point I
have been stressing by pointing out that there is no current "psychological
115
study of memory consolidation," and hasn't been since its cellular and molecular mechanisms have been discovered. In everyday contexts, on the other
hand, we are more lenient about the number and variety of explanations we
offer and accept. This is one way that everyday practices differ from scientific
ones.
At bottom, scientific practice in cellular and molecular neuroscience
refuses to grant the psychological any genuine causal explanatory role once a
real neurobiological successor is firmly in place. This lesson is apparent in
the scientific example from Chapter Two. As that example illustrates, in real
science the mental/psychological loses its status as genuinely causalmechanistic once we know how the lower-level mechanisms work.
Furthermore, the remaining purely heuristic roles for psychological causal
explanations hardly constitute a demand for a special class of theoristsphilosophers of mind-to construct a special domain of theories-theories of
mental causation-to serve any useful purpose.
Finally, scientific practice at least suggests a similar fate for
cellular/molecular causal mechanisms. Once biochemistry achieves causal
explanations of the tertiary folding configuration of the proteins involved, and
clever experimenters manipulate these proteins directly at the biochemical
level to produce specific behavioral effects , molecular neurobiological
explanations of the sort I presented above will also lose their (current) status
as causally explanatory. They will become purely heuristic. Similarly for the
future biochemistry vis-a-vis future thermodynamics or electrodynamics or
whatever. If worries are beginning to chum as you contemplate the full scope
of the reductionism I am urging, hold onto them. I promise to address them
later in this chapter, after they emerge again when we draw consequences
from our detailed scientific example for a second philosophical conundrum,
the multiple realization of the psychological on the neurobiological. Before
we tum to that, however, we have another "levels" question looming , this one
internal to neuroscience itself.
3 WHAT ABOUT COGNITIVE NEUROSCIENCE?

3.1 "Levels" questions within neuroscience
With this talk about levels of scientific investigation and the effect
that accomplished lower level explanations have on what at one time counted
as causal -mechanistic (now higher level) accounts, we can expect questions
about the relationship between cognitive neuroscience and cellular/molecular
neuroscience. Recall from Chapter One that cognitive neuroscience is the
interdisciplinary melding of psychology, neuropsychology, neuroanatomy,
116
linguistics, computer science , and artificial intelligence. It now uses such sexy
and powerful tools as "massively parallel" neural network modeling and
computer simulation, functional neuroimaging, and the mathem atics of
dynamical systems to interpret "global" activity patterns across neuron
populations. Unlike the "psychology of memory consolidation," cognitive
neuroscience is currently thriving. But without question its investigations are
pitched at a "higher level" than are those of cellular and molecular
neuroscience.
This leads to a very interesting issue. Familiar questions about the
relationship between psychology and neuroscience-in fact, the very ones we
saw at work in the mental causation debate-have direct analogs concerning
the levels within current neuroscience itself, and these questions are just as
stark here as they are across the psychology-neuroscience divide . Consider a
single example, concerning the "autonomy" of higher-level theorizing and
explanation. This has been a stock question in the philosophy of psychology
and cognitive science for three decades. Yet Stephen Kosslyn has recently
raised this challenge within neuros cience itself, writing: "Cognitive
neuroscience is a good illustration of how the whole can be more than the sum
of its parts" (1997 , 158). Although he admits that "cognitive neuroscience
must move closer to neurobiology," he still insists that " it will not simply
become neurobiology" becau se "cognitive neuroscience adds methods and
techniques to study, and conceptualize, how the brain gives rise to cognition
and behavior" (1997, 160). His "levels" relations within the brain sciences
themselves will sound familiar to philosophers of psychology. Indeed, they
are the very claims routinely made by autonomists about the psychological.
There are two broad approaches toward addressing levels questions.
One takes its departure from the philosophy of science. It explicates an
abstract intertheoretic relationship and applies it to questions about specific
scientific levels. This approach leads to familiar theories about and disputes
over relations like reduction, mechani sm, supervenience, emergence, realization, and instantiation. The other approach works from within empi rical
science itself. It seeks to employ the different experimental methods and data
analy sis techniques used in the levels whose relation is at issue. The results
are transdisciplinary research projects that addre ss some phenomenon using
resources from a variety of levels . The hope is that by seeing how these
resources work together and interact, we will come to see how the levels
relate from which they are drawn. I contend that this second approach is a
much more fruitful way to address levels questions. To defend this, I'll next
describe a transdisciplinary research proje ct that my group is pursuing,
indicate how we combine the methods and techniques from neuroscientific
levels ranging from single cell physiolog y, computational modeling and
computer simulation, and functional neuroimaging, show some promising
117
preliminary experimental results, and then close by drawing some lessons that
such transdisciplinary projects shed on "philosophical" disputes about
autonomy versus reduction across scientific levels.
3.2 Searching for the cellular mechanisms of the sequential

features of higher cognition
Neuropsychologists have known for decades that frontal cortical
regions subserve many "higher" cognitive processes (Kolb and Whishaw
1996, chap. 14). Examples include language production, complex motor
sequencing, planning, and problem solving. Higher cognitive processes are
cha racteristically sequential. They proceed from one idea or representation to
another in a way that seems to respect and exploit the contents of earlier steps
and upcoming target steps. To grasp my half-full beer glass on the cluttered
bar, behind the pretzel basket that my buddy 's arm is hovering above, I must
make a sequence of hand and arm movements in the proper order. To verbally
utter an understandable (not to say grammatical) English sentence, I must not
only get the right sounds spoken, but also in the right order. The question is,
how do we do this? How do neurons in frontal region s generate the sequential
aspects of higher cognition? What are the cellular mechanisms of these
characteristic features?"
To addre ss these questions scientifically, we followed the standard
practice of first searching for a model neural circuit. Three conditions seemed
obvious for a neural circuit to serve as a fruitful model in the search for the
cellular mechanisms of sequential cognitive proces sing.
I.
Its components need to be located in frontal cortex.
As mentioned above, many higher cognitive and conscious proce sses have
been known for some time to depend on intact frontal circuits .
2. It must generate sequential outputs.
Sequential features of higher cognition are our explanatory goal. Outputs of
the model circuit need not be "cognitive," but they must at least share the
sequential features of higher cognition. Finally ,
3. Much must be known about its cell -physiological and anatomical
details.
118
What is lacking, and very difficult to get in alert, behaving cognizers, is

knowledge of these details. A model system that meets all three of these
conditions could yield a cell -physiological mechanistic account of how its
sequentially structured outputs are generated that comfortably generalizes to
other frontal regions that subserve higher cognition and consciousness-or at
least it might suggest novel testable predictions about these.
The primate saccade command circuit meets all three conditions. Saccades are a type of rapid eye movement that continuously relocate the fovea
(the area of the retina with highest visual acuity) on different features of a
visual display. Human and nonhuman primates saccade 3-5 times per second.
Most saccades are executed involuntarily, but saccades can be controlled
voluntarily and consciously. The primate saccade command circuit involves
both frontal and posterior parietal cortical regions. The frontal regions receive
retinotopically coded visual inputs from the dorsal visual stream and project
efferents to midbrain, brain stem, and neostriatum. 10 The bilateral frontal eye
fields (FEF) of premotor cortex, encompassing primarily Brodmann's area 8a
in humans, are prominent in this circuit. This component of the circuit thus
meets condition (1) above . 1I
The role of PEFs in saccade commanding has been well studied in
nonhuman primates at the single-cell electrophysiological level (Bruce and
Goldberg 1985; Goldberg and Bruce 1990). Thus this circuit meets condition
(3) above on a fruitful model for investigating the cellular mechanisms of the
sequential features of cognitive processes. These studies show that FEFs
contain neurons with pre- and post-saccadic activity . Pre-saccadic neurons
have movement fields analogous to the receptive fields of sensory neurons.
They fire action potentials at maximal frequency prior to saccades of a
specific amplitude and direction, near-maximal frequency to saccades of
closely related dimensions, and diminish to baseline firing rate prior to
saccades of different dimensions. Post-saccadic neurons begin discharging
action potentials over baseline rate immediately after saccade onset. Their
activity also displays similar movement fields. More than 1/3 of the neurons
that Goldberg and Bruce (1990) found with pre-saccadic activity also had
post-saccadic activity. Their pre- and post-saccadic movement fields were
directly opposed. If one of these neurons was most active prior to a saccade of
a particular amplitude and direction (e.g., up and to the left), it was most
active after a saccade of exactly the opposite amplitude and direction (e.g.,
down and to the right) . Since many PEF neurons fire at different rates both
before and after a single saccade, presumably the FEFs send saccade
command dimensions to midbrain and brainstem circuits coded as a vector
average of all active neurons (the average of each active neuron's preferred
movement dimension times its spiking frequency).
119
Goldberg and Bruce's (1990) results using the two-step saccade

paradigm suggested that the PEFs compute at least two-step saccade
sequences "on the fly" or "ballistically," without the use of reflective
feedback. This computation makes use of the retinotopic dimensions of the
two stimuli from the initial origin encoded in pre-saccadic FEF activity and
the post-saccadic activity of the cells after the first saccade. Given FEF
neurons' pre- and post-saccadic movement fields, the primate FEFs appear to
compute the dimensions of the second saccade by implementing vector
subtraction in eye movement space (Figure 3.1).12 This sequential processing
in primate FEFs thus meets condition (2) on a fruitful model for the cellular
mechanisms of cognitive sequential processing, at least for two-step
sequences. Further evidence for the existence of multiple-step saccade
sequencing comes from human eye tracking studies. When a human face is
presented as a visual stimulus, normal adults respond with a characteristic
pattern of saccades that move systematically to the two eyes , the nose tip, and
the lips. This pattern is contrasted with that of human schizophrenics, whose
saccades to the same visual stimulus are scattershot, seemingly random, and
unsystematic. Schizophrenics lack the organized sequential features of normal
saccadic responses. For a recent study of this effect, see Loughland, Williams,
and Gordon (2002) .
What about the cellular mechanisms of saccade sequences involving
more than two steps? Single-cell electrophysiological work directly addressing this question does not exist, but work on two other frontal regions
suggests an intriguing possibility. For more than a decade, Patricia GoldmanRakic and her colleagues have explored activity in individual neurons in the
dorsolateral prefrontal cortex (DLPFC) of rhesus monkeys during delay
periods of spatial working memory tasks (Funahashi, Bruce, and GoldmanRakic 1989; Funahashi, Chafee, and Goldman-Rakic 1993).13 Their behavioral tasks involve saccade and anti-saccade delayed responses. Monkeys
fixate on a central spot on a computer screen and a light is flashed in the
periphery. They maintain fixation on the central spot for a short delay period
(from 1.5 up to 6 seconds). In the saccade task, monkeys must then saccade to
the remembered location of the peripheral stimulus. In the anti-saccade task,
they must saccade in the opposite direction away from the remembered
location of the peripheral stimulus. During the delay period, Goldman-Rakic
and her colleagues record single-cell activity in DLPFC neurons that appears
to hold information "on line" about the spatial location of the target. Results
with the anti-saccade task verify this interpretation. DLPFC neurons that are
active during the delay period of the saccade task to a specific stimulus
location are also active during the delay period of the anti-saccade task to the
same stimulus location, even though the monkey must make exactly the
opposite oculomotor response. This indicates that activity is tied to direction
120
A<3, 5>
-(FP~A)
<-3, -5>
A~B
<3, -7>
B <6, -2>
FP~B
<6, -2>
Figure 3.1. Vector subtraction in eye movement space. x-axis represents a saccade target's
degrees of hori zontal remove from fixation point (FP), y-axis represents degrees of vertical
remov e. A is the first saccade target, B is the second. In the double-step saccade paradigm
emplo yed by Goldb erg and Bruce, targets A and B have both appeared and extinguished before
the first saccade to A has begun (du ring the latency period of the first saccade in the sequence).
Thus the FEFs must comput e the dimensions of the seco nd saccade from the oculomotor
dimensions of the first saccade and the retinotopi c location of the second target. Vector
subtraction [(FP -7 B) - (FP -7 A)] comp utes the dimension s of A-7 B: <6, -2> - <3, 5> = <3,
-7> . Post-saccadic activity in FEF neuron s followin g the first saccade provides the dimensions
of - (FP -7 A) (i.e., (A -7 FP), pre-saccad ic activi ty in FEF neuron s with visual field properties
provides the dimensions of (FP -7 B). Their summed activity yield s the dimensions of A -7 B.
See Bickle et al. (2000 ), especially Figures 4 and 5, for full details .
and amplitude of impending saccade; otherwise the same neuron 's delay
period activity would differ significantly in the saccade and anti-saccade tasks
in response to the same stimulus location. Goldman-Rakic and her colleagues
label this physiological activity as the "working memory fields " of individual
DLPPC neurons.
More recently, Courtney et al. (1998) have used functional magnetic
resonance imaging (fMRl) to locate a spatial working memory area specific to
human frontal cortex. This area is in the superior frontal sulcus , directly
adjacent to the anatomical location of the human PEPs. They measured
activity increases in these regions durin g delay periods while humans
121
performed a spatial working memory task , but no increase when humans

performed an object identification wo rking memory task. As is the primate
DLPFC, synaptic connections between this region and FEFs are extensive.
An intriguing hypothesis combining these studies is that multiple step
(> 2) saccade sequences could also be computed via vector subtraction and
executed "ballistically," based on interactions between activ ity in FEFs and
these frontal spatial working memory areas (FWMs). Pre- and post-saccadic
activity in PEFs is sufficient to compute the dimensions of two-step saccade
sequences. But since post-saccadic activity coding for the opposite dimensions of the sacc ade just executed ceases as soon as the next saccade is
initiated, somehow FWM areas need to hold "on line" locations of upcoming
targets in the sequence and the post -saccadic dimensions of earlier saccades.
Do they? And how can this speculation about the cellular mechanisms of
multiple step saccade sequences be developed and explored experimentally?
3.3 Cognitive neuroscientific resources to the rescue:

Biological modeling and functional neuroimaging
Biological modeling (with computer simulation) is an ideal tool for
developing testable hypotheses like the one we sought. Neurobiologist Gary
Lynch and computer scientist Richard Granger nicely distinguish this type of
modeling from another type that is more prominent in cognitive science and
better known to philosophers:
Recently, the question of network properties inherent in
cortical design has been explored with two types of cortical
simulations. The first , which might be termed abstract network modeling, employs theoretical models that use a few
assumptions about neurons and their connectivities and then,
through computer simulations, seeks to determine if parti cular, often quite complex, behaviors emerge. The biological
postulates are usually simplified ... The second line of
modeling research, biological modeling, seeks to exploit the
rapidly growing body of information about the detailed
anatomy and physiology of simple cortical networks. ... In
biological models, design features are added to simulations
for biological reasons only, independent of their potential
computational attractiveness or complexity. (1989, 205-206) .
In this spirit of biological modeling, we constructed a model of pre- and post saccadic activity in FEFs and frontal working memory regions (Bickle et al.
122
2000, especially Figures 6 and 7). Our computational architecture and

parameters were derived directly from established cellular physiology from
single-cell recordings. Computations in the Vector Subtraction Core derived
from FEF single-cell pre- and post-saccadic physiological discoveries by
Goldberg and Bruce. Computations in the Working Memory Store derived
from single-cell working memory field phys iological discoveries by Goldman-Rakic and her colleagues. The model computes multiple-step saccade
sequences from an initial fixation point until space in the Working Memory
Store is exhausted (corresponding to time constraints of cellular activity in
FWM regions) (Figure 3.2).14
Here, however, we confronted a common problem with biological
modeling. To get realistic performance from our computer simulation (like
that illustrated in Figure 3.2), we had to make some computational
assumptions that could not be justified neurobiologically. A prominent
assumption of this sort is the order of activity in the model's different
components as the number of steps in a saccade sequence increases, e.g. , from
two to four. (We call this an increase in "saccade sequence burden.") Based
on Goldberg and Bruce's electrophysiological data, we assumed that saccade
command activity during two-step sequences is restricted to FEFs, which is
realized exclusively in the computations of our Vector Subtraction Core. But
as the number of saccades in a sequence increases, to three steps or more, our
model requi res activity in both the Vector Subtraction Core (derived from cell
physiological activity in primate FEFs) and the Working Memory Store
(derived from cell physiological activity in primate frontal working memory
regions). There is no physiological evidence that this is the order of regional
activation in the primate brain during saccade sequencing. Our model also
assumed a roughly monotonic increase in the level of FEF activity as saccade
sequence burden increases. Activity occurs in more FEF afferents during later
steps in a saccade sequence. In addition to new visual input from the dorsal
stream, FEFs also receive spatial and eye movement input from frontal
working memory neurons. Again , no direct physiological evidence supports
this assumption. But in the spirit of biological modeling, if we treat these
assumptions as novel hypoth eses about unknown frontal neural mechanisms
of saccade sequential processing, then fMRI using Blood Oxygenation LevelDependent (BOLD) contrast is an ideal technique for testing them
empirically. IS
One methodological advantage for functional neuroimaging of basing
a study on existing cell-physiological data and a neurocomputational model
derived from them is that one can simply adapt the behavioral task used by
the physiologists for humans in the magnet. One doesn't need to cook up a
new task from scratch. We did this with Bruce and Goldberg's (1985) doublestep saccade paradigm. Six healthy adult subjects executed four cycles of
123
A
<3, 10>
<9, -2>
~<12,8>
-B
-~~~
~.
<5, -4>
<-4, -8>
Figure 3.2. Results with a computer simulation of our neurocomputational model on a 4-step
saccade sequence. Vector subtraction core first computes the dimensions of a saccade from
origin to A: <0,0> occupies the previous step nodes (since this is the first saccade in the
sequence) reflecting post-saccadic activity in FEF neurons, <3,10> occupies the next target
nodes reflecting pre-saccadic activity in FEF neurons, so the next step nodes compute the
vector sum <0+3,0+ 10> = <3,10> and the simulation executes a saccade of those dimens ion.
For A-7B, <-3,-10> occupies the previous step nodes, <12,8> occupies the next target nodes ,
so the next step nodes compute the vector sum <-3+12,-10+8> = <9,-2> and the simulation
executes a saccade of those dimensions from A, landing on B (reflecting the activity of postsaccadic FEF neurons activated after the first saccade , in keeping with Goldberg and Bruce's
1990 discoveries). For B-7C , <-3,-10> occupies nodes in the first layer of the working memory
store reflecting working memory field activity in FWM regions, <-9,2> occupies the previous
step nodes, and <5,-4> occupies the next target nodes, so the next step nodes compute the
vector sum <-3+-9+5 ,-10+2+-4> = <-7,-12> and the simulation executes a saccade of those
dimensions from B, landing on C. For C-7D (dotted line), <-3,-10> occupies nodes in the
second layer of working memory , <-9,2> occupies nodes in the first layer, <7,12> occupies the
previous step nodes, and <-4,-8> occupies the next target nodes , so the next step nodes compute
the vector sum <-3+-9+7+-4, -10+2+12+-8> = <-9,-4> and the simulat ion executes a saccade of
those dimensions from C, landing on D. (The bold line from C back to the origin demonstrates
the role of the "Return to Origin" computation and the Significance Activation Mechanism,
derived from cell-physiological properties of ACC and "suppression site" FEF neurons , not
discussed in this book..) Reprinted from Bernstein el al. 2000, Figure 8, 149, with permission
from John Benjamin Publishing.
124
blocks of five 2-step , five 3-step, and five 4-step saccade sequencing trials,
and a baseline motor task (Figure 3.3) . Each target dot followed the previous
one by only 100 milliseconds. We chose these timing values to insure that all
stimuli were presented during the latency period of the first saccade in each
sequence, to require "ballistic" processing and to place a demand on working
memory capacities." Each trial began when a red dot appeared for 500
milliseconds in the center of a black background. This was followed immedi ately by a sequence of 2, 3, or 4 yellow dots at 100 millisecond intervals, each
at a randomly generated position in one of eight octants and one of two radial
distances from the central fixation dot. After the final target was presented, a
black screen appeared and remained until the duration of the trial. Each block
consisted of five trials, each trial lasting 6 seconds, for a block duration of 30
seconds, a cycle duration of 2 minutes , and a total task duration of 8 minutes
(plus an additional thirty seconds at the beginning so that subjects could
acclimate to the magnet). Subjects were instruc ted to fixate on the red dot as
soon as it appeared, and when the yellow dots appeared to saccade from one
target to the next in the correct order of their appearance. They were explicitly
instructed to move their eyes , not ju st imagine or think about where to move
them. As the number of targets in a sequence increased, the task put
increasing demands on subjects' working memory. Each subject reported
difficulty performing the 4-step task. We used a bilateral finger-tapping motor
task as a control condition. Thi s task provides known activation in the motor
strip, which we used as reference data specific to each subject.
We acquired activation data at 3 Tesla using BOLD-sensitized T2 *weighted, grad ient-echo EPI. Put in its simplest terms , the BOLD signal takes
advantage of the different magnetic properties of oxygenated versus
deoxygenated hemoglobin. The ratio of these values gives an indirect measure
of neuron (and glial cell) activity at a particular spat ial resolution or "voxel
size" (in this study, 3 x 3 mm), since increased cell activity requires additional
oxygen carried by hemoglobin for intracellular metabolic processes (Arthurs
and Boniface 2002). We acquired whole brain fMRI images in 24 slices every
1500 milliseconds for the entire task duration (8 minutes 30 seconds, so 340
fMRI data points). The first twenty images were disc arded, corresponding to
the first 30 seconds of fMRI data collection when a subject was acclimating to
the scanner. Immedi ately after acquiring activation data, we performed a
three-dimensional Modified Driven Equilibrium Fourier Tran sform whole
brain scan in an axial plane to provide high-resolution anatomical images .
This enabled us to co-regi ster activation maps of individual subjects and to
normalize structural images to Talairach spatial coordinates. This yields
spatial resolution of structural images in three dimensions to I x 1.5 x 1.5 mm
and provides excellent anatomical resolution and contrast between gray and
white matter (Holland et al. 2001).
FP - 500 IDS
A - 100 IDS
B -100 IDS
125
C -100 IDS
D - 100 IDS Execute saccades rest
Figure 3.3. Visual display during a 4-step saccade sequencing trial. See text for discussion.
Data post-processing was done using Cincinnati Childrens Hospital

Image Processing Software (CCHIPS ) (Schmithorst and Dardzinski 2000).
We performed cross-correlations for each subject on a voxel-by-voxel basis
between BOLD signal intensity time course and the reference function .
Voxels in which the BOLD signal was above a statistically appropriate
correlation threshold were overlaid on the subject's co-registered anatomical
image. The resulting statistic parametric maps were then transformed into
Talairach space for composite analysis, averaged across all subjects, and a
composite activation map based on the average correlation value was then
displayed for each comparison (e.g., 2-step versus 4-step, 3-step versus
baseline task, etc .).
To identify regions subserving saccade sequential processing, we
adopted two strategies. First we identified regions that were activated
significantly (p < 0.01) during any of the saccade sequencing periods (2-step,
3-step, or 4-step) compared with the control motor task. Second, we identified
voxels in the composite data where the amplitude of the BOLD-sensitized
fMRI signal was correlated significantly with the number of saccade steps per
trial (i.e., with increasing saccade sequence burden). These strategies clearly
126
identified the PEFs and the two regions of frontal cortex that had previously
been identified as (spatial) working memory area s, namely , a region in
DLPFC and another around the superior frontal sulcus . I ? Voxel-by-voxel
correlations between the BOLD signal and number of saccade steps yielded
composite activation maps. We transformed correlation values into a t-statistic and averaged across all subjects. Voxels that exceeded a significance of p
< 0.01 were overlaid across the averaged anatomical data set (Bickle et al.
2001, Figure 5).
We also plotted time courses for each subject for five Regions of
Interest (ROI s): FEFs , DLPFC, superior frontal sulcus, posterior parietal
cortex, and anterior cingulate cortex . For each ROI, we graphed normalized
BOLD signal values reflecting level of activation against tMRI data point in
the block design. Because the BOLD signal change s over time due to brain
blood flow that has nothing to do with the task , we "detrended" (also known
as "drift corrected") each time course to filter out this low frequency
component and remove any baseline drift. The procedure is to fit a quadratic
function to the data from each voxel and then subtract the linear and quadratic
components from the measured time course data." To account for factors like
hemodynamic and attention lag, we also threw out the first five data points
(frames) for each epoch. (All this is standard tMRI image analysis.) Using
this corrected data, we then computed composite activity for each ROI during
2-step , 3-step, 4-step, and baseline task performances (Figure 3.4).
The se results suggest numerous conclusions about the biological
plau sibility of our neurocomputational model of saccad e sequential processsing . First, our model is based on cell-physiological properties of regions that
are indeed active during saccade sequencing: the PEFs and two frontal
working memory areas ." Second , FEF activity does increase in monotonic
fashion with increasing saccade sequencing burden , from the 2-step task to
the 3-step and from the 3-step to the 4-step (Figu re 3.4A) . Finally, activity in
frontal working memory regions also increases with saccade sequencing
burden, and our data suggests a dissociation between DLPFC and superior
frontal sulcus activation. Activity increase in DLPFC is most prominent in the
shift from 2-step to 3-step saccade sequences, while activity increase in
superior frontal sulcus is most prominent in the shift from 3-step to 4-step
sequences.i" As noted above , given the speed of stimuli presentations, the 4step task places a heavy burden on working memory mechanisms. All these
results suggest that some purely computational assumptions of our neurocomputational model do have biological plausibility when treated as testable novel
hypotheses about the cellular mechanisms of saccade sequential processing in
frontal circuits.
Frontal Eye Fields
3 -Step
4-Step
Superior Frontal Sulcus
:~_
:f~rtEIj
2-Step
127
..
99.9
99 8
99.7
99.6
Baseline
'
2-Step
3-Step
4-Step
Baseline
Dorsolateral Prefrontal
Cortex
""~
10~~
99 .9
99.8
2-Step
3-Step
4-Step
Baseline
Figure 3.4. "Detrended" ("drift corrected") normalized composite mean BOLD signal intensity
in three regions of interest during 2-step, 3-step, and 4-step saccade sequencing and control
finger tapping trials. See text for discussion.
Data from this preliminary study also reveal an experimental design

flaw; but upon analysis, results from this flaw serendipitously help to verify
the biological plausibility of our model of saccade sequencing. Consider the
high level of FEF activity during the baseline motor task (Figure 3.4A above).
What accounts for this? During that finger-tapping task, subjects were staring
at a blank black screen. Guess what humans (and other primates) do when
they stare out into darkness? They saccade at roughly the normal rate, 3 times
per second. This explains the high level of FEF activity in control as
compared to experimental tasks. In the latter, subjects are forced to make only
a limited number of saccades, after which they return their eyes to the center
of the screen, anticipating the next saccade sequencing trial." But consider
also the lower level of activity in the frontal working memory regions during
the baseline task (Figure 3.4B, C). This suggests that although subjects were
saccading frequently during it, their saccades were not guided by working
memory information about either stimulus spatial location or oculomotor
dimensions of recently executed saccades. In other words, saccades during the
control task were random, one-step saccades, not organized sequences. So
saccade sequencing (beyond two steps) does seem to require interactions
128
between FEFs and frontal working memory region activ ity, just as our model
hypothesizes.
3.4 Philosophical lessons from transdisciplinary neuroscience

What lessons do this transdisciplinary cell -physiological, biological
modeling and neuroimaging study suggest for the "big" issue s I raised at the
beginning of this section about the relationship between cognitive and cellular
neuroscience? Bear in mind that figuring out the cellular mechanisms of
saccade sequential processing is not our ultimate explanatory goal. We chose
to model and explore the frontal saccade command circuit becau se it itself
was a model circuit for exploring the cellular mechanisms of sequential
cognitive processing. Expansion of the frontal lobes is widely believed to
underlie many of the higher cognitive abilities that distinguish humans from
other primates. However, the neuronal circuitry across much of the human
frontal lobe remains poorly understood at the cell-physiological level, due in
part to technical limits on studying human brain function and to the difficulty
of identifying and characterizing the higher cognitive functions. The usual
procedure in science, when confronting these kinds of difficulties, is to focus
on a simpler system that shares characteristic features of the processes in
question but whose components are more readily accessible experimentally.
The primate saccade command circuitry prov ides exactly this sort of model
system for the sequential properties of higher cognition (including our streams
of con scious experiences). Saccading is neither typically "cognitive" nor conscious (although it can be under cognitive and conscious control), but it is
sequential. By coming to understand the cell-physiological mechanisms by
which saccade sequences are computed and executed, we seek to uncover
testable hypotheses about the cellular mechani sms of the sequential aspects of
highe r cognition. One hypothesis resulting from our pilot study is that sequ ential processing (cognitive or not) is vector subtraction implemented neurally.
It is now commonplace in cognitive neuroscience to interpret activity acro ss a
population of neurons in terms of vector space representations and transformations (Churchland and Sejnowski 1992). Vector subtraction computes the
dimensions of future paths through the appropriate space, and is realized
neurally by the interactions between pre-movement, post-movement, and
working memory activity in the individual neurons compri sing the population
(Bickl e et al. 2000).
This judgment is strengthened by two features specific to our transdisciplinary study. First, cytoarchitecturally-with regard to cell types and
distribution, and both laminar and columnar structure- the frontal saccade
command regions are composed of "standard frontal cortex with a distinct
129
granule layer." They thus possess exactly the cellular resourc es as most of
frontal cortex (Parent 1996). These similarities suggest that the computational
strategies implemented in the cell properties and connectivities of the prima te
saccade sequencing system-vector subtraction with an interactive working
memory store-are available to many other frontal cortical regions known to
subserve sequential cognitive processing. Any cognitive process that can be
characterized mathematically as a pathway through a multi-dimensional
vector space could be implemented neurally via vector subtraction by the
appropriate set of pre-vector, post-vector, and working memory field neurons .
The scope of this way of "mathematizing" brain function in contemporary
cognitive neuroscience (to say nothing of its applications in "connectionist"
cognitive science more generally) suggests testable empirical hypotheses for
physiological investigation. If the PEFs and frontal working memory regions
implement saccade sequential processing in this fashion , the cytoarchitectural
similarities with the rest of frontal cortex suggest that the latter might
implement it in similar cellular mechanisms. Second, the continued biological
justification that the tMRI studies give to what were purely computational
assumptions of our model increases the plausibility of using the model to
generate testable hypotheses about the cell-physiological mechanisms of the
sequential features of other frontal cognitive and conscious processes. For a
given sequential cognitive process, representable mathematically as trajectories through a vector space, can single-cell investigations find neurons with
the appropriate pre-activity, post-activity, and work ing memory field s to
implement iterated vector subtraction (in the fashion that the saccade
command circuit appears to)? What single-cell neurophysiologist wouldn't
like such leads about neuron response properties to look for during cognitive
processing tasks?
To summarize the discussion so far: saccade commanding and execution are (typically) neither cognitive nor conscious, but their outputs are
sequentially organi zed. In addition, these components and circuits are well
understood at the single-cell physiological level, located in frontal cortex, and
have been characterized by a successful neurocomputational model, some of
whose purely computational assumptions have now been verified biologically.
(Our transdisciplinary research project demonstrates the last two points.) Like
any fruitful scientific model, ours suggests testable hypotheses for future
research toward uncovering the cellular mechanisms of the sequential aspects
of higher cognition and conscious experience . Can we be sure that our model
has focused upon the essentials of even this single feature of cognition ? Can
we be sure that we aren 't being misled by the cellular mechanisms of a
simpler system producing sequential outputs? Can we be sure that other
strategies for uncovering the cellular mechanisms of higher cognition won't
be more succe ssful (e.g. , Bechtel and Richardson 's 1993 "decomposition and
130
localization" strategy or Craver and Darden' s 2001 strategy for "discovering

mechanisms")? Our answers are no, no, and no. But these negative answers
are for the unexciting reason that we are doing science, where no algorithm
for successful discovery exists.22
What about cognitive to cellular neuroscience reduction versus the
"autonomy" of higher level methodologies, explanations, and theories? Transdisciplinary projects employing methods and results from a variety of levels
(like ours) addres s this issue empirically. We use neurocomputational
modeling and functional neuroimaging to answer questions that it is difficult
to imagine addressing solely at the single-cell physiology level. If that is all
that "autonomy of the higher level" amounts to, then we treat neurocomputational modeling functional neuroimaging as "methodologically
autonomous" from cell physiology. But that typically isn't all that proponents
of "higher level autonomy" want, at least among philosophers. They want
autonomy of theory or explanation; they claim that mechanisms uncovered by
higher level investigations are "independent" of the lower-level details. That
sort of autonomy is clearly no part of transdisciplinary science-in-practice, at
least of the sort we pursue. We use the higher level methodologies, resources,
and techniques as a way to get to the cellular mechanisms of cognitive
processing. And I contend that our approach is characteristic of most real
transdisciplinary neuroscience, as opposed to philosophers' and cognitive
scienti sts ' fantasies about how it proceeds. For us, higher level theories and
methods have a useful and seemingly ineliminable heuristic role to play in the
search for lower level mechani sms and reductions. They greatly increase the
desc riptive base for lower level experiment and theorizing. They tell us where
in the brain to look and they guide us in constructing behavioral tasks that
isolate the crucial dependent and independent variables. But that is all they
do, and all they can do. When they've exhausted this descriptive and
methodological function they fall away, much like Wittgensteins ladder. r'
Higher level accounts reduce to lower level mechanisms, after the former
function to help us discover the latter. If you think that they do more , then
your intuition is no part of transdisciplinary research in mainstream neuroscience and you really are a "levels" dualist, whether or not you admit to it.
Features "of the whole," "of the system," "of the population," are discovered
to be nothing but complex sequences, combinations, and interactions that
occur between the individual components, specifiable ultimately in lower
level terms and interactions. What else could they be ?
Nothing prevents us from resolving more general levels questions
across cognitive science and neuroscience in the same way. Cognitive
scientific investigations, methodologies, theories, and explanations are
essential heuristics in the search for lower level neuronal mechanisms. They
generate data, descriptions, and behavioral tasks as part of transdisciplinary
131
research projects that it is difficult to imagine gaining from purely cellular and
molecular investigations. This much "methodological autonomy" is an ineliminable part of standard scientific practice. But that is all the "autonomy" that
transdisciplinary scientific practices warrant, and that much is consistent with
ruthless reductionism.
4 PUTNAM'S CHALLENGE AND THE MULTIPLE

REALIZATION ORTHODOXY
"It is remarkable that the cAMP signal transduction pathway, including its
nuclear components, seems to be required for memory-related functions in
each of these species and behavioral tasks" (Yin et al. 1994, p. 55).
I mentioned in Chapter One (section 4.1) that one of the most
influential challenges to psychoneural reduction in the philosophical literature
rests upon multiple realization. There we saw ways that reductionists have
tried to undermine anti-reductionist conclusions urged from it. Mainly,
reductionists have granted the multiple realization premise and argued that
anti-reductionist conclusions drawn from it are invalid. Few have challenged
the truth of multiple realization. It is time to see why, and in light of the
detailed example from the previous chapter and additional scientific details
hinted at in the quote that prefaces this section, to construct this more direct
assault. I am convinced that this is the definitive reply to one of the most
influential arguments in late-20 th century philosophy.
Multiple realization in philosophy of mind begins explicitly with
Hilary Putnam. In defense of the "functionalist" view of mind that he developed and championed throughout the 1960s, Putnam laid down a challenge to
his chief competitors: the early "central state materialists" who advocated
identifying mental with neural types (properties, states, events). Using "pain"
as his paradigmatic mental type, Putnam writes:
Consider what the brain -state theorist has to do to make good
his claims. He has to specify a physical -chemical state such
that any organism (not just a mammal) is in pain if and only if
(a) it possesses a brain of a suitable physical-chemical structure; and (b) its brain is in that physical-chemical state. This
means that the physical-chemical state in question must be a
possible state of a mammal 's brain, a reptilian brain, a mollusk's brain (octopuses are mollusca, and certainly feel pain) ,
etc. (1967 , p. 45)
132
Since the brain -state theori st makes the same claim about every mental state ,
his liability is even greater. Putnam continues: "If we can find even one
psychological predicate which can clearly be applied to both a mammal and
an octopus (say, "hungry" ), but whose physical-chemical correlate is different
in the two cases, the brain state theory has collapsed" (ibid.) . It is
"overwhelmingly likely," Putnam asserts, that we can find such a statemany , actually.
Thus multiple realization entered into the philosophy of mind. The
premise asserts that a given psychological kind (property, state , event) is
realized by distinct physical kinds . Providing a precise definition of 'realization ' has proved difficult. Must it be a necessary truth that the realizing
state, property, or event obtain only if the realized state, property or event
obtain, or is contingency enough? If necessity is required, which strength of
neces sity is sufficient (physical, metaphysical, logical)? Despite these
controversies, multiple realization has achieved consensus as a true and
crucial premise in an argument against mind-brain identity theory and psychophysical reduction. This status has remained even as funct ionalism, the view
that spawned the argument, has given way to nonreductive physicalism. One
important reason for its staying power, given the concerns in this book , is that
neuroscience itself seems to provide examples of creatures whose behavior is
describable using the same psychological concepts but whose nervous systems are very different. The realizing neural states across the creatures would
not be identical. Hence multiple realization seems verified empirically.
In the remainder of this chapter, I aim to challenge the truth of
multiple realization as it applie s to creatures here on earth . From the perspective of behavioral and systems neuroscience, multiple physical realizations of
shared psychological kinds seem obviously to obtain . But as neurosci ence's
core over the past two decades has shifted , increasingly to the level of molecular manipul ations and investigations, multiple realization at the systems level
gives way to evolutionarily conserved, shared mechanisms across otherwise
vastly different species . The molecular mechanisms determining neuron
activity and plasticity are the same in invertebrates through mammals. In light
of these discoveries and the direct ties that have been forged to behavioral
effects, the emerging " links" between molecules and mind cast doubt upon
multiple realization. I will flesh out this argument by supplementing the
scientific details from the previous chapter with additional ones from studie s
on invertebrates, in particul ar on fruit flies and sea slugs. These recent
discoveries answer Putnam' s challenge. As "overwhelmingly unlikely" as it
might seem from the philosopher's armchair (and the systems neuroscientist's
laboratory bench), cellular and molecular neuroscientists are discovering
"physical-chemical state s" that serve as shared mechanisms for shared
psychological events across biological phyla. And principles of molecular
133
evolution suggest that more of these will be discovered as these scrences

progress. Multiple realization, meet molecular neuroscience.
Some observations and caveats are immediately in order. First,
Putnam's original challenge was an empirical scient ific challenge. He claimed
that (circa the empirical data and theories of 1967) unitary physical-chemical
states were unlikely to be found in the variety of earthly organisms for which
it seems reasonable to assume identical psychological events. To address his
challenge, one must show that science is finding such unitary physicalchemical states. One can only show this by presenting scientific details.
Philosophers who reject the relevance of empirical science will not be
impressed. But they've probably long since put down this book, and they
should equally not be impressed by Putnam's challenge and its antireductionist conclusion. Putnam's challenge rests directly on intuitions about
scientific facts and what empirical investigations will or will not discover ."
Unfortunately, over the three and one-half decades since Putnam first
issued his challenge, its empirical grounding has fallen out of focus . Antireductioist philosophers nowadays typically speak of multiple realizability
and lace their discussion with "thought experiments" involving fantasized
cognizers like silicon-based aliens and artificially intelligent electronic robots.
This is because most philosophers assume that identity holds across "all
possible worlds," either necessarily or as a matter of scientific law. Putnam
(1967) himself strapped the "brain-state theorist" with this additional burden,
writing that any proposed physical-chemical state must also be "a state of the
brain of any extraterrestrial life that might be found that will be capable of
feeling pain" (1967, p. 45). Jerry Fodor (1974) exploited this "nomological"
assumption in his important elaboration and extension of Putnam's original
argument. This explains the popular "thought experiments" in the multiple
realizability literature of beings that share our psychological kinds but lack
our organic brains at any level of physical description.
This broader sense of multiple realizability and philosophers' "possible world" fancies do not concern me . I don't know whether identity holds
across "all possible worlds," or even across all "physically possible worlds." I
don't know the "conceptual" or "nomological limits" of our psychological
concepts. But I take comfort in the fact that you don't, either, regardless of the
strength of your intuitions. Jerry Fodor once remarked that
self-confident essentialism is philosophically fashionable this
week. There are people around who have Very Strong Views
('modal intuitions,' these views are called) about whether
there could be cats in a world in which all the domestic
felines are Martian robots, and whether there could be Homer
in a world in which nobody wrote the Odyssey or the Iliad.
134

Ducky for them; their epistemic condition is enviable, but I
don't myself aspire to it. (1987, 16)
I concur completely and apply this attitude to the "possible world" scenarios
that transform multiple realization from an interesting empirical problem into
multiple realizability, a "conceptual puzzle" about "the scope of our psychological concepts." As you might have already gathered from my neoCarnapian outlook sketched in Chapter One , I steer clear of pragmatically
fruitless questions. I'll worry about brainless yet pained or belief-entertaining
aliens and robots as soon as one crosses my path. My concern is with existing
earthly creatures. If the scope of my concern is too narrow for your
philosophical sentiments, so be it. Scientists don't give a hoot for
philosophers' Very Strong Modal Intuitions about kind identity across
possible worlds, and their enterprises are doing just fine .
I'll make one more remark on this point and then move on. Since the
heyday of "central state material ism" in the 1950s and early 1960s, "brainstate" theorists have emphasized the contingent-read: this worldly-nature
of their identity or reductionist claims . No serious "brain-state" theorist over
the past forty years has claimed that neural expressions will provide
synonyms for psychological expressions, any more than the "lightning is
atmospheric electron discharge" hypothesis claims to. It might be fashionable
in post-Kripke philosophy to insist that all identity claims hold necessarily,
but fortunately scientists don't bother reading Kripke and keep right on
making and testing identity claims that purport to hold in the real world.
Despite this unfortunate detour into pragmatically fruitless metaphysics that
the multiple realization issue took, however, there still remains within it a
genuine and unanswered empirical challenge. Answering this challenge is my
goal.
The next point to note at the outset is that no paradox is looming in
my project. That molecular neuroscientists are finding shared mechanisms
that underlie shared psychological kinds across species does not imply that all
species possess similar psychological capabilities. That implication would be
a reductio ad absurdum for the ruthless reductionism of current mainstream
neuroscience. Any account that leaves out the psychological differences
between, say, humans and sea slugs, will be seriously incomplete as an
account of cognition and behavior! Fortunately, nothing is implied or
suggested in the account about to be presented about vast psychological
differen ces across species. The multiple realization challenge only speaks to
shared psychological kinds. Obviously, psychological differences must have
distinct mechanisms of some sort.
Current cellular and molecular neuroscience does have a going story
about these differences. This story has been characterized by Richard
135
Hawkins and Eric Kandel as a "cell-biological alphabet" out of which different "words" and "sentences" are constructed to explain different psychological capacities. f They write :
Do the [intracellular] mechanisms so far encountered form
the beginning of an elementary cellular alphabet? That is, can
these units be combined to yield progressively more complex
learning processes? We would like to suggest on theoretical
grounds that such an alphabet exists and that certain higherorder forms of learning generally associated with cognition
can be explained in cellular-connectionistic terms by combinations of a few relatively simple types of neuronal processes.
(1984a, 386)
They draw their cell-biological letters out of earlier research on the sea slug ,
Aplysia califomica, but argue for their applicability to cognitive learning in
mammals:
We propose that higher forms of learning may utilize the
mechanisms of lower forms of learning as a general rule ; and
second, we speculate that this may occur because higher
forms of learning have evolved from lower forms of learning.
... Thus, whereas individual neuron s may possess only a few
fundamental types of plasticity that are utilized in all forms of
learning, combing the neurons in large numbers with specific
synaptic connections (as occurs, e.g., in mammalian cortex)
may produce the much more subtle and varied processes
required for more advanced types of learning. (l984b, 391)
However, I mention this approach merely to show that no paradox is looming
in my reply to multiple realization. Given that its challenge focuses on shared
psychological kinds , a discussion of multiple realization is not the place to air
complaints about this approach toward explaining psychological differences.
Finally, it is impossible (obviously) to establish the unique realization
of each type of psychological state across all the specie s that possess it. At
present, we don't know enough about the underlying neural mechanisms for
many types, especially the molecular mechani sms where I am claiming that
the unique realizations lie. Still , if we can find one prominent shared
psychological kind that appears to be realized differently in the nervous
systems of different species possessing it, but which turns out actually to be
uniquely realized by shared molecular mechanisms, this single empirical
example would bolster a general hypothesis. This argument will be even
136
stronger if that example illustrates a principle, the evolutionary conservation

of molecular mechanisms, that holds generally, on independent grounds. I
contend that such an example already exists in current molecular accounts of
memory consolidation .
We are about to embark again into detailed molecular neuroscience.
The sledding will be no easier here than it was in Chapter Two, although
readers quickly will be struck with a feeling of deja vu. ("Haven't we heard of
these molecules, signaling pathways, and effects on long-term memory
behavior before?") But then, did you really expect a challenge with the
staying power of multiple realization to yield to a simple empirical reply?
5 MOLECULAR MECHANISMS OF NONDECLARATIVE

MEMORY CONSOLIDATION IN INVERTEBRATES
5.1 Single-gene fly mutants for associative learning
In the mid-1960s, Seymour Benzer introduced a technique for
generating chemical mutations of single genes. This work drew upon the
beginnings of biotechnology that has since mushroomed into the techniques
used today to manipulate the mammalian genome (discussed in Chapter Two,
sections 4.2 and 5.2 above) The fruit fly, Drosophila melanogaster, proved to
be an excellent experimental preparation. Benzer's group developed an olfactory shock-avoidance conditioning procedure for Drosophila (Quinn et al.
1974) and soon produced the first single-gene learning and memory mutant,
dunce (Dudai and Quinn 1976). Quinn continued this mutagenesis approach
with shock-avoidance procedures and produced four other Droso-phila
learning and memory mutants. Other training and testing procedures for
Drosophila were developed, including several operant procedures (Connolly
and Tully 1997). At present, more than twenty fly learning and memory
mutants have been identified (Dubnau and Tully 1998).
A common conditioning procedure begins by trapping a group of flies
in a chamber. Training involves exposing them to odor A (the CS+) paired
with a shock (US) through the chamber's landing apparatus. This is followed
by exposure to odor B (the CS-) that is not paired with a shock. "Spaced"
training intersperses training sessions with rest periods. "Massed" training
presents sessions consecutively without rest intervals. Trained flies are then
transferred to the choice point in aT-maze where they "choose" between odor
A (CS+) in one arm and odor B (CS-) in the other, presented simultaneously.
Greater than 90% of wild-type (non-mutated) flies avoid the CS+, while less
than 5% avoid the CS- (Tully and Quinn 1985). Spaced training with 10
137
training sessions produces long-term memory for the CS+-US pairing in flies
that is measurable for more than seven days . This conditioning assay has been
used to "dissect" the specific learning and memory deficits of induced
Drosophila mutants , e.g., by varying the nature of the training sessions and
the time between training and T maze choice test.
Biochemical analysis of dunc e and rutabaga mutants suggests that the
intracellular cyclic adenosine monophosphate (cAMP) "second messenger"
pathway is crucial for Drosophila olfactory learning and memory (Levin et al.
1992). In a series of recent experiments, Tully 's group has elaborated the role
this pathway plays in Drosophila memory consolidation. They cloned a
Drosophila gene , dCREB2, which transcribes a number of related protein
products (Yin et al. 1995b). One, dCREB2-a, is a protein kinase A (PKA)responsive transcriptional activator. Another, dCREB2-b, is a repressor of
PKA-responsive transcriptional activation. Readers of the previous chapter
are familiar with the role of the related mammalian genes and their protein
products in postsynaptic E-LTP, L-LTP, and declarative long-term memory
consolidation.
Using germline transformational techniques standard in invertebrate
molecular biology for two decades, Tully's group generated Drosophila
mutants that overexpress dCREB2-b under the control of a heat shock
promoter (hs-dCREB2-b) (Yin et al. 1994). This promoter is only turned on to
transcribe the transgene's protein product after mutant adult flies are exposed
to heat shock . Experimental mutants get exposed to heat shock; control
mutants do not. This is a standard practice to control for possible
developmental effects of the transgene's insertion that might affect behavior
but are not specific to learning and memory. When expressed, the mutant
tran sgene guides the production of an overabundance of dCREB2-b, the CRE
repressor. This blocks the expression of various genes induced by the cAMPPKA cascade. Using the Tully and Quinn (1985) olfactory association
procedure (described above), Yin et al. (1994) showed that inducing the hsdCREB2-b transgene disrupt ed long-term memory in mutant flies after spaced
training. Induced hs-dCREB2-b mutants displayed significantly poorer
retention of the CS+-US association when tested at both 24 hours and seven
days later, compared to both wild-type flies and hs-dCREB2-b mutan t controls
(flies with the transgene inserted that were not exposed to heat shock, and so
in which the transgene was not expressed). Despite this long-term memory
deficit, however, induced hs-dCREB2-b mutants displayed normal initial
learning, olfactory acuity and shock reactivity (all assayed using standard
techniques) . On a massed training regime that produces 3-hour but not 24hour retention in wild-type flies, induced hs-dCREB2-b mutants performed
similarly to both wild-type and non-induced hs-dCREB2-b mutant controls.
Clearly , overexpressing the hs-dCREB2-b transgene produces an effect
138
specific to memory rather than perception and selectively disrupts

consolidation of long-term olfactory conditioning.
In a subsequent study, Tully's group inserted and expressed another
heat-shock promoted transgene, hs-dCREB2-a. This mutation overexpresses
dCREB2-a, the activator isoform of the protein . In wild-type and non-induced
(non-heat shocked) hs-dCREB2-a mutant controls, optimal long-term memory
(tested seven days after training) in the Tully and Quinn T maze requires 10
spaced training sessions. Induced hs-dCREB2-a mutants, however, showed
comparable long-term retention after only a single training session. Again,
using standard assays, olfactory acuity and shock reactivity were normal in
the induced hs-dCREB2-a mutants compared to wild-type and non-induced
hs-dCREB2-a mutant controls. This result indicates that this transgene's
overexpression also produces specifically a memory effect. Finally, long-term
memory formation was not enhanced (compared to controls) in transgenic
flies expressing a mutant CREB isoform that is resistant to phosphorylation
by PKA. Based on their combined results from these studies, Yin et al.
conclude that "opposing functions of CREB activators and repressors act as a
"molecular switch" ... required to form [long-term memories]" in Drosophila
olfactory conditioning (l995a, p. 110).
Figure 3.5 illustrates the current model of memory consolidation in
Drosophila. Neurons in the mushroom body receive sensory inputs from both
the olfactory and the foot shock sensory pathways. The latter come in through
a modulatory neuron releasing a catecholamine as its neurotransmitter
(dopamine, DA, or serotonin, 5-HT). The CS+ pathway activated by odor A
generates action potentials in specific mushroom body neurons and a
subsequent rise in calcium ion influx (Ca 2+). (This is a standard neuronal
effect I described in Chapter 2, section 3.2.) Neurotransmitter release by the
presynaptic olfactory neuron generates action potentials in the postsynaptic
mushroom body neuron. As the action potential traverses the length of the
mushroom body neuron 's axon (and backpropagates to the soma and
dendrites), it opens voltage-gated Ca 2+ channels and Ca 2+ flows into the cell
through these opened gates by the forces of diffusion and electrostatic
pressure. Intracellular Ca 2+ interacts with calmodulin (CaM), producing
increased amounts of Ca 2+-CaM complex. Adenylyl cyclase molecules have a
Ca 2+-CaM complex binding site; binding at these sites primes these molecules
for subsequent intracellular interactions. In the paired CS+-US condition,
catecholinergic release by the modulatory interneuron in the US pathway
occurs during this rise in intracellular Ca 2+, Ca 2+-CaM complex, and adenylyl
cyclase priming. The DA or 5-HT receptor on the mushroom body neuron is
metabotropic, coupled to a G protein complex that also activates adenylyl
cyclase. The combined results of CS+-US pairing makes adenylyl cyclase
molecules in specific mushroom body neurons more efficient at converting
139
adenosine triphosphate (ATP) into cAMP. Thus in these specific neurons,

more cAMP molecules are available to bind to sites on the regulatory subunits
of more cAMP-dependent protein kinase A (PKA) molecules. cAMP binding
frees up more PKA catalytic subunits. In the short term, more freed PKA
catalytic subunits phosphorylate more potassium (K+) channels in the neuron
membrane, closing these channels and preventing K+ efflux. The result is
broadened action potentials, with subsequent increases in the amount of
neurotransmitter released by specific affected neurons . PKA catalytic subunits
are also thought to facilitate neurotransmitter release by direct actions on
release machinery in presynaptic active zones (by a mechanism still being
investigated) . This increase in neurotransmitter release strengthens responses
in other specific neurons further downstream in the motor output pathway,
leading to odor avoidance (behavior). Mushroom body neurons in the CS
pathway, which aren't followed by the US and subsequent DA or 5-HT
binding, lack the combined step that renders adenylyl cyclase molecules more
efficient at ATP-cAMP conversion. Events further down stream involving
these neurons, and ultimately behavioral effects (odor avoidance), are reduced
substantially. All this molecular neuro science should be familiar, since these
are the molecular mechanisms of mammalian E-LTP, only here transposed
into a presynaptic key. (Compare Figures 2.4, 2.6 and 3.5.)
Familiarity deepens when we consider the effects of repeated CS+-US
pairing s over the appropriate time period (as occurs, e.g., in Tully and
Quinn's 1985 multiple session "spaced" training). The repeated paired stimuli
continually increase the concentration of freed PKA catalytic subunits in
presynaptic terminals of specific mushroom body neurons. Eventually this
reaches levels high enough for catalytic PKA subunits to translocate to the
neuron's nucleus. There these molecules phosphorylate the CREB isoforms.
During the rest intervals of spaced training, the net function of phosphorylated
CREB activators comes to exceed that of CREB repressors, initiating a
cascade of immediate early gene expression. Phosphorylated dCREB2-a
molecules bind to cAMP response elements (CRE sites) on these genes'
control regions, initiating transcription. This initiates translation of new
proteins that yield long-lasting changes to structure and function of
presynaptic terminals of specific mushroom body neurons, namely , increased
amounts of neurotransmitter released by individual action potentials and more
active zones (release sites). The result is increased efficacy at specific
mushroom body-follower neuron synapse s in the motor output/odor avoidance pathway. Ultimately this increased efficacy produces changes in motor
output. Since this plasticity in neurotransmitter release rates is driven by new
gene expression and protein synthe sis, it is much more long-lasting than the
transient, short-term changes that are localized to synap ses and independent of
new gene expre ssion. Quite literally , the procedures that "consolidate" long-
140
dCREB-repressor
dCREB-aclivator
CREB
~ Catalytic
subunit
CalCaM
Regulatory
subunit
rutabaga
amnesiac
Figure 3.5. Molecular mechanisms of synaptic facilitation in Drosophila. Presynaptic

facilitation via these molecular pathways leads to enhanced excitatory
neurotransmitter release onto postsynaptic follower neuron , and hence increased
postsynaptic activity to afferent stim-uli. Reprinted from Trends ill Genetics, 15,
"Genetic approaches to memory storage," 463-470, Copyright 1999, with permission
from Elsevier Science.
term memory in Drosophila generate permanently altered neurons in terms of

their fundamental protein make-up. Previously dormant genes get expressed
and new protein products result. The key step is the availability of PKA
catalytic subunits in numbers sufficient to translocate to the cell nucleus . As
we are beginning to see, this is a common theme across biological classes and
types of memory.
141
5.2 Consolidating nondeclarative memory in the sea slug,

Aplysia
Work with induced Drosophila single gene mutants was a crucial first
step toward discovering both the molecular-genetic and biochemical aspects
of learning and memory. But measuring changes in fruit fly neurons and
circuit organization presents daunting technical challenges. With its relatively
simple nervous system, readily accessible neurons, and more easily measured
behavioral repertoire, the marine invertebrate Aplysia californica (a sea slug)
has also been a central experimental preparation in cellular and molecular
studies of learning and memory for one-quarter century. Indeed, some of the
mechanisms of the current Drosophila model were first discovered experimentally in Aplysia. Aplysia research has also yielded increased knowledge
about many molecular details first discovered in Dro sophila.
For example, it is now clear from Aplysia research that the cellular
and molecular mechanisms of classical conditioning, both short-term and
long-term, are elaborations on those of sensitization, a simpler form of
learning and memory. Behaviorally, sensitization is a heightened responsiveness in an organism's entire range of defensive reactions following a
noxious stimulus. In Aplysia, the gill withdrawal reflex following a tail shock
has been a common laboratory preparation (Frost et al. 1985). In normal
contexts, a weak tactile stimulus to Aplysia siphon (e.g., touching the fleshy
spout lightly with the tip of a paintbrush) produces a moderate gill withdrawal
that quickly habituates with repeated stimuli (moderate in terms of the speed
of the gill withdrawal into the mantle cavity and the amount of gill
withdrawn). Following a single electric shock to the tail, however, Aplysia
withdraw their gill more quickly and completely into the cavity to the weak
tactile stimulus. This effect persists for minutes after a single shock and can
be ' extended with additional shocks (making this preparation another
laboratory analog of stimulus repetition). Five or more shocks presented in
spaced training yields sensitization of gill withdrawal that persists for two or
more days, up to two weeks (Bailey and Chen 1989).
In terms of its cellular organization, this Aplysia sensitization circuit
begins at sensory neurons innervating the siphon and others innervating the
tail. Siphon sensory neurons synapse directly on gill motor neurons and on
interneurons that also synapse on gill motor neurons. Tail sensory neurons
synapse on a special class of modulatory interneurons that in turn synapse on
siphon sensory and gill motor neuron soma (cell bodies) and presynaptic
terminals. Sensitization starts with activity in these modulatory interneurons
driven by the tail shock, especially on the presynaptic terminals of the siphon
sensory-gill motor synapses. This activity produces increased neurotransmitter release by the siphon sensory presynaptic terminals upon subsequent
142
Nucleus
CRE
CRE1I-2
Long term
CRElI-Ie
Ubiqui lin
hydrola""
ClEBP
Ublquitin
protMsome
Shat term
Figure 3.6. Mo lecul ar mechani sms of syna ptic facilitat ion in Aplysia (sea hare) . Molecul ar
genetic effects produ cing long -term facilitation in presynap tic neurotransmitter release machinery were first disco vered in this experimental preparation, but have since been found in
Drosophila. See text for discussion . Reprinted from Trend s in Genetics, 15, "Genetics
approaches to memory storage," 463 -470 , Cop yright 1999, with permi ssion form Elsevier
Science.
143
siphon stimulation, and in turn heightened gill motor neuron activity and
behavioral response (gill withdrawal). The intracellular molecular mechanisms of sensitization are exactly the ones involved in Drosophila olfactory
conditioning (Figure 3.6; compare with Figure 3.5).
A nice methodological feature of the Aplysia monosynaptic siphon
sensory-gill motor circuit is that its components can be extracted and reconstituted in vitro in dissociated cell culture. The modulatory interneurons
driven by the (prior) tail shock release a catecholamine, serotonin (5-HT), as
their neurontransmitter. Montarolo et al. (1986) showed that short-term facilitation in this monosynaptic preparation could be induced with a single puff
(lmicromole, or uM) of 5-HT directly onto the presynaptic terminal, and that
long-term facilitation measurable for more than 24 hours could be induced by
five puffs of 5-HT delivered over one -and-one-half hours. In both cases,
facilitation was measured experimentally as a percentage increase in excitatory postsynaptic potential (EPSP) in the gill motor neuron. The 5-HT binds
to metabotropic receptors in the siphon sensory presynaptic terminals,
activating an intracellular G protein complex and priming adenylyl cyclase to
generate more cAMP from ATP. The additional cAMP binds to the regulatory
subunits of PKA molecules, freeing the catalytic subunits. In the short term
(i.e., following a single puff of 5-HT), the catalytic PKA subunits phosphorylate K+ channels in the presynaptic terminals and act directly at active zones.
When the weak siphon stimulus occurs subsequent to these tail shock-induced
short-term changes, the resulting action potentials in the siphon sensory
neurons are broadened.. More neurotransmitter is released, more binds to
receptors on gill motor neurons, action potential rates increase in gill motor
neurons, and the gill withdrawal response is dramatically stronger and quicker
than under conditions of no sensitizing stimulus (Figure 3.6). Once again , all
these molecular mechanisms are familiar.
Short-term classical conditioning in Aplysia is an elaboration of these
same molecular mechanisms. The principal difference between sensitization
and classical conditioning is the order of stimulus presentations. In sensitization the aversive stimulus comes first, followed by the neutral stimulus. In
classical conditioning this order and timing is exactly reversed. A common
assay for Aplysia conditioning also pairs a tactile siphon CS with a tail shock
US. Optimal conditioning produces an even higher level of excitatory neurotransmitter release by siphon sensory neurons than sensitization produces. In
the short term, this stimulus-specific increase results from the influx of Ca 2+
into the siphon sensory presynaptic terminals by the action potentials generated by the prior siphon stimulus . The rest of the molecular story will be
familiar. Ca 2+ binds with intracellular calmodulin (CaM) to produce increased
levels of Ca 2+ -CaM complex. Additional Ca 2+ -CaM molecules bind to
additional adenylyl cyclase molecules, priming their efficiency for converting
144
ATP into cAMP. When the 5-HT released by the modulatory interneurons
(driven by the now subsequent tail shock) binds to metabotropic receptors in
the siphon sensory presynaptic terminals and activates the intracellular G
protein complex, the primed adenylyl cyclase molecules convert far more
ATP into cAMP. Increased cAMP yields increased free PKA catalytic
subunits, which phosphorylate more K+ channels in the siphon sensory
presynaptic terminals and enhance neurotransmitter release. Increased neurotransmitter release upon subsequent CS presentations yields greater gill motor
neuron response. And greater gill motor neuron activity generates a quicker
and stronger gill withdrawal, even quicker and stronger than that produced by
sensitization due to the primed adenylyl cyclase molecules generated by the
now prior siphon stimulus (Figure 3.6).
But it is in the study of the molecular basis of the consolidation
switch that the methodological advantages of the Aplysia experimental preparation really payoff. It is not only that the same mechanisms discovered in
Drosophila olfactory conditioning have been verified directly and elaborated
further in these more accessible neurons and circuits. Also, it is here that
discoveries in invertebrates connect up closest with the molecular
mechanisms of consolidation found in the wider variety of mammalian
learning and memory.
Multiple spaced puffs of 5-HT to the presynaptic terminals in vitro, or
multiple tail shocks (sensitization) or siphon touch-tail shock pairings
(classical conditioning) in vivo, produce enough free PKA catalytic subunits
via the cAMP second messenger pathway that these molecules trans locate to
the presynaptic neuron 's nucleus. Notice once again that these procedures are
laboratory equivalents of repetition, known through behavioral studies to
produce memory consolidation. Using standard techniques from molecular
biology, Bartsch et al . (1998) cloned an Aplysia CREBl gene and characterized its nucleotide sequence and the amino acid sequence of its predicted
protein products. One of these products, the CREB la polypeptide isoform,
displayed 95% amino acid sequence homology to mammalian CREB proteins,
meaning that 19 out of every 20 amino acids in the protein sequences were
identical across these widely divergent species. Furthermore, the key
phosphorylation consensus site in the Aplysia protein's phosphorylation (P)
box, the site where freed PKA catalytic subunits induce their effects, is
completely conserved between Aplysia CREB la and mammalian CREB.
Every amino acid is identical across the P box sequences.
To investigate the role of this protein product in Aplysia con solidation, Bartsch et al. (1998) used the monosynaptic siphon sensory-gill
motor neuron circuit in dissociated cell culture. At two, four, or six hours
before delivering either one or five spaced 5-HT puffs , they injected AsIV or
AsIV/V, both antisense oligonucleotides, into the sensory neuron . These
145
antibodies target specific nucleotide sequences corresponding to various

exons in CREBla messenger RNAs (mRNAs), interfering selectively with
gene expression. They then measured percent change in excitatory
postsynaptic potential (EPSP) amplitude in gill motor neurons, comparing
activity in monosynaptic circuits containing injected versus uninjected
neurons exposed to the same regimen of 5-HT puffs. Circuits containing the
injected neurons exposed to a single 5-HT puff were normal compared to
uninjected neurons when measured ten minutes later. Thi s result indicates
normal short-term facilitation (which does not require CREB-induced gene
expression and protein synthesis). But long-term fac ilitation was abolished by
anti-CREB la antibodies when measured 24 hours later, indicating that the
CREB la protein is selectively necessary for inducing long-term facilitation in
the ApZysia siphon sensory-gill motor circuit.
Bartsch et al. (1998) also extended the evidence that ApCREB la is a
transcriptional activator for memory consolidation. They showed that injec ting recombinant CREBla in combination with AsIVIV antibodies combined
with five spaced 5-HT puffs rescued long -term facilitation, returning EPSP
amplitude in gill motor neurons nearly to control levels when measured 24
hours later. They also showed that Aplysia CREB la is the limiting factor in
short-term to long-term synaptic facilitation. Injecting recombinant CREB la
protein into the presynaptic neuron coupled with a single 5-HT puff produced
nearly as much facilitation (EPSP amplitude increase in gill motor neurons) as
did five spaced 5-HT puffs to uninjected neuron s. Finally, they showed that
PKA-phosphorylated recombinant CREB la is sufficient for producing longterm facilitation. Monosynaptic circuits containing injected neurons showed
the same increase in EPSP amplitude 24 hours later without any 5-HT puffs
compared to that induced by the standard five spaced puffs to uninjected
neurons. This increase was completely abolished by co-injection of either an
RNA synthesis inhibitor (actinomycin-D) or a protein synth esi s inhibitor
(anisomycin) into the sensory neuron. Standard molecular techniques also
demonstrated that Aplysia CREB la proteins are phosphorylated in vivo
following 5-HT exposure.
The role of transcription repressor CREB2 in blocking consolidation
of Aplysia learning and memory (sensitization and classical conditioning) has
also been clarified experimentally. Bart sch et al. (1995) cloned ApCREB2, a
transcription factor with a predicted amino acid sequence containing a basicleucine zipp er (bZIP) domain that interacts with Aplysia CCAAT enhancer
binding protein (ApC/EBP).26 It also contains consensus sequences for MAP
kinase phosphorylation similar to tho se of human CREB2 and mouse ATF4
proteins. Standard molecular techniques show that ApCREB2 is expressed in
sensory neurons, is a binding protein at numerous CRE sites , and repre sses
ApCREBla-mediated activation at CRE sites (all in vivo) . When a single 5-
146
HT puff is paired with an injection of ApCREB2 antiserum to the sensory

neuron in vitro, long-term facilitation measured 24 hours later was comparable to that following the standard five puff treatment to uninjected neurons.
This effect is blocked by co-application of either a protein synthesis inhibitor
(anisomycin) or an RNA synthesis inhibitor (actinomycin-D), indicating that
the ApCREB2 antiserum-induced facilitation has the key properties of
transcriptionally dependent long-term facilitation. Florescent micrographs of
sensory neurons 24 hours after ApCREB2 antiserum injection and a single 5HT puff display the characteristic new varicositie s present in uninjected
neurons following five puff treatment. However, ApCREB2 antiserum
injection does not affect short-term facilitation measured one minute after a
single 5-HT puff (compared to uninjected controls). Bartsch et al . (1995)
conclude that ApCREB2 is a transcriptional repressor and that the ratio of
ApCREBIa to ApCREB2 in presynaptic terminals determines the con solidation of short-term memory into long-term memory. This was exactly the
mechanism suggested by Yin et al.'s (1994, I995a) experiments with
Drosophila single-gene mutants. A plausible mechanism of CREB2 inhibition
is MAP kinase phosphorylation, which blocks its repressor activity. Exactly
as in mammals (Chapter Two, section 4.2 above), PKA catalytic subunits
don't bind directly to ApCREB2 molecules, but do influence MAP kinase
levels in the neuron ' s nucleus.
The evidence from Aplysia is clear. Both phosphorylated CREB Ia
and CREB 2 are key transcriptional factors for consolidating sensitization and
conditioning into their long-term forms . The y are the next step forward in the
cAMP-PKA intracellular path that influences synaptic change in invertebrates. But on which genes are these molecules acting? What protein products
are being controlled by their fluctuating ratios ? Again, the Aplysia
experimental preparation proved advantageous. And the results will be
familiar.
The role in memory consolidation of the immediate early gene
ubiquitin carboxyl-terminal hydrolase (uch) and its protein product was first
shown conclusively in Aplysia. Bergold et al. (1990) had shown that two-hour
exposure to 5-HT lowers the concentration of PKA regulatory subunits but
not that of catalytic subunits when both were measured 24 hours later , that
cAMP is probably the second mes senger mediating 5-HT intracellularly, and
that this regulatory mechanism requires new protein synthesis. They conclude
that this alteration in PKA subunit ratio somehow keeps the kinase in a
persistently active state to produce the persi stent phosphorylation seen in
long-term facilitation . Hegde et al. (1993) showed further that degradation of
an Aplysia PKA regulatory subunit (R) requires both ubiquitin protein and the
proteasome complex it binds with, and that vertebrate PKA regulatory sub units (R 1 and R n) can also be degraded via the ubiquitin pathway. Building on
147
these results, Hegde et al. (1997) showed that 5-HT, using cAMP as its
second messenger, induces a neuron-specific ucli gene in Aplysia (Ap-uch)
whose protein product, ubiquitin hydrolase, has a similar amino acid sequence
to a class of human uch. PKA phosphorylation-dependent ApCREB la is the
transcription activator for this Ap-uch immediate early gene. Its protein
product, Ap-uch, has enzymatic activity dependent on the same residue as its
human homologue and associates with the proteasome. Hegde et al. (1997)
also clarified the role of Ap-uch in long-term synaptic facilitation by injecting
sensory neurons of the Aplysia monosynaptic sensory-motor neuron circuit in
vitro with antisense oligonucleotides that block the synthesis of Ap-uch.
Circuits containing injected neurons treated with five spaced 5-HT puffs show
no increase in synaptic facilitation when tested 24 hours later. However, the
injections have no effect on short-term facilitation following a single 5-HT
puff. The ubiquitin-enhanced proteasome degrades PKA R subunits, keeping
the freed catalytic subunits in a persistently active state and subsequently
enhancing neurotransmitter release for up to twelve hours following standard
five puff spaced 5-HT treatment. Once again, readers of the detailed example
in Chapter Two (sections 4 and 5) will be struck by a feeling of deji: vu.
Invertebrate-vertebrate similarities don't stop here. CCAAT enhancer
binding protein (ApC/EBP) is a second transcription factor rapidly expressed
in Aplysia sensory neurons following 5-HT treatment. Alberini et al . (1994)
cloned the gene that encodes ApC/EBP. Its putative protein product is
homologous in its amino acid sequence to rat C/EBP, especially in its bZIP
domain. It contains a common consensus sequence within this domain for
phosphorylation by both PKA and Ca 2+-calmodulin-dependent kinase II
(CaMKII). The gene's regulatory region contains a CRE site, indicating that
its expression could be activated or repressed by CREB proteins. The protein
product ApCIEBP binds to numerous sites on a variety of early- and lateresponse genes further downstream. Alberini et al . (1994) also found that
ApC/EBP is induced by 5-HT application in sensory neurons as an immediate
early gene product, with cAMP serving as the second messenger. Long-term
but not short-term synaptic facilitation is blocked selectively in vitro when
sensory neurons are injected with an oligonucleotide that competes with
ApCIEBP at CRE binding sites. Injections of ApC/EBP antisense RNA into
the sensory neurons, which selectively inhibits ApC/EBP synthesis, likewise
blocks long-term but not short-term facilitation. So do injections of antiserum
BCA, a specific antibody against ApC/EBP. Alberini et al . (1994) also found
that ApCIEBP needs to bind to its target regulatory elements for 9-12 hours to
induce its long-term effects; this time frame is throughout the entire stabilization period for memory consolidation. These data, along with ApC/EBP's
binding affinity for a number of late-response genes known to transcribe
protein products necessary for presynaptic structural changes underlying
148
lasting increases in neurotransmitter release rates, confirms its role as a

transcription activator for protein synthesis-dependent long-term sensitization
and conditioning. Furthermore, ApC/EBP is a downstream product of an
immediate early gene for which CREB proteins are transcriptional effectors
and repressors.
CIEBP completes the current model of memory consolidation in
Aplysia (Figure 3.6 above). " Note that despite the experimental elaborations
made possible by the Aplysia preparation, these mechanisms of nondeclarative memory consolidation (sensitization and classical conditioning) are
identical to those of Drosophila olfactory conditioning. For fruit fries and sea
slugs at least, multiple realization is nowhere to be found in the current
molecular explanations of nondeclarative learning and memory . Perhaps that
is not surprising. These species exhibit only the simplest forms of learning
and memory in the simplest nervous systems. What is surprising-or
"remarkable," as the Drosophila biologists claimed in the quote at the
beginning of section 4 of this chapter-is that these same intracellular
pathways, transposed into a postsynaptic key, underlie both E-LTP and its
consolidation to L-LTP in forms of declarative memory specific to mammals
(compare Figures 2.4, 2.6, 3.5, and 3.6). In the key molecular and genetic
components, these homologies across species-from fruit flies and sea slugs
to mammals-obtain down to the level of identical amino acid and nucleotide
base pair sequences. In these quite specific respects, these "homologs" across
species are much more precise and fine-grained than are standard "homologies" cited in evolutionary biology, e.g., human and whale forelimbs. " Here
the molecular and gene compositions and the intracellular pathway interactions are shared across species. These shared features obtain despite vast
differences in brain size, organization, site of principal effect (presynaptic or
postsynaptic), behavioral repertoire, and even "cognitive logic" of the distinct
types of memories being consolidated (declarative versus nondeclarative).
Putnam's challenge has been answered empirically for one psychological
kind, memory consolidation, that from the perspective of systems neuro science seems obviously multiply realized. There is a "physical-chemical
state," the cAMP-PKA-CREB molecular biological pathway, that uniquely
realizes memory consolidation across biological classes, from insects to
gastropods to mammals.
To understand the extent to which neurobiologists take these shared
molecular mechanisms seriously, consider that in the Discussion section of
one of their Aplysia publications, Bartsch et al. (1995) speculate that CREB2
inhibition might be a mechanism for the human "flashbulb memory " effect.
"F lashbulb memory" occurs when information is stored after a single, usually
emotionally charged occasion. This information can be retrieved for long
periods after the event, sometimes for a lifetime. Personal memories from
149
"when Kennedy was shot" or, for us younger folks , "when the Challenger
exploded," are common examples. Bartsch et al. (1995) note that in mammals, the surprising and emotionally charged stimulus recruits activity in the
amygdala and other catecholinergic modulatory systems. Perhaps these systems , via the familiar intracellular second messenger pathways induced by
their modulatory neurotransmitters, temporarily relieve the repressive effects
of CREB2 (and other transcriptional repressors), thereby "priming" the
intracellular long-term facilitation and potentiation machinery to act in a
fashion that normally requires multiple stimulus presentations. One experimental result reported in Bartsch et al. (1995) is pertinent to their speculation.
When they blocked ApCREB2 activity, they induced the same level and
persistence of long-term facilitation in vitro with a single 5-HT puff that
required 5 spaced puffs in untreated neurons. One piece of evidence they cite
for their speculation about human "flashbulb memory " based upon an Aplysia
in vitro study is the shared intracellular pathways induced by modulatory
neurotransmitters initiating CREB-related synaptic facilitation and potentiation in fruit flies, sea slugs , and mammals. Putnam's challenge has not only
been answered scientifically; scientists even use this answer to suggest novel
explanations and predictions about related psychological phenomena.
6 EVOLUTIONARY CONSERVATISM AT THE

MOLECULAR LEVEL: THE EXPECTED SCOPE OF
SHARED MOLECULAR MECHANISMS 29
With the current models of memory consolidation in a variety of
species now before us, along with an overview of experimental data
supporting them and a consequence drawn again st multiple realization, antireductionists might switch to a burden of proof argument. Molecular
neuroscience suggests a unitary "switch" for memory consolidation across all
its forms and biological instantiations, and that is surprising. But as Putnam
himself emphasized more than thirty years ago (and we noted expli citly
above), the "brain state theorist" must show this same result for every
psychological kind . The multiple realization premise in the standard antipsychoneural reduction/identity argument requ ires only that the relation hold s
for some psychological kinds. Even if we rest rict our concern to the empirical
aspect of Putnam' s challenge, doesn't the burden of proof still lie on the
reductionist's shoulders? Surely molecular neuroscience can 't yet discharge
this burden of showing unitary molecular realizer s for every psychological
kind shared across species!
150
Of course it can't, at least not directly. Memory consolidation is one

of molecular neuroscience's current parade cases. Similar results for other
psychological kinds aren't yet on offer. But this case exemplifies a general
principle of molecular evolution that we should expect to obtain in the
discipline's future successes, namely, the slower rate of evolutionary change
in the functionally important ("functionally constrained") regions of enzymes,
proteins and genes. Such a principle of evolutionary conservation is implicit
in the often-cited dictum, "evolution does not start from scratch," but instead
builds upon genotypes or phenotypes already present in populations. It is also
hinted at in the following quotes from noted scientists. Insect biologists
Dubnau and Tully remark that
In all systems studied, the cAMP signaling cascade has been
identified as one of the major biochemical pathways involved
in modulating both neuronal and behavioral plasticity.
Molecular characterization of the [Drosophila] learning
mutants dnc [dunce] and rut [rutabaga] offers a striking
convergence of data with studies of learning in A. californica
[Aplysia]. More recently, elucidation of the role of CREB mediated transcription in long-term memory in flies, LTP and
long-term memory in vertebrates, and long-term facilitation
in A . californica suggest that CREB may constitute a universally conserved molecular switch for long-term memory.
(1998,438; my emphasis)
Even more directly to this point, Squire and Kandel remark:
These several findings have provided a new set of insights
into the evolutionary conservatism underlying the molecular
underpinnings of mental processes. The simplest memory
capabilities, and those that seem to have appeared earliest in
evolution, seem to be nondeclarative memories related to
survival, feeding, mating, defense, and escape. As a variety of
additional types of nondeclarative memory and then declarative memory evolved, the new memory processes retained
not simply a set of genes and proteins, but entire signaling
pathways and programs for switching on and stabilizing
synapse connections. Moreover, these common mechanisms
have also been conserved through the evolutionary history of
species: they are found in both simple invertebrates such as
Drosophila and Aplysia and complex mammals such as mice.
(1999, 155; my emphases)
151
While suggestive, however, these remarks are vague. Happily,

developments in molecular evolution can give them real scientific substance.
The functional regions of molecules participating in particular types of
intracellular biochemical pathways are the slowest to change across species
sharing the ancestor that first possessed them. From well-established general
principles of molecular evolution, finding the same molecular mechanisms at
work in memory consolidation across present species turns out not to be so
"remarkable." Nor will it be surprising to find similarly shared molecular
mechanisms underlying other psychological processes that contribute to
organisms' fitness.
Consider first a paradigm illustration of the general principle." The
insulin protein, a hormone that transports glucose into cells, is formed by
removal of a central region (the C protein) from the proinsulin molecule. The
two remaining ends of the proinsulin molecule then bond to form insulin.
Comparing the amino acid sequences of the functional insulin molecule and
the C protein across a variety of existing species shows that the C protein
evolved six times more rapidly than the functional regions: 2.4 x 10-9 amino
acid replacements per year in the C protein , or 24 per one hundred million
years, compared with 0.4 x 10-9 in the insulin molecule, or 4 per one hundred
million years . (This is in a polypeptide containing only 51 amino acids in its
entire sequence; see Ridley 1998, 179-180). As diabetics diagnosed prior to
the early 1990s know, before human insulin generated through recombinant
DNA technology became readily available, injectable functional insulin came
from cattle and pigs . Human ancestors diverged from cattle and pig ancestors
long ago, and yet the functional insulin molecule remains virtually identical
across these existing species. Insulin is but a single example of this
empirically confirmed principle of extremely slow evolutionary change in
functionally constrained regions of intracellular molecules." A related
principle holds for entire genes and proteins. The amino acid sequences of
"housekeeping" proteins, especially ones that function in basic metabolic
processes in all cell types (like insulin), evolve at much slower rates in all
regions combined than do proteins with more specialized cellular functions or
limited to specific cell types (see Ridley 1998, Table 7.1, 156).
These general principles of molecular evolution were already so
secure empirically more than two decades ago that Kimura (1983 , but going
back originally to papers published in the late 1960s) used them as one
argument for the neutral theory of molecular evolution. The neutral theory
holds that most evolutionary changes at the molecular level-for example,
changes in the amino acid sequence of a protein-are selectively neutral ,
conferring no increase in fitness . This view is in contrast to a selectionist
152
theory of molecular evolution, which holds that evolutionary changes to

molecular structures typically are advantageous, and hence a product of
natural selection . Kimura (1983) insisted that the slower rate of evolutionary
change in the functionally constrained regions of enzymes and proteins was
best explained by the neutral theory. The neutral theory explains these facts
by the higher probability that a change in the amino acid sequence in
functionally less important regions will be neutral (neither deleterious nor
advantageous). Any change in a functionally constrained region will almost
certainly be deleterious to an organism's fitness, since it will probably affect
the molecule's binding and interactive capacities. A change to a functionally
unimportant region has a much greater chance of not affecting the molecule's
binding capacitiesr"
Selectionists answered this challenge by applying a general argument
that natural selection typically favors small changes in a trait over large
changes. This general argument employs an analysis of adaptation usually
attributed to R.A. Fisher. It is general in that it purports to apply to all types of
traits, be they molecular or "macro." Consider a two-dimensional space with a
quantitative measure of some trait x on the x-axis and a measure of its fitness
on the y-axis (Figure 3.7). Assume that there is some value for x such that
individuals of some species possessing x to that degree are optimally fit, with
fitness declining away gradually from that value . This analysis creates a "hill"
of fitness values for trait x. At any given time, the individuals of a welladapted species will possess x to a degree that groups them somewhere
around (but not at) x's peak fitness value. A small change in trait x, in terms
of the length of a line segment along the x-axis from an individual's current
location, will either move the individual slightly down the fitness hill (e.g., be
deleterious) or slightly up (be advantageous). A large change in trait x will
either move the individual further down the fitness hill (be very deleterious)
or over the top and down the other side-leaving the individual lower on the
fitness hill than it was before the large change in x. Large changes to a trait's
measure in a well-adapted species are thus almost always deleterious (to the
individuals undergoing them) . Natural selection favors small changes.
Applied to molecular evolution, a change to a protein 's functionally unimportant regions might be a "small" change, and could (on rare occasions)
contribute to increased fitness (move the individual up the fitness hill slightly)
by "fine-tuning" the molecule's interactions. Natural selection will select
these changes. A change to a molecule's functionally constrained regions, on
the other hand, will typically be a "large" change. It will affect directly and
significantly the protein's capacity to interact in intracellular processes. Thus
it will either take the organism down the fitness hill directly, or over the top of
the fitness hill and to a point down the other side lower than the point
occupied prior to the change. In either case, natural selection will almost
153
never select a change to a protein's functionally constrained regions. The

chance of it selecting a "smaller" change to the amino acid sequence in a
molecule's functionally unimportant regions is higher, and that explains the
slower rate of evolutionary changes in molecules' functionally constrained
regions.
F
i
I
I
I
e
s
s
small mutation I
I
I
large mutation
Character
Figure 3.7. Fisher 's model of adaptive evoluti on: the fitnes s hill. Fitness of a trait for an
individual is graphed against degree to which the trait is possessed . The unb roken line
repre sent s the degree of the character posse ssed by indi viduals of a hypothetical well-adapted
species. A mut ation, that is, a change in the value of x, change s the bearer' s fitness. Small
changes to x have a higher prob ability of being positively adapt ive (raising the bearer' s fitness)
than large chan ges. See text for full explanat ion. (See Ridley 1996, Figure 87 .4, 182.)
The debate between neutral and selectionist theories of molecular

evolution rages on to this day (Ridley 1998, chapter 7). Fortunately, we don't
need to take a stand on it. The lesson we wish to draw from the empirical facts
of molecular evolution holds regardless of which explanation is correct. The
fact is, the rate of evolution for the functionally important region s of proteins
involved in basic cellular metabolic processes is remarkably slow . This is the
fact that both theories seek to explain. The amino acid sequences in these
154
molecules and the intracellular processes they participate in tend to remain

constant across species that share the ancestor that first possessed them.
How does this feature of molecular evolution support continual
discoveries of shared cellular and molecular realizations of psychological
kinds across species? Psychological kinds that contribute to survival and
reproductive fitness must have some effects on neuron activity, on changing
rates (or patterns) of action potentials in the neurons comprising various
anatomical pathways and circuits. These commodities are neural currencies
for information exchange within the nervous system and to the muscles that
orchestrate behavior (Chapter Two, section 3.2 above). They are common to
sensory transduction, neuronal communication, and output to motor effectors.
They lie behind the basic "rate law" through which a given neural pathway
represents information about distinct events (as differential rates of action
potentials) . Thus if psychological kinds have genuine causal effects on
behavior (and are not just causally inefficacious epiphenomena), and so are
factors that confer fitness on individuals, they must be realized in processes
that affect action potential rates in individual neurons. In tum, a neuron's
action potential rate depends directly on its basic metabolic processes governing intracellular energy exchanges and transformations, often through the
effects of second messenger signaling cascades activated by neurotransmitters
and hormones external to the neuron and impinging on its membrane
receptors ." This means that the molecular mechanisms of any psychological
kind that is causally efficacious and confers fitness on its possessors must
engage the functionally constrained regions of "housekeeping " proteins
involved in basic cellular metabolism. And as we just saw, these are the
regions of biological molecules that evolve the slowest. Thus we should
expect cellular and molecular neuroscience to find common molecular
mechanisms across existing species that share psychological processes. At
bottom, an active neuron is an active neuron, be it a fruit fly's, a sea slug 's, or
a mammal's. And these varying neural pathways and circuits that account for
psychological differences across species are nothing but interconnected and
interacting populations of membrane-bound molecular tricks for regulating
ionic conductance, whose subcelllular and extracellular signaling components
were selected for or drifted upon long before the large-scale variations
evolved. Contemporary neuroscience is increasingly at the point where it can
"link" processes occurring at these lower levels experimentally to behavior. It
is this increasing behavioral evidence of specific psychological effects due to
specific molecular manipulations within shared, evolutionarily-conserved
intra- and interneuronal signaling and transmission pathways that make these
34
accounts explanatory, and hence reductive, of psychological kinds.
These general principles of molecular evolution inform another
argument against multiple realization. The genes and proteins common to dif-
155
ferent cell types also evolve the slowest, and the cAMP signaling cascade and
CREB-mediated transcription are certainly not unique to neurons. The cAMP
signaling cascade activated by receptors coupled to intracellular G proteins is
the classic second messenger system of molecular biology . It is active in cells
of virtually every tissue type. In the paragraph that introduces the term
"second messenger," the authors of a popular current undergraduate biology
textbook write : "Cyclic adenosine monophosphate, or cAMP, is a wellstudied second messenger that activates protein kinases in many different
kinds of cells" (Purves et al. 1998, p. 850). Their diagram of a "typical"
intracellular second messenger system (on the same page) illustrates a G
protein-activated cAMP signaling pathway. Similarly CREB, along with
cyclic adenosine monophospate response element modulator (CREM), constitutes the main class of gene regulatory activators of the cAMP signaling
pathway. Gene regulatory programs induced by CREB control a variety of
biological processes in a wide range of tissues besides neurons, including T
cell development in the immune system, spermatogenesis, and the regulation
of blood pressure through angiotensin (Haus-Seuffert and Meisterernst 2000) .
In the same textbook just cited, Purves et al. (1998) use cAMP-activated
CREB-modulated gene transcription as their illustrative example of how
surface receptors can trigger gene transcription in a variety of cell types . So
the empirical facts underlying the arguments of the last two chapters are ones
that academic biologists now teach to their freshmen majors! Not only are the
molecular mechanisms of memory consolidation unified across biological
species, but they also occur in cellular and developmental processes
throughout the body .
From the molecular perspective, then, there is nothing inherently
special about neurons. They are cells specialized to conduct electrochemical
potentials down their lengths and affect this capacity in other neurons and
(ultimately) muscle fibers. Neurons are neither "wonder tissue" nor a unique
evolutionary "creation." They are a collection of molecular processes that are
also at work in other biological tissues. The same molecular mechanisms
underlying their collective functions (like memory consolidation) are at work
in cells performing a variety of biological functions in other organ systems.
Multiple realization in nervous tissue across species? Not even. When it
comes to the underlying molecular mechanisms that drive everything
biological, there isn't much "multiple realization" even across tissue types!
These discoveries give real substance to a prescient claim that Eric
Kandel made more than two decades ago. Based on the barest suggestions
about shared underlying molecular mechanisms available then, he invites us
to "conceive of learning as ... a late stage of neuronal differentiation" (1979,
p. 76): in other words, as one of the developmental processes that eventually
individuate neurons!" Since the beginnings of its "molecular revolution,"
156
leading neuroscientists have been aware of shared mechanisms across not

only species but also tissue types and biological processes. Multiple realization in its broadest sense is inconsistent with readily-available molec-ular
biological facts circa 2002. And ruthless reductionism is the vision guiding
most cellular and molecular neuroscientists, along with most molecular biologists in general.
The proponent of empirical multiple realization might try one more
counter-argument. Many philosophers are convinced that multiple realization
of psychological kinds is true despite the fact that "unitary" microphysical
accounts of the multiple realizers must exist. Few would dispute that the
"distinct" physical events realizing one and the same psychological kind have
commonalities at the level of atomic shells, electrons, and subatomic particles.
The anti-reductionist's point is that unique realization at that level of
description and explanation is irrelevant to psychology. Why not extend this
reasoning up to the biochemical/molecular biological level? The anti-reductionist can grant that there will be "unitary" mechanisms there , shared across
existing species. But why should that be any more interesting for the point at
issue-psychoneural reductionism-than are the assumed commonalities at
the microphysical level?
The problem with this reasoning is that assumed unitary mechanisms
at the microphysical level are irrelevant to the multiple realization of
psychological kinds . This is because at present microphysical descriptions
have no explanatory relevance for behavior vis-a-vis psychology. Right now
we can neither "explain behavior" nor manipulate it experimentally in
properly specific fashion by mucking around directly at the level of
microphysical posits. At present we cannot perform experimental manipulations by intervening directly on the microphysical level to generate specific
behavioral effects for even the restricted sorts of behavior employed in
controlled psychological studies, e.g., forced choices in a T maze or time
spent freezing in a novel environment. We can 't (now) explain psychological
generalizations at the microphysical level, either. i" But after nearly two
decades of molecular neuroscience, we can explain quantitative behavioral
data at the level of biochemical pathways and intracellular molecular
mechanisms (in conjunction with facts about cellular constitution and
anatomical circuits and pathways). And we can interfere directly with these
molecular mechanisms in controlled experiments that yield specific behav ioral data. We saw in the previous chapter how to explain quantitative data
about the two key behavioral features of memory consolidation using the
molecular mechanisms of LTP, the role of stimulus repetition and the effects
of retrograde interference. Unlike microphysics, molecular neuroscience has
explicit explanatory power for behavioral data, here and now. Unitary mech anisms already found at this level thus count legitimately against claims of
157
multiple realization in a way that speculative future reductions to microphysics do not.

Still, there is no reason for ruthless psychophysical reductionists to
rest on the laurels of current cellular and molecular neuroscience. Biochemistry has yet to solve the "folding problem" of amino acid strings for proteins'
tertiary structures, but advances on specific instances are being made. And
once we have a grasp of that , it is likely that explanations of what current
molecular biology has to assume (without explanation) will come forth . At
some time in my future professional career, will 1 get to write a book for
philosophers on the comprehensive biochemistry of behavior, cognition, and
consciousness, with molecular neuroscience then an "essential heuristic"? The
really ruthless reductionist in me hopes so!
7 CONSEQUENCES FOR CURRENT PHILOSOPHY OF

MIND
It is now time for scientifically inspired philosophers to give up the
multiple realization argument. Molecular neuroscience, the core of the discipline for nearly two decades, is showing that shared molecular mechanisms,
conserved evolutionarily across present-day species, realize shared psychological features and processes. Experimental evidence is now clear for the
"consolidation switch" from short-term memory to long-term memory, and
this case reflects general principles of molecular evolution we should expect
to discover for all molecular mechanisms affecting fitness. There should also
be no shrinking from the consequences of rejecting multiple realization. With
it goes the strongest-perhaps the only-empirical argument against psychoneural reduction. This leaves nonreductive physicalism, the most popular
solution to the mind-body problem among current philosophical orthodoxy,
without empirical support.
How might a scientifically inspired nonreductive physicalist respond?
One possibility is to dig into current cellular and molecular neuroscience and
find empirical evidence for multiple realization ." We've seen that this
approach cuts against general principles of evolutionary conservatism at the
molecular level. It will also require serious argument for why we should
identify the psychological features across species with diverging molecular
mechanismsr" This is an issue that few proponents of multiple realization
have felt compelled to address, although at least one critic has raised it
explicitly (Zangwell 1992) . But at least nonreductive physicalists adopting
this strategy could still count themselves as scientifically inspired . The other
alternative is to give up on scientific inspiration and count oneself among the
anti-empiricist metaphysicians of current philosophy. This is a classification
158
that many philosophers of mind try hard to avoid. However, relying on an

image of neuroscientific practice and results dating back to the 1970s (prior to
the discipline's "molecular revolution") or on "possible world" intuitions and
science fictional scenarios to defend one 's key premises (e.g ., multiple
realization) are hallmarks of "armchair" philosophy. A choice is beginning to
loom for philosophers of mind: either state-of-the-art neuroscience, which is
ruthlessly reductionistic, or anti-empirical armchair metaphysics. Middle
ground is disappearing quickly with the advancement of cellular and
molecular neuroscience.
NOTES
1 The other is explaining consciousness. We'll address it in Chapter Four.
Heil and Mele (1993) remains a good introduction to these issues and the philosophical
exotica they have spawned. Trent Jerde pointed out to me that the cognitive neuro scientifi c
empirical literature also contains work on "downward causation." He cites Pardo , Pardo , and
Raichle (1993) as an important contribution, as they study how self-control of menta l states can
direct neural responses. Jerde admits that a regress is looming . What about the neuronal
processes constituting (or at least causally affecting) the self that is directing these responses?
However, these empirical studies might hold promise for fruitful philosophical reflection,
especially in light of the arcane concepts that have dominated purely philosophical discuss ions
of mental causation.
Horgan presents the claims to follow in a slightly different orde r. I present them in this order
to emphasize the conditional nature of claims 2 and 3 (in my numbering) , with 4 asserting the
antecedent of the implied conditionals.
3
We 'll investigate multiple realization in great detail in sections 4-6 of this chapter.
Horgan (2001) articulates and defends a version of causal compatibilism by applying David
Lewis's ([ 1973] 1983) observations about implicit, contextually variable discourse parameters
to concepts like 'cause' and 'causal explanation.'
Of course there remains a "psychology of memory/" Consolidation is not the only important
feature of memory , just the one that first yielded to "ruthless" cellular/molecular reduction .
It seems reasonable to take "results published in Cell, Neuron, or similarly influential and
respected mainstream neuroscience journals" as an adequate condition on "experimentally
verified ." If you doubt this , pick up an issue yourself.
The "where to look" role is prominent in functional neuroimaging. using techniques that are
prevalent in current cognitiv e neuroscience. I'll discus s this point in the next section of this
chapter.
8
It is worth noting that more than one century ago William James (1890 ) noticed similar
sequential features in our "streams" of conscious experience. Given the apparent importance of
frontal cortex in the neurobiology of consciou sness, by discovering the cellular mechanisms of
sequential features of cognition, we might also be drawing a bead on the ones underlying
sequential features of consciousness, too. Bickle et al (200 I) charac terize these specific
159
sequential features and emphasize these similarities across cognitive processes and our
"Jamesian" conscious streams .
to I describe the "dorsal" and "ventral" visual streams in some detail in Chapter Four, section 3
below . See especially Figures 4.2 and 4.3. That level of detail is not important for the current
discussion.
11 For FEF location , see Figure 4.1 in Chapter Four below . For a good "textbook" overview of
the primate oculomotor system , see Goldberg et al. (1991).
12 I am skipping over many scientific details here because they have already appeared in print.
See Bickle et al. (2000).
13 For the anatomical location of DLPFC , sec Figure 4.1 in Chapter Four below . Sections I and
2 of that chapter contain a detailed discussion of Goldman-Rakic's and her colleagues work .
For current purposes, this paragraph will suffice .
14 In our complete model, we also developed a "Return to Fixation" mechanism that breaks off
execution in the middle of a multiple-step saccade sequence and computes the dimensions of a
saccade back to the original fixation point (Bernstein et al. 2000) . We derived components of
this additional feature directly from single-cell electrophysiology of "suppression site" neurons
in FEFs (Burman and Bruce 1997) and structural MRI and neuropsychological assessment of
two patients with anterior cingulate cortex (ACC) lesions encompassing the "cingulate eye
fields " (Gaymard et al. 1998). Since we arc just beginning to explore this component of our
biological model with fMRI in behaving humans, I won 't discuss it here in any detai l.
15 The next seven paragraphs describe the experimental task, IMRI data collection procedures,
and preliminary data analysis first reported in Bickle et al. (2001) . Since that report, we have
fully processed and analy zed the preliminary data sets, so results reported here arc new . I
include some technical details to illustrate the complexity of even a quite simple functional
neuroimaging task. A scientific manuscript describing the methods and results is currently in
preparation. This project is completely collaborative, so the next seven paragraphs should be
considered co-authored by Malcolm Avison, Vince Schmithorst, Anthony Landreth, and Scott
Holland. Please note that co-authorship (and their full agreement with my arguments) docs not
extend to the final subsection of this chapter! I also thank Kathleen Akins for helpful written
comments on a paper length treatment of the philosophical and scientific arguments of this
section .
16 This timing was also designed to make the 4-step sequences very difficult, to begin probing
activation in anterior cingulate cortex .
17
Both strategies also identified the anterior cingulate cortex.
The fit function is a second order polynomial Y = Ao + AIX + AzXz, where Ys are the pixel
values and X is the data point (frame) number (I, 2, ..., 320) . The linear and quadratic
components subtracted away arc AIX + A zx2 .
18
19
And the anterior cingulate cortex .
zo We are conducting a follow-up study to explore these different time courses of activation in
these two prominent frontal working memory regions during saccade sequencing.
ZI We've now corrected this experimental flaw by introducing a fixation point during the
control task!
160
22 Thanks to Huib Looren de Jong and Maurice Schouten for emphasizing to me the importance
of these worries.
For nonphilosophers, Ludwig Wittgenstein ended his first major work, the Tractatus LogicoPhilosophicus, with the following remark : "My propositions serve as elucidations in the
following way: anyone who understands me eventually recognizes them as nonsensical, when
he has used them-as steps-to climb up beyond them. (He must, so to speak, throwaway the
ladder after he has climbed up it." ([1919]1961). Please note that my appeal to Wittgenstein is
metaphorical. As should be clear from the discussion in the text, I am not charging cognitive
neuroscientists with literally asserting nonsense!
23
There is always the issue about how much scientific detail to include in a book addressed to
an interdisciplinary audience. More detail , comparable to the amount presented in the last
chapter, is coming in this chapter. But the details are necessary , first and foremost to show how
Putnam 's empirical, scientific challenge has actually been met These details also speak to one
of the general themes of this book, that impressive research and explanation of behavioral data
is taking place in current cellular and molecular neuroscience, and philosophers of mind and
cognitive science aren't aware of it. Thanks to Trent Jerde and John Symons for advising me to
remind readers of the "bigger picture" that all the "gory details " aim to illuminate.
24
25 In my (1998 , chapter 5), I characterize this story as "combinatorial reduction." My account is

designed explicitly to incorporate Hawkins and Kandel's evidence and arguments into a general
theory of intertheoretic reduction.
Short sequences of amino acids determine which of a handful of DNA-binding domain motifs
a given transcription factor possesses. Leucine zippers consist of a stretch of amino acids with a
leucine residue in every seventh position . DNA binding occurs at a stretch of positively
charged residues adjacent to each zipper. CREB proteins and C/EBP possess the leucine zipper
motif.
26
27 However, Bartsch et al. (2000) have recently found another transcriptional activator in
Aplysia neurons , Activating Factor (ApAF). ApAF is phosphorylated by PKA catalytic
subunits and forms dimers with both ApCREB2 and ApC/EBP. These new results show that
ApAF is a candidate memory enhancer gene further downstream from the CREB proteins. The
(molecular) beat goes on ...
28
Thanks to Huib Looren de Jong and Maurice Schouten for suggesting this standard example.
29
This section was improved by discussions with Marica Bernste in and Robert Skipper.
The next few paragraphs draw on Ridley (1998), chapter 7. This is a standard current
textbook on evolut ionary theory .
30
3 1 Another textbook example is the heme region of the hemoglobin molecule . See Ridley,
chapter 7, for detailed discussion.
Though even here . most changes will be deleterious, since an amino acid replacement will
typically affect the folded protein's tertiary structure . The neutral theory can thus explain the
slow rate of evolutionary change even in proteins ' functionally unimportant regions .
32
33 Note that this fact holds for all current theori es of neural coding , not just for frequency/rate
coding. (See Chapter Two, section 3.2 above .) The early chapters of Kandel, Jessell , and
Schwartz (2000) are a good introduction to the full range of basic metabolic processes in
neurons . Those in Shepherd (1994) provide a more compact presentation.
161
As discus sed briefly in section 4 of this chapter, psychological differences across species are
accounted for by different sequences and combinations of these cellular and molecular
events-different sequences and combinations of the "cell biological alphabet"-made
available by the more complex circuits and anatomies in "higher" cognitive species . However, I
also repeat from that earlier discussion that psychological differences are not at issue in the
multiple realization challenge; psychological similarities are.
34
See Shepherd (1994) for a good (though increasingly dated) primer on the shared molecular
and molecular-genetic mechanisms of synapse plasticity and neuron development.
35
This nonexplanatory feature of current physics vis-a-vis psychology and behavior might be a
reason why no one took seriously Paul Churchland's (1982) attempt to undercut the multiple
realization argument by arguing that "reductive unity" for psychological kinds will ultimately
be found in thermodynamics. Churchland not only failed to offer any real empirical evidence
for this possibility, but even more importantly it is difficult to see (now) how thermodynamics
could explain (in any genuine sense) concrete behavioral data-like, e.g., that from the Kandel
lab's work with transgenic mice .
36
37 Ken Aizawa suggested this in a commentary at the 2002 Southern Society for Philosophy
and Psychology annual meeting . He claimed that , e.g., "protein kinase A" is defined
functionally and multip ly realized physically . I demurred, on empirical grounds. We agreed to
leave the question open , pending further discussion.
38 This work might be part of a general project that Jim Bogen suggests as a response to my
arguments, that of developing a taxonomy of psychological kinds to see if anything systematic
(and non-hand-waving) can be found that determines which kinds can and which cannot be
"ruthlessly reduced. " I would look forward to grappling with any proposed taxonomy .
CHAPTER FOUR
CONSCIOUSNESS
Consciousness is one psychological phenomenon thought by many to

lie beyond the explanatory reach of cellular and molecular neuroscience.
Recently some philosophers have appealed to features of consciousness to
revive psycho-physical dualism (Jackson 1983; Nagel 1989). Others use them
to urge "new mysterian" skepticism about our human cognitive capacity to
solve the consciousness-brain problem (McGinn 1989). Some find in the
"hard problem" of consciousness a call to revolutionize physics (Chalmers
1996; Penrose 1994). It is common for even those who are optimistic about
neuroscience's explanatory potential to insist that explaining consciousness
will require "whole brain" resources from cognitive neuroscience, such as
sophisticated neuroimaging techniques, computational modeling in massively
parallel neural networks, and dynamical/complex systems mathematics to
analyze and interpret the results (Churchland 1995, Freeman 2000; Hardcastle
1995). Orthodoxy in the philosophy of mind and cognitive science holds that
the techniques of traditional neurophysiology and their recent supplements
from molecular biology won't be up to the task, even if cognitive neuroscience ultimately is. (Indeed, even the latter conditional remains deeply
controversial.)
There are, however, alternative voices. Perceptual neurophysiologist
William Newsome, for example, insists that "we have not yet begun to
exhaust the usefulness" of traditional neurophysiology's "single unit
approach," especially "the recent trend toward applying the single unit
approach in behaving animals trained to perform simple cognitive tasks"
(Newsome 1997,57). He lists perceptual, attention, learning and memory, and
motor planning tasks; each category has an obvious link with consciousness.
More recently, Newsome has asserted that standard electrophysiological
methods, updated with new technologies, provide the ultimate test for
neuroscience's "most remarkable hypothesis":
The most remarkable hypothesis of modern neuroscience is
that the entirety of our personal experience-from our
perception of the external world to our experience of internal
thoughts-result solely from patterned electrical activity
among the several billion neurons that comprise the central
164

nervous system. Ultimately, the most stringent test of this
hypothesis is to create realistic experiences and mental
operations artificially, by directly activating known circuits of
neurons in the brain in the absence of the external inputs that
normally elicit such mental operations." (Liu and Newsome
2000, R598; my emphases)
Newsome's experimental work on this project is the current standard. He uses

tungsten stimulating electrodes to elicit activity in tiny clusters of neurons.
Microstimulation has been a standard technique in electrophysiology for
decades, but technical developments and background knowledge now enable
exp erimenters (like Newsome) to induce electrical activity directly in just a
few highly specialized neurons , with remarkable behavioral results . (In
section 5 of this Chapter, I'll present some of the scientific details and draw
explicit consequences from them for the philosophy of consciousness.) And
notice that in the above quote, Liu and Newsome's explanatory target is
"personal experience." Consciousness in all its glory is at center stage for
these reductionistic neuroscientists. Finally, notice that they speak of "creating realistic experiences" as the ultimate test of neuroscience's guiding
hypothesis. This feature is in keeping with the theme developed in the last two
chapters, namely, the need to manipulate directly at the cellular and molecular
level to induce specific behavioral result s as a condition on the cellular/
molecular account's being deemed explanatory.
In fact, there is much work in current reductionistic neuroscience,
particularly at the cellular level, that speaks to philosophical concerns about
consciousness. Unfortunately, these results are not known to philosophers (or
to many cognitive scientists), who presume to speak of neuroscience's
"explanatory potential and/or limits. " All these results constitute a progressive
scientific approach aiming ultimately at exp laining consciousness by
explaining piecemeal its basic features. They amount to a convincing case
again st the orthodox idea that consciousness is beyond the pale of
reductionistic neuroscience, or that reductioni stic neuroscience has nothing to
contribute to its scientific investigation. The purpose of this chapter is to make
this case. Here I'll discuss:
1. experiments and results revealing the "working memory fields" of
indi vidual primate prefrontal neuron s,
2. the effects of explicit attention on action potential profiles in individual visual neurons,
and
3. microstimulation studies on tiny clusters of visual and somatosensory
neurons that induce phenomenological experiences in primates.
CONSCIOUSNESS
165
The last topic even hooks up with the clinical neurological literature on
human patients undergoing brain surgery while awake. As before, the
scientific details aren't easy, but the payoff of the trek is worth the effortunless one is committed to a mysterian philosophy of consciousness come
what may. '
1 PREFRONTAL NEURONS POSSESS WORKING

MEMORY FIELDS
Recall from Chapter Two above that short-term memory retains
information only temporarily and is very susceptible to distraction and
interference. It is now common among cognitive psychologists to divide
short-term memory into a number of distinct kinds, processes or capacities.
One of these is working memory , which is usually thought to be a temporal
extension of immediate memory. Working memory holds recently recalled or
acquired information "on line" in service of ongoing cognitive tasks,
including comprehension, reasoning, and problem solving (Baddeley 1994).
Often-cited everyday examples include retaining a telephone number one was
recently presented while dialing or waiting to dial, or composing and speaking
a complicated sentence. These everyday tasks are nicely reflected in equally
common metaphors for working memory, including "the mind 's blackboard"
or "global workspace." Like immediate memory, working memory has a
limited capacity for the number of distinct items occupying it at any given
time-psychologist George Miller's legendary "magic rule of seven [items],
plus or minus two" is often cited for it-but a vast access to all types of
memory items. Working memory also has a limited time scale: on the order of
up to 20 seconds, but perhaps extendable to a couple minutes with active
rehears al, after which the items are forgott en or consolidated into longer term
memory forms . Psychologist Alan Baddeley (1986) famously divides working
memory into three components. A central executive workspa ce is supported
by two "slave" systems: the phonological loop allows recycling or rehearsal
of small bits of verbal information, while the visuospatial sketch pad stores
images like faces or spatial layouts. The key idea is a temporary memory
capacity, constantly updated for content as cognitive tasks and demands
change that require readily available information. Distinct neural information
processing systems appear to possess their own working memory regions
(Goldman-Rakic 1996). Neuropsychologist Larry Squire and neurobiologist
Eric Kandel suggest that "working memory actually consi sts of a relatively
large number of temporary capaciti es, each a property of one of the brain's
specialized information-proces sing systems" (1999, 85).
166
At least one prominent psychologist, Bernard Baars , stresses the

continuity of working memory and consciousness. Baars writes: "All unified
models of the mind have a small "working memory" that is closely associated
with conscious processes. Working memory is that inner domain in which we
rehearse telephone numbers to ourselves or, more interestingly, in which we
carryon the narrative of our lives" (1997, 41). In his "theater model" of the
human mind, working memory is analogous to the stage containing possible
contents of current and upcoming conscious experiences, competing and
cooperating for the "spotlight" of consciousness to shine upon them. The
stage is limited in the number of actors that can be on it at any given time, but
offers vast access to many different actors. Like consciousness itself, the
procession of actors across the stage of working memory is typically serial
and sequential, forming a stream pertinent to ongoing cognitive activities
mostly occurring off the stage and in the audience of unconscious processes
(Baars 1997, chapter 2). However, one need not share Baars's love of the
theater metaphor to notice the close connection between working memory and
conscious experience.' The common sense examples of working memory all
involve first-person phenomenological consciousness. A friend barks out a
telephone number to you: "6-9-8-6-8-7-4 !" The telephone is inside the house.
You rush in, rehearing the numerical sequence in conscious inner speech. The
mobile phone is not on its pad. You keep up the constant stream of conscious
inner chatter (and maybe even outer chatter, to which you also consciously
attend). There's the phone! You consciously rehearse the number one more
time as you dial. Working memory is inexorably tied up with ongoing
conscious experience. Hence by unraveling the neural basis of working
memory, including its cellular mechanisms, we are honing in on the neural
basis of at least some of conscious experience.
The delayed response paradigm has proven to be an excellent
experimental set-up for exploring the neural basis of working memory. In
general, these tasks begin with a sequential presentation of the item or items
to be recalled, followed by a delay period during which the item must be
recalled, followed by recall or response performance cued to the initial item.
For working memory studies the delay periods are short, typically only a few
seconds. Target and test items can include shapes (e.g., faces), letters, words,
or markers of spatial locations. Goldman-Rakic et al. (2000, Figure 50.1, 734)
review some common designs of this general paradigm for working memory
studies involving both human and nonhuman primates. One design that has
proven especially illuminating for single-cell studies on nonhuman primates is
the oculomotor delayed response task (ODR). In a version that tests working
memory for spatial locations, the monkey fixates on a central point while a
visual stimulus is flashed quickly in his periphery. The monkey maintains
central fixation during a delay period after stimulus presentation, at the end of
CONSCIOUSNESS
167
which the central fixation point is extinguished and the monkey must quickly
respond by saccading (moving his eyes) to the spot of the stimulus target's
remembered location. This task allows experimenters to record the monkey's
exact direction of gaze throughout the task and also to know the exact
retinotopic location of the target stimulus. The fixation requirement during the
delay period insures comparable behavior on every trial and forces the
monkey to rely upon mnemonic rather than postural cues .
Patricia Goldman-Rakic and her colleagues (Funahashi et al. 1989)
used this ODR task on rhesus monkeys surgically prepared for chronic singleneuron recording while awake and alert. (This is the work I hinted at in
Chapter Three, section 3.2 above.) To monitor eye position precisely, they
implanted a search coil under the conjunctiva of one eye in each monkey and
used a standard computerized monitoring technique. Fixation spot and visual
stimuli are presented on a dark computer screen. The fixation point is a filled
white circle that usually appears in the center of the screen , while the
peripheral target stimuli are filled white squares. Each trial begins after a 5
second intertrial interval with the appearance of the fixation spot. After the
monkey establishes fixation on the spot for 750 milliseconds (as monitored by
the intraocular search coil technique), a visual cue appears for 500 milliseconds at one of eight peripheral locations (with location randomized over
individual trials) . Extinction of the visual cue is followed by a delay period
from one-and-one-half to six seconds . The fixation spot remains illuminated
during the visual cue and the delay periods; the monkey must maintain
fixation on it throughout both periods or the trial is scrubbed and the monkey
does not receive a reward for correct performance. Extinction of the fixation
spot marks the end of the delay period. The monkey must respond within 500
milliseconds by saccading to the remembered location of the visual cue . A
correct response requires a saccade that ends within a 6 diameter window
surrounding the visual cue 's actual location . The monkey receives a .2 ml
drop of sweetened water as a reward for each correct response. To insure
continued motivation, the monkey has been denied all liquids in its home cage
for the previous twenty-four hours and works to satiety during each testing
day (150-250 ml of liquid reward). Trained monkeys perform very well on
this behavioral task, usually displaying over 90% correct responses for all
visual cue locations across all delay periods (up to six seconds, the maximum
delay period tested).
During trials, Funahashi et al. (1989) recorded from single cells using
tungsten microelectrodes from 319 neurons in the prefrontal cortices of three
rhesus monkeys. 288 of these neurons were located within or surrounding the
caudal (back) half of the principal sulcus (PS) in dorsolateral prefrontal cortex
(DLPFC) (Figure 4.1); these locations were confirmed by a later histological
study. 170 neurons of these 288, or nearly 60%, displayed ODR task-related
168
activity, in that their average discharge rate (spikes per second) during at least
one task phase (fixation, visual cue, delay, or response) differed significantly
from their intertrial interval average. Of these 170 ODR task-related PS
neurons, 87 (30% of the total PS sample , 51% of task-related sample) had
significant delay period activity differences. 69 of these 87 showed
directionally-selective delay period activity differences that were statistically
significant compared to intertrial activity rate only when the visual cue had
been presented at one or two of the eight target locations. 50 of the 87 delay
period neurons showed statistically significant increases in activity; 46 of
these 50 showed directionally selective increases. In these 46 cells, activity
rose quickly within 100 milliseconds after visual cue presentation (at the very
beginning of the delay period) and ceased within 100-150 milliseconds after
saccade initiation (response). Even those neurons showing significant delay
period activity increases to more than one remembered target location
typically had a preferred location that elicited maximal activity increase .
Graphs of direction-selective individual PS neuron activity, with cue location
on the x-axis and normalized measure of spiking frequency increase during
the delay period on the y-axis, could be fit with to a Gaussian curve. For the
50 cells exhibiting directionally tuned delay period increases, the fit
Gaussians tended to be narrowly tuned. These cells showed high activity
increases for one cue location and actual decreases for all others. This makes
their delay period activity computationally similar to the receptive field
properties of visual neurons selective for some parameter of the external
stimulus (e.g., line orientation, motion direction; see my discussion in section
3 of this chapter, especially Figure 4.4A).
Funahashi et al. (1989) also found that varying the length of the delay
period (from 1.5 to 3 to 6 seconds) had no significant effect on PS neurons'
activity. Directional selectivity remained constant and activity rate remained
elevated throughout all delay period lengths they tested. In addition, for the
handful of directional selective delay period PS neurons that were recorded
from during one or more error trials , in which the monkey saccaded to a
location different from where the visual cue had appeared, available data
indicated that their responses were significantly depressed during the delay
period on error trials compared to correct trials .
Based on these results , Goldman -Rakic and her colleagues attribute
"memory fields" to these prefrontal PS neurons and hypothesize abo lit the
cellular mechanism of working memory:
Each neuron with directional delay period acnvity had a
"mnemonic" receptive field: only when the cue was presented
in that field did the neuron show excitation or inhibition
during the subsequent delay period. Moreover, directional
CONSCIOUSNESS
169
PPC
IFC
IT
Figure 4.1. Schematic gross anatomy of primate prefrontal cortical regions (and other cortical
regions) investigated by Goldrnan-Rakic and her colleagues. Labels: DLPFC, dorsolateral
prefrontal cortex ; FEF, frontal eye fields; ORB, lateral orbitofrontal cortex ; PPC, posterior
parietal cortex ; IT, infcrotcmporal cortex ; PS, principal sulcus ; ARC, arcuate sulcus ; OS,
orbital sulcus ; IFC, prefrontal inferior conve xity. Figure created by Marica Bernstein , see
O'Scalaidhe et al., 1997, figure 3B, 1137.
delay period activity expanded when the delay was lengthened and faltered on occasional trials when errors were
made. Therefore we propose that this area of the visual field
be termed the memory field of the neuron analogous to the
receptive fields of visual neurons or the movement fields of
oculomotor neurons. Memory fields may be the cellular
expression of a working memory process that allows mnemonic information to guide behavior. (Funahashi et al 1989,
345; my emphases)
In the final sentence (italicized above), the authors explicitly cite Alan
Baddeley as the cognitive-psychological authority on the kind of working
memory they are proposing a cellular mechanism for. As I stressed earlier in
this section, this is the kind of working memory tightly affiliated with
170
consc ious experience; and now we have a propo sed cellular mechanism for a
visual form of it, tied directly to single-cell data in behaving primates .
However, Funahashi et al. (1989) note two lacunae in their initial
study . First, they can't rule out the possibility that delay period activity
increases in PS neurons code for oculomotor dimensions of the upcoming
saccade, rath er than for the remembered spatial location of the visual cue.
Thi s alternative interpretation would block the link I'm stressing with a kind
of working memory tied closely to conscious experience, since we are not
con sciously aware of impending oculomotor commands. All the experimental
data they gathered is strictly consi stent with this deflationary alternative
interpretation.' Second , the mechanism by which a given PS neuron's
memory field is constructed was not known, although the dense anatomical
connections between DLPFC , posterior parietal regions known to process
spatial visual information (being part of the well-known "dorsal," "where" or
"how" visual stream) and some subcortical thalamic and limbic regions were
suggestive (Figure 4.1 above and Figures 4.2 and 4.3 below).
The first of these lacunae was solved by a beautiful follow-up study
(Funahashi et al. 1993). Goldman-Rakic and her colleagues compared
responses of directional selective DLPFC neurons during delay periods on the
ODR task of the (1989) study and an anti-saccade ODR task (AS-ODR) . On
the latter task , the monkey has to make a saccade in the direction exactly
opposite the location of the visual cue after the extinction of the fixation
signal indicates the end of the delay period. A filled white circle as fixation
spot indicates an ODR trial; a white cross fixation cue indicates an AS-ODR
trial. If delay period activity in directionally selective PS neurons codes for
the oculomotor coordinates of the upcoming saccade, then it.. should be similar
on ODR tasks with the visual target at one location-s-say, 90 from fixation
spot, or vertically upwards-and on AS-ODR tasks with the visual target in
the opposite location-e.g., 270 from fixation spot, or vertically downwards.
Those two stimuli prompt correct saccades to the same location, namely 90
from fixation spot (vertically upwards). On the other hand, if delay period
activi ty in directionally selective PS neurons code s for remembered spatial
location of the visual cue, then a cell should have similar activity in the ODR
and the AS-ODR tasks where the target cues are in the same location-say,
90 from fixation point (vertically upwards)-despite the different saccade
dimensions required for correct performance (e.g., 90 vertical upwards in the
ODR trials , 270 vertical downwards in the AS-ODR trials).
Funahashi et al. (1993) record ed from 108 prefrontal neurons in two
rhesus monkeys during both ODR and AS-ODR trials. 51 of these neurons
had statistically significant directionally selective delay period activity
increases in the ODR trials. Of these 51 neurons, 44 could be localized
histologically (after the behavioral and electrophysiological studies) to the PS
CONSCIOUSNESS
171
or immediately adjacent cortex. 30 of these 44 neurons, or 68%, displayed

delay period activity linked to visual cue location, not to oculomotor dimensions of upcoming saccade. (They responded in the second fashion described
in the paragraph above .) The authors conclude from this result that "the
preponderance of stimulus-dependent delay period activity in this area is
strong evidence of a prefrontal specialization for ideational processing that
does not rely on a motor code and is not mediated by motor signals"
(Funahashi et al. 1993, 754) .
These results on AS-ODR trials also strengthen the case for the
hypothesis that activity in PS neurons with memory fields provides the
cellular mechanism for the type of working memory tightly associated with
conscious experience. The AS-ODR task requires the sort of mental inversion
also required on such characteristically human cognitive tasks as the Stroop
task or the Wisconsin card sorting task; performance on all three is seriously
impaired in humans with prefrontal cortical damage." Each requires that a
habitual, sensory-driven response be suppressed, while an instruction-guided
alternate response gets selected. As Funahashi et al. (1993) note, it is common
in the human cognitive psychological literature to appeal to working memory
as, "a workspace for manipulation of symbolic representations" to explain
performance on such "mental inversion" tasks. Once again they cite Alan
Baddeley's work as their precedent for the type of working memory at issue.
Even more clearly than on their earlier study (Funahashi et al. 1989), this
follow-up study requires the kind of working memory closely tied to human
first-person phenomenological consciousness. And again, they propose a
cellular mechanism for it-activity during the delay period in PS neurons with
directionally selective "working memory fields"-that emerges directly from
single-cell primate neurophysiology in vivo during behavioral performance.
Hence at least one psychological phenomenon closely tied to conscious
experience yields to "the single-cell approach."
2 CONSTRUCTION AND MODULATION OF MEMORY

FIELDS, FROM CIRCUIT CONNECTIVITIES TO
RECEPTOR PROTEINS
What about the second lacuna in Funahashi et ai's (1989) original
study, the unknown mechanisms of memory field construction in DLPFC
neurons? The authors suggest that a circuit-anatomical account of principal
sulcus (PS) neurons' afferents and efferents provide a hint. These neurons
receive afferents from posterior parietal cortical neurons that comprise the end
of the "dorsal" visual stream. Since the ground breaking work of Ungerleider
172
and Mishkin (1982) and of Goodale and Milner (1992), this pathway has been
known to process information about the location of visual stimuli and the use
of this information in coordinated movement. Previous studies suggested
similar activity profiles between DLPFC and posterior parietal neurons during
delayed response tasks, but Chafee and Goldman-Rakic (1998) were the first
to compare these patterns directly. They recorded from the lateral intraparietal
area (7ip) in the posterior parietal cortex and prefrontal cortical area 8a in the
frontal eye fields under identical behavioral conditions, within the same
hemispheres in two rhesus monkeys performing the ODR task (see Figure 4.1
above). They recorded from 252 posterior parietal and 235 prefrontal neurons
whose firing rates changed significantly during at least one phase of the ODR
task (compared with intertrial interval rate). The only differences they could
find pertained to incidence and timing of firing . Parietal neurons responding
to visual cue presentations fired earlier (and with a slightly higher frequency)
than prefrontal neurons selective for that phase. A larger proportion of
prefrontal neurons displayed delay period activity. But activity patterns in
parietal and prefrontal neurons that displayed similar preferences for task
phases were virtually identical. Both regions contained neurons with delay
period activity and the spatial tuning properties of the neurons across these
regions were the same (Chafee and Goldman-Rakic 1998, Figure 8). The
authors note that these results suggest a principle of domain specificity by
which prefrontal neurons share specific information through cortical networks
of which they are a part. The fact that posterior parietal damage is not
associated with spatial working memory deficits also suggests that these
response similarities need not imply functional similarities. But they do
provide an interesting hypothesis about the construction of working memory
fields of prefrontal neurons, which are driven by similar neuronal activity in a
class of neurons from which they receive direct inputs .
More recently, Goldman-Rakic and her colleagues have extended this
reasoning with an intriguing suggestion about the evolutionary significance of
prefrontal neuron activity:
Organisms that primarily respond to sensory input and motor
output may have developed the capacity to hold information
on line by extending sensory responses to persist after the
termination of sensory stimulation and thereby to flexibly
instruct responses mediated by stored information, that is,
information not available in the immediate stimulus
environment. Ultimately, the elaboration of this process may
have contributed to the human capability to behave
independent of their immediate stimulus milieu and thereby
CONSCIOUSNESS
173
to flexibly prepare for and think about future consequences of

their actions. (2000, 740)
Expansion of prefrontal cortex is prominent in human evolution. With each
experiment we are inching closer through working memory to full-blown
human conscious experience.
This progress is crucial for the reductionist arguments in this book,
because the remaining steps to report from Goldman-Rakic and her colleagues' work on the neural basis of working memory takes us increasingly
down explanatory levels, to local cellular circuitries and ultimately to
subcellular components. First, Funahashi et al. (1991) hypothesized that the
different neurons that respond to different phases of the ODR task lie in
distinct layers of a single cortical column in which all component cells have
similar preferred remembered spatial locations. (This suggestion deepens the
analogy with visual cortical organization.) These columns also include
inhibitory interneurons that interact with pyramidal cells." Goldrnan-Rakic
and her colleagues (Wilson et al 1994) showed that these interneurons also
display directional selectivity in the ODR task, often the inverse of pyramidal
cells in their immediate proximity. So as an interneuron increases its activity
rate, activity in surrounding pyramidal cells decreases. This suggests that
local feedforward inhibition plays a role in constructing the memory fields of
prefrontal pyramidal cells. Anterograde (forward-tracing) and retrograde
(back-tracing) anatomical techniques have also revealed horizontal connections across cortical columns within specific prefrontal regions (Kritzer and
Goldman-Rakic 1995). These results reveal "bands" that resemble isoorientation columns in primary visual cortex. Based on these local
connections, Goldman-Rakic (1995) proposed a model of working memory
circuitry composed of clusters of pyramidal neurons with similarly tuned
working memory fields, directly connected to each other (across cortical
columns) by their excitatory axon collaterals. Inhibitory interneurons provide
reciprocal interconnections between pyramidal neurons with opposite favored
directional selectivity. These local circuit properties account for many of the
single-cell results first reported in Funahashi et al. (1989) .
The reductionist beat goes on. At the same time Goldman-Rakic was
formulating her local model of memory field construction in prefrontal
pyramidal cells, her lab was also descending down to the level of a specific
receptor on these neurons for the neurotransmitter dopamine: the dopamine
D 1 receptor. Primate prefrontal cortex receives dense dopamine projections
through the mesocortical pathway from the substantia nigra through the mid brain and forward to frontal cortex . Dopamine depletion in primate prefrontal
cortex has been associated experimentally with behavioral deficits in a
delayed spatial memory task, and prefrontal neuronal activity during a
174
delayed response task has been augmented by dopamine applied iontophoretically (in tiny currents, to individual neurons , through special
microelectrodes) and attenuated by a dopamine antagonist applied similarly
(see Sawaguchi and Goldman-Rakic 1994 for references). To elaborate on the
role of specific dopamine receptors in working memory, Sawaguchi and
Goldman-Rakic (1994) employed the ODR behavioral task using delay
periods from 1.5 to 6 seconds and a control task in which the visual cue
remained illuminated during the delay period. On control trials , the saccade
response is sensory-guided. Any behavioral impairment on ODR trials but not
on control trials would thus indicate a specific working memory deficit,
unaccompanied by a sensory or motor deficit.
Two rhesus monkeys were surgically prepared for precise eye
location monitoring (using the standard intraocular search coil technique) and
intracranial injections of drugs directly onto prefrontal regions while performing the behavioral tasks. Drugs injected during performances included two
specific dopamine D 1 receptor antagonists, SCH23390 and SCH 39166; a
nonselective dopamine receptor antagonist, haloperidol; an inactive analogue
of SCH23390, SCH23388; a selective 5-HT -2 (serotonin) receptor antagonist,
ketanserin; a selective dopamine D2 receptor antagonist, sulpiride; a
dopamine D2/D3 receptor antagonist, raclopride; and sterile saline . Six visual
cue locations were used for both ODR and control trials . Drug injection sites
were confirmed by a histological study after the experiment.
The various drugs were injected into 45 sites across DLPFCs of the
two monkeys. Haloperidol injections, the nonspecific dopamine receptor
antagonist, and injections of both specific D 1 receptor antagonists, had
significant effects on ODR performance at 22 sites, all clustered in a region in
or immediately adjacent to the PS. Applications both decreased the accuracy
of memory guided saccades and increased the latency of saccadic response
(Sawaguchi and Goldrnan-Rakic 1994, Figures 3, 4, 5). Typically these deficits appeared 1-3 minutes after an injection , peaked after 20-40 minutes , and
declined back to non-injection levels after 60-90 minut es. Injections of these
drugs at DLPFC sites other than the PS had no effect on ODR performance.
These deficits were always restricted to one or two target locations per
effective injection site, a result in keeping with the earlier discovery of
"working memory fields" of individual PS neurons . Memory-guided performance deficit s were correlated directly with length of delay period. Longer
delay periods (up to six seconds, the maximum delay period tested) were
associated with increasingly inaccurate memory -guided saccades (to specific
visual cues locations) and longer latencies (Sawaguchi and Goldman-Rakic
1994, Figure 6). The deficits were also dose-dependent. Larger injections (up
to 60 micrograms of solution, the largest doses tested ) were associated with
increasingly inaccurate memory-guided saccades and longer latencies
CONSCIOUSNESS
175
(Sawaguchi and Goldman-Rakic 1994, Figure 7). On the other hand, these
dopamine D I receptor antagonists had no effects on accuracy or latencies of
sensory-driven saccades on the control trials , at any delay length or dosage.
Their effect was specifically on work ing memory, not the sensory or motor
aspects of the ODR task. None of the other drugs-the selective dopamine D2
receptor antagonist, the D2/D3 receptor antagonist, the 5-HT-2 receptor
antagonist, or saline-had any effects on accuracy or latency of memoryguided saccades on the ODR trials or sensory-guided saccades on the control
trials, even when injected directly into the effective PS sites of the dopamine
D 1 receptor antagonists.
Based on these results, Sawaguchi and Goldman-Rakic remark that
"it seems reasonable that the behavioral impairment induced by [D 1] receptor
blockade relates specifically to the working memory process now recognized
to be a cardinal function of prefrontal circuits," especially ones involving PS
neurons with directionally selective working memory fields. (1994, 522) .
Furthermore, their findings "provide evidence that the activation of the
mesocortical dopamine system activates D I receptors of the monkey PFC,
thereby modulating the mnemonic process associated with the PFC' (1994 ,
524; my emphasis). And given that mesocortical dopamine fibers preferentially form synapses with pyramidal cells in primate prefrontal cortex, and
that pyramidal cells containing a phosphoprotein associated with D 1 receptors
are primarily output neurons in cortical layer VI-projecting to thalamus and
other cortical regions, ideally for communicating working memory information to ongoing and upcoming cognitive processing-Sawaguchi and
Goldman-Rakic extend their explanation down an additional level. "It is,
therefore, plausible that the mesocortical dopamine system specifically
activates projection neurons located in the deep layers of PFC via
postsynaptic D 1 receptors on their basilar and/or apical dendrites in the upper
strata" (1994 , 525) . This is increasingly a subce llular account of the mechanisms of working memory-of a type , I remind you, that is closely
associated with conscious experience. Reductionism marches on!
The role of dopamine D 1 receptors on memory field construction and
modulation was further clarified-and reduced-by another follow-up study
using the ODR task in primates outfitted surgically for iontophoretic appli cation of drugs and single -cell recordings (Williams and Goldrnan-Rakic
1995). Their technique used a "quad-barreled" carbon-fiber microelectrode
with one barrel outfitted for extracellular neuronal recording and the other
three outfitted for drug delivery in miniscule quantities directly onto the
neuron being recorded from . Williams and Goldrnan-Rakic discovered that
SCH39166 delivery, the selective dopamine D 1 receptor antagonist, at less
than 30 nanoamperes (nA) injection current accentuates delay period activity
increases in directionally selective PS neurons with memory fields during
176
ODR trials; but delivery at greater than 50 nA injection current virtually

obliterated delay period activity, even after a stimulus in the neuron 's favorite
visual cue location. The drug had no effect on a neuron's activity during any
phases of sensory-guided control trials . Raclopride, the dopamine D2/D3
receptor antagonist, had no significant effects on neuron activity during any
periods of either ODR or control trials. Williams and Goldman-Rakic infer
from these results that "the normal action of dopamine is to constrain
neuronal activation during performance of a working memory task" (1995,
575) . They then carry their hypothesized mechanism of memory field
modulation down to the intracellular level : "A known mechanism of
inhibition by D 1 action is the attenuation of a slow inward sodium current
which normally supports activation of the cell by excitatory inputs . . .
Blocking the D 1 receptor may simply disinhibit specific excitatory input to
the same cells" (1995 , 575).6
So far we've only seen these cellular and molecular mechanisms at
work for the type of visual spatial working memory employed in ODR and
AS-ODR tasks. Is there evidence for cells with memory fields and their
molecular modulation in other types of working memory also affiliated
closely with conscious experience? Goldman-Rakic and her colleagues
(Wilson et al. 1993) found similar results in single-cell studies of neurons in
the inferior prefrontal convexity below the PS (see Figure 4.1 above). They
used a delayed response task keyed to non spatial information about the color
and form of a visual stimulus. This region was a promising candidate for such
a study because it receives direct input from the inferotemporal cortex, the
end of the "ventral" visual stream involved in object identification (based on
such features as color and form) . Lesions in this prefrontal area have been
linked to memory deficits for objects. Finally, sensory neurons in the
prefrontal cortex have foveal receptive fields, the area of highest acuity on the
retina for visual parameters that figure in object identification. Monkeys were
surgically prepared for chronic single-cell recordings in this area while
performing an ODR task that randomly interposed spatial or pattern memory
trials. On pattern DR trials, a stimulus pattern appears in the screen center
(after the monkey establishes fixation on the central spot). The pattern is
extinguished and the delay period ensues. After the delay period, the monkey
must make a rightward saccade response for one type of visual pattern and a
leftward saccade response for another type. Spatial DR trials are likewise
simplified to only two visual cue locations, one left and one right of the
central fixation point. Thus the two trials require the same response, but differ
in the type of working memory that guides them.
Wilson et al (1993) found 31 neurons in the prefrontal inferior
convexity that displayed delay period activity. 24 of these responded
selectively during the pattern DR trials and 6 responded during both spatial
CONSCIOUSNESS
177
and pattern trials. Most of these neurons were selective to only one of the two
patterns (i.e., their activity during the delay period of pattern DR trials did not
increase when the other pattern was presented). Pictures of face s (both
monkey and human) were particularly favo red stimuli for these neurons.
Wilson et al. (1993) found some that were responsive during the delay period
of pattern DR trials only to specific faces , with no response after others. This
last result is especially intere sting for the point stressed just above. Activity
profiles of prefrontal inferior convexity neurons share numerous features with
those in the inferotemporal (IT) cortex. The prefrontal inferior convexity
receives dense projections from IT cortex, suggesting a similar "circuit
property" contribution to those neurons ' memory fields. Wilson et al. write:
"These connections presumably provide signals about the attributes of foveal
visual stimuli on which prefrontal circuits operate" (1993 , 1957). One
difference is crucial, however. Activity in IT neurons, even to favored objects,
declines with object familiarity. Wilson et al. (1993) observed no such decline
in prefrontal inferior convexity neuronal responses. Such sustained activity
even with object familiarity would be fitting for neuron s that hold such
information transiently on line for other cognitive processes when the visual
stimulus is no longer present.
More recently, Goldman-Rakic and her colleagues (O'Scalaidhe et al.
1997) have found further evidence for the modular organization of prefrontal
cortex in a single-cell study with faces presented as visual stimuli . Although
there was no explicit memory component to their task, they continued
recording from prefrontal neurons for a few seconds after stimulus offset. In a
deliberate attempt to record from as wide a range of DLPFC and lateral orbital
prefrontal sites as possible, they recorded from over 1700 sites in three rhesus
monkeys trained only to maintain fixation while a foveal stimulus is presented
and for a few seconds afterwards. They found 46 neurons whose activity
increa sed at some point during or immediately after face presentation, and
which gave no statistically significant response increases to any other type of
patterned stimulus. 44 of these 46 face-sel ective neurons were located in the
prefrontal inferior convexity or the immediately surrounding lateral orbitofrontal cortex. The other two were located in a region of the frontal eye fields.
None were located along the principal sulcus (PS) or in the immediately
surrounding superior frontal cortex, the predominant sites of neurons with
spatial working memory fields . Hence all these face selective prefrontal
neurons lie in regions that receive dense projections from IT cortex. Many of
these neurons showed a tenden cy to begin firing immediately after the face
stimulus extinguished. This activity lasted throughout the entire 2500
milliseconds until the next stimulus appeared. Two of the monkeys in this
study had been trained (as part of another study) to perform ODR tasks ; but
the third had never been trained on any memory task . The authors conclude
178
that "the capacity for face-selective persistent firing and delay-period activity
does not depend on intention to make a respon se, but appears to reflect an
intrinsic property of the neurons' responses to visual stimuli" (0' Scalaidhe et
al. 1997, 1137).
Finally, Goldman-Rakic and her colleagues have combined all this
detail-single-cell physiology, circuit connectivities, and subcellular receptor
effects (including NMDA receptors)-into a computational model and com puter simulation that examines "the synaptic mechanisms of elective persistent activity underlying spatial working memory in the prefrontal cortex"
(Compte et al. 2000 , 910) . Their results include activity profile s from
simulated pyramidal neurons that "reproduces the phenomenology" of the
single-cell recordings in behaving primates engaged in ODR tasks (see their
Figure 4). The model descends levels in its biological realism and comparable
performance measures down to neurotransmitter receptor channel contributions to neuron oscillations during delay periods (see their Figure 6).
Simulated interneurons also display movement fields like those predicted by
the local circuit model based on experimental single -cell results (see their
Figure 7). The network's working memory performance is even resistant to
distractions (see their Figure 8) and yields a number of novel predictions
testable by single-cell in vivo and neural tissue slice in vitro experiments
(Compte et al. 2000, 922).
Collectively, this research is a first-rate collection of state-of-the-art
neuroscience, pursued methodologically at multiple levels but always aimed
at explanatory mechanisms at the lowest level that technology and theory
permit pursuing at any given time. Single-cell neurophysiology sets the stage
for this developing explanation of primate working memory capacities. The
explicit tie between this kind of working memory and conscious experience,
stressed in both cognitive and "folk" psychology, indicates that at least some
features of consciousness are not beyond reductionistic neuroscience, as many
philosophers assume them to be.
3 EXPLICIT ATTENTION AND ITS UNREMARKABLE

EFFECTS ON INDIVIDUAL NEURON ACTIVITY
Although he is best known in philosophy for espousing elim inative
materiali sm, Paul Churchland nevertheless advocates the reality of conscious
experience and its status as a key expl anatory goal of neuroscience. He writes:
"Consciousness is at least a real and an important mental phenomenon, one
that neuroscience must acknowledge as a prime target of its explanatory
enterprise" (1995 , 213) . He lists "steerable attention " as one of its "salient
dimen sions," writing that "consciousness is something that can be directed or
CONSCIOUSNESS
179
focused-on this topic instead of that, on these things rather than those , on
one sensory pathway over another, even if one's external sensory perspective
on the world is held constant (1995, 214)." Hence the discovery of
neurobiological mechanisms for explicit selective attention and its effects will
be another step forward toward explaining principal features of
consciousness.
However, it should be noted explicitly that Churchland himself
advocates a "connectionist-inspired" neurocomputational explanation of
selective attention based on "recurrent" (feedback) projections from higher
back down to lower layers of neuron-like processing units (1995, 215-226,
especially 217-218). While his account shares some rough structural features
with the neurophysiological details I'll present in this section, the level at
which he pitches his "neurocomputational perspective"-activation vectors
across populations of densely-connected neuron-like processing unitssuggests that he too is pessimistic about explaining features of consciousness
at the cell-biological level. "Connectionist AI" uses some distinctively
unbiological principles and posits , and its level of abstract modeling has been
eschewed by practicing computational neuroscientists at least since compartmental modeling became popular in the mid-1990s. In light of his recent
writings , Churchland's attitude toward the ruthless reductionism espoused in
this book is unclear.
In this section I'll begin with the neural effects of explicit conscious
attention. For that is where the "single-cell approach" has paid remarkable
dividends. But these results in tum guide us toward a mechanistic account of
attention itself, specifically for visual attention (a robustly conscious form of
the phenomenon!) but generalizable beyond that single modality . And philosophically, the payoff of this research is important. At the level of the
individual neuron, explicit attention appears to elicit a very mundane effect, as
mundane as that elicited by simply turning up the salience or contrast level of
extern al stimuli. Hence the second-to-Iast intuition on which consciophiles
might build a case-that consciousness, even if ultimately explainable in
neural terms , is at least a very special kind of neural event and cause-is
being dismantled by reductionistic neuroscience. (We 'll examine consciophiles' last available intuition , the problem of qualia and subjective
experience, in the final sections of this chapter.)
Psychologists have studied the behavioral effect s of explicit
conscious attention for decad es. These studies have confirmed a handful of
gene ralizations. Directing explicit conscious attention to a spatial location in
anticipation of a stimulu s there both decreases response time to stimulus onset
and lowers the threshold of detection. A common experimental paradigm goes
back to Michael Posner (1980). Subjects are cued to expect a stimulus at one
location with a high probability, and with only very low probabilities at other
180
locations. They are then tested for their capacity to detect stimuli of various
intensities or for the speed of their responses to stimuli appearing at all these
locations. Subjects are faster to respond and are capable of detecting less
intense stimuli at cued rather than uncued locations, a phenomenon Posner
(1980) refers to as "attentional benefit." Downing (1988) extended Posner's
paradigm and showed that attention directed to spatial locations also affects
sensitivity and bias for brightness, orientation, and form discriminations there .
This sensitivity decreases with distance from the cued location. More
recently, Rossi and Paradiso (1995) showed that preferential processing
measured behaviorally by detection performance on near-threshold stimuli is
also specific to features independent of location. Subjects are better able to
detect peripheral gratings at specific orientations presented at near-threshold
intensities when performing a central attention task involving stimuli oriented
to that same degree. They obtained similar results on near-threshold peripheral stimuli at specific spatial frequencies.
John Maunsell and his colleagues have studied the effects of explicit
selective attention on cortical representations of visual information. As he
stated nearly a decade ago, such findings "are changing the way we view the
visual cortex," from an extractor of sensory attributes encoded in retinal
images to an active processor that selects a limited portion of the visual image
for concentrated attention and reshapes it to accentuate current interests
(1995, 764-765). Maunsell is also committed experimentally to the "singlecell approach," writing that while technologies continue to develop for
investigating human functional brain organization, "animal models remain the
only source of detailed information about how neurons encode visual
information" (1995 , 765) . As far back as a decade ago, his lab perfected a
technique to isolate the effects of explicit attention on individual neuron
activity throughout all regions of the visual processing hierarchies. ' They use
rhesus monkeys, surgically prepared for chronic single-cell recordings while
alert and awake, as their experimental preparation. As Goldman-Rakic and
her colleagues are accomplishing for working memory , Maunsell's group is
using "the single-cell approach" to reveal the neural basis of another
phenomenon closely associated with consciousness . This approach, we are
assured by philosophers of mind and many cognitive scientists, can't possibly
help us "explain consciousness."
Maunsell and his colleagues use a delayed matching-to-sample task as
their behavioral measure of attention. The monkey fixates on a central spot
and is prompted by a cue to attend to one region of visual space . The monkey
then depresses a button to indicate readiness and sample stimuli appear 500
milliseconds later at two locations . Their stimuli include orientation bars,
moving spots, or color patches. One stimulus appears at the spot that the
monkey was prompted to attend , the other appears somewhere else in the
CONSCIOUSNESS
181
monkey's visual field. The samples appear for 500 milliseconds and extinguish, after which a delay period of 500 milliseconds ensues, followed by the
appearance of test stimuli. The monkey's task is to indicate whether the test
stimulus at the cued location exactly matches the sample stimulus presented
there (e.g., bars at the same orientation, motion in the same direction, same
color patch) . He either releases the "readiness" button within 500 milliseconds
to indicate match, or continues to depress the button for at least 750
milli seconds more to indicate no match . The monkey must maintain fixation
on the central spot during the entire trial, with eye position measured precisely
using the intraocular search coil technique. He is rewarded only if he
maintains fixation on the central spot throughout the trial and correctly
indicates whether sample and test stimuli match or don't at the cued location.
Matches and nonmatches at the two locations are uncorrelated, so the monkey
gains no advantage by attending to the wrong (uncued) location. Given the
time constraints, this is a difficult attention task , yet trained monkeys
regularly approach 90% correct trials .
While monkeys perform this attention task , Maunsell and his
colleagues record activity in individual neurons throughout striate and
extrastriate cortex using standard extracellular techniques and analysis. The
region of cortex being recorded from determines the nature of sample and
target stimuli employed and the receptive fields of individual neurons being
recorded from determine the spatial location of the two sample and test
stimuli . One stimulus is located directly within the receptive field of the visual
neuron being recorded from (the region of visual space in which stimuli elicit
activity above baseline action potential rate) ; the other stimulus is located in a
non-overlapping region directly across the central fixation point from the
neuron 's receptive field. Experimenters can then compare activity in an
individual neuron under conditions of attention to its receptive field (the
"attended" mode, as measured by correct performance on the matching-tosample task when prompted to attend to the location of the neuron's receptive
field) and conditions of attention directed elsewhere (the " unattended" mode,
as measured by correct performance when prompted to attend to the other
location). By presenting the same visual stimuli to the monkey in both
"attended mode" and "unattended mode" match ing trials, any differences in
neuronal respon ses across the two trials reflect attention ("behavioral state")
effects on single neuron activity.
In a (1995) review, Maunsell points out that simple versions of this
behavioral task and neurophysiological recording had demonstrated signifycant attention effects on the activity in neurons as far back in the visual
processing hierarchy as extrastriate area s V4 and MT (see Figures 4.2 and
4.3). He gives data from one neuron in area V4 that responded to the same
stimulus in its receptive field with roughly a 50% increase in action potential
182
rate when the monkey was attending to that stimulus than when it attended to
the other stimulus (1995 , Figure 2). He also shows an effect in MST when the
monkey was forced to attend to only one of two motion stimuli both occurring
within the neuron ' s receptive field . The neuron's action potential rate
increased to nearly 100 spikes/second when the attended motion stimulus
moved in its preferred motion direction, but quickly declined to below
baseline response levels when the attended stimulus suddenly changed
direction (1995 , Figure 3). Maunsell concludes that "the widespread statedependent modulations revealed by these studies show that the overall pattern
of activity in V4 and MST can change markedly depending on what aspect of
the visual scene is the focus of attention" (1995 , 767-768). He even argues
that there are reasons to think that the studies available circa 1995
underestimate to full range of state-dependent modulations on neuronal
responses. We should take the attention modulations on neuronal activity that
had been found so far as marking the "lower limit" of state-dependent
contributions.
. .
(IT
Temporal
Figure 4.2. Schem atic illustration of the gro ss anatomy of some sensory portion s of prim ate
cortex. Abbre viations: VI , primar y visual cortex ; V2, V3, V4, regions of extrastriate corte x; IT,
inferolemp oral corte x; MT, middle tempo ral corte x; MST , med ial superior corte x; PPC ,
posterior parietal cor tex; S I, primary somatosensory cortex. Repr inted (with additional label s)
from Fundamen tal Neuroscience , M.zigmond , F. Bloom , S. Landi s, J. Roberts, and L. Sq uire
(Eds.), 844, Cop yright 1999. with permission from Elsevier Scienc e.
CONSCIOUSNESS
183
Ventral
dtj..
:-----------+-_.-.
Figure 4.3. Simplified circuitry diagram of neural regions contained in the primate "dorsal" and
"ventral" visual streams . Abbreviations are the same as in Figure 4.2, except : Sc. (midbrain)
superior colliculus ; LON, (thalamic) lateral geniculate nucleus ; PIT, posterior inferior temporal
corte x; AIT, anterior inferior temporal cortex ; VIP, ventral intraparietal area; 7a, Brodmann's
area 7a (in posterior parietal cortex); FEF, frontal eye lieIds. (Figure created by David
Winterhalter.)
These findings of increased neuronal activity when a subject attends

to a sensory neuron's receptive field correlate directly with psychological
findings about the behavioral effects of explicit selective attention. But there
184
are two ways that sensory neuronal activity modulated directly by attention
state could produce faster reaction time and better discrimination at threshold
stimulation. The simplest way is just to increase the neuron's action potential
rate to all degrees of the stimulus parameter it responds to. Consider, for
example, a visual neuron in area V4 that responds to stimulus orientation. The
activity of such a neuron to orientation patterns in its receptive field fits a
Gaussian curve with maximal activity (action potentials/second) elicited by
one specific orientation, slightly less activity to related orientations, dropping
down to baseline activity (or below) for opposing orientations (see Figure
4.4A). If attention modulates neuronal activity in this simple manner, the
frequency of action potentials generated to a stimulus while attending to it
will be greater at most or all degrees of stimulus orientation, as reflected in
the increased height of its entire tuning curve under attentive as compared to
unattentive conditions (see Figure 4.4B). The neuron's stimulus selectivity is
unaltered, as reflected in the similar widths of its tuning curve under attended
and unattended conditions. This effect is referred to as "multiplicative
scaling" and it is a common neuronal effect. A similar effect in sensory
neurons can be induced by simply increasing the salience of the external
stimulus. The behavioral effects of attention, namely lower response thres holds and speed , would result because multiplicatively scaled neuronal
responses have better signal-to-noise ratios and thus signal stimulus features
more reliably. However, this mechanism would be deflationary for consciophiles because explicit conscious attention would then not be "specially" or
"uniquely" realized neurally . It would be just another mechanism, albeit an
internal/endogenous one, for "turning up the gain" on individual neuron
activity. (And for that, there are numerous known neurobiological mechanisms.) Ho-hum for explicit conscious attention, from the brain's perspective.
On the other hand, explicit conscious attention could generate these
behavioral effects by virtue of a much more robust and unique neuronal
modulator. Perhaps it alters directly the stimulus selectivity of individual
sensory neurons. Perhaps it sharpens effected neurons' tuning curves,
increasing their activity to the most preferred stimulus degree and closely
related ones, but dampening normal activity to those further removed (see
Figure 4.4C). This would make neuronal activity in affected neurons signal
more precisely the attributions of the attended visual stimulus. Sharper tuning
curves provide more fine-grained representations of stimulus dimensions,
which would improve both detection threshold and speed to attended
locations and account for the behavioral data. Consciophiles also could be
heartened if this is the cellular mechanism of attention modulation because
increasing neurons' stimulus selectivity is not a common neurophysiological
CONSCIOUSNESS
185
c
t
p
o
t
a
I
," ,
,,, ,,
,
,,, ,,
,,
,,
,,
,,,
,,
,
,,
,,,
,,
,
,,
,
s
e
c
Degree of stimulus orientation
Figure 4.4 . Tun ing cur ves of an orientation-selective V4 neuro n. A. Response under normal
conditions fits a Gaus sian curve, with maximal response (highest action potentials per seco nd)
to stimuli at a single orientation, near-maximal respon ses to orientations close to its most
preferred stimulus, decl ining to responses blow baseline response levels to orientations
increasingly different from its most preferred stimulu s. B. "Multiplicative scaling," a general
increase in the neuron 's response to stimuli of any orientation, is one possible effec t of explicit
selective attention to the neuro n' s receptive field. C. A "sharpe ned" tuning curve is a second
possible effec t of explicit selective attention to the neuron ' s receptive field . Scal ing and
sharpening changes are exagge rated in B and C to better illu strate the possible effects. See text
for further discussion and references to graphs of changes actually measured in V4 and V I
neuron s under conditions of explicit selective attention to their receptive fields. (Figures
created by Marica Bernstein.)
186
dynamic. It might even hearten mysterians, since known neurobiological

mechanisms for this effect are not so readily apparent.
To test these competing explanations, Carrie McAdams and Maunsell
(1999) have recently employed a version of the delayed matching-to-sample
behavioral measure of attention described above with single-cell recordings in
areas V4 and V 1. Orientation bars always appear in the receptive field of the
neuron being recorded from, while color patches appeared in the opposite
location . The monkey is cued in the usual way about which location to attend
to. "Attended mode" constitutes attention to the orientation bars in the
neuron's receptive field while "unattended mode" constitutes attention to the
color patch outside the neuron's receptive field. Monkeys must maintain
fixation on the central spot throughout each trial and correctly indicate (by
releasing or continuing to depress the button) whether the sample orientation
bars match the test orientation bars (vertical or horizontal) when cued to
"attended mode," or whether the sample color patch matches the test patch
when cued to "unattended mode." By comparing individual neuron activity in
"attended" and "unattended" modes to the same combination of visual
presentations, McAdams and Maunsell isolate the modulatory effects of
explicit selective attention on individual neuron activity. The time constraints
on the behavioral task clearly elicit subjective conscious attention by human
performers. By comparing a neuron 's activity across the entire orientation
tuning range under attended and unattended conditions, the debate over the
nature of attention's modulatory effect can be resolved. Does explicit
conscious attention elicit multiplicative scaling or a sharpened tuning curve?
Experimental results with over 200 orientation-selective V4 neurons
and 121 V 1 neurons clearly confirm the multiplicative scaling hypothesis
(Figure 4.4B above; see McAdams and Maunsell 1999, Figures 2, 4, 5, 6, 7,
and 10). For tuning curves of both individual neurons and averages across
populations, the amplitude of attended responses (frequency of action potentials) compared to unattended responses to the same orientation stimulus was
statistically significantly greater, for nearly all degrees of orientation. The
only exceptions were orientation degrees that generated no response over
baseline in the neuron. Explicit conscious attention to a visual neuron's
receptive field enhances its action potential rate to virtually every degree of
the stimulus parameter it responds to. However, the standard deviations of its
tuning curves to the entire range of orientation degrees remain constant across
"attended" and "unattended" modes. This means that the two tuning curves
have nearly identical widths. Hence explicit conscious attention does not
affect a neuron 's stimulus selectivity. Finally, neurons' "attended" and
"unattended" tuning curves had nearly identical asymptote values. This means
that explicit conscious attention had no effect on neurons' responses to
" unpreferred" orientation degrees. These results yield a decisive conclusion .
CONSCIOUSNESS
187
Directing explicit conscious attention to the location of a given sensory

neuron 's receptive field simply elicits multiplicative scaling. It only "turns up
the gain" on the neuron's response to all degrees of the stimulus parameter,
without sharpening its stimulus selectivity. Intere stingly (and unlike some
other studies that presumably used less sensitive measures and visual stimuli)
McAdams and Maunsell (1999) found significant multiplicative scaling with
attention all the way back to individual neurons in VI, the primary visual
cortex, which is the first region of cortex to receive visual inputs in mammals
via the lateral geniculate nucleus of the dorsal thalamus. However, they could
not rule out that these small effects in V 1 are caused by subtle position
differences instead of attention modulation.
McAdams and Maunsell explicitly note that selective (conscious)
attention therefore has the same effect on neurons as procedures as
(metaphysically) mundane as increasing stimulus salience and contrast. They
write: 'The phenomenological similarity between the effects of attention and
the effects of stimulus manipulations raises the possibility that attention
involves neural mechanisms that are similar to those used in processing
ascending signals from the retinas , and that cortical neurons treat retinal and
attentional inputs equivalently" (1999,439). Recently Charles Gilbert and his
colleagues (Gilbert et at. 2000) have described extensive feedback projections
from higher regions in the visual processing hierarchies and long-range
horizontal connections within a given region that provide an explanation of
endogenously-generated attention effects, as far back in the visual processing
hierarchy as V 1. This "plexus" of horizontal connections link neurons with
widely separated receptive field locations, but with similar favored dimension
degrees. In other words , cells that are maximally responsive to the same
degree of a visual stimulus parameter, e.g., orientation direction, but to stimuli
at very different locations in the visual field , are linked acros s cortical
columns. This network is formed by the axon s of cortical pyramidal neurons
and the longest connections can span 5-6 mm, with multiple synapses from
one end of the axon to the other. Functionally, this extends the area of visual
space represented by the area of cortex from which individual neurons
integrate input by an order of magnitude greater than the neurons' own
receptive fields . "Contextual" influences, including attention, are likely to be
mediated in part by these networks of horizontal connections and by feedback
connections from higher cortical areas. Gilbert et at. even propose a cellbiological mechanism for these modulatory effects: "One possible mechanism
underlying the attention effects is a gating or modulation of the synaptic
effects of long range horizontal connections by feedback connections from
higher cortical areas " (2000, 1224; my emphasis). This is what explicit
selective conscious attention is, according to state-of-the-art current
neuroscience.
188
Even more recently, McAdams and Maunsell (2000) have separated

the correlates of spatial and feature-directed attention effect s on individual V4
neurons, similar to the effects investigated behaviorally by Rossi and Paradiso
(1995) described earlier in this section. Two rhesus monkeys surgically
outfitted for eye position monitoring and single-cell recording while alert and
awake were trained to perform two versions of the delayed matching-tosample task. The first was a purely spatial attention task, where both the
receptive field of the V4 cell being recorded from and the opposite location
contained orientation stimuli. The second was a spatial and feature task,
where the receptive field of the V4 cell being recorded from contained orientation sample and test stimuli, while the other location contained color sample
and test stimuli. ("Attended mode," "unattended mode," and "correct trial"
were defined as above.) By subtracting the average action potential frequency
of a V4 neuron in "unattended mode" from "attended mode" and dividing by
the average value in "unattended mode" for "space" and "space and feature"
trials, any increase in the "space and feature" value for a given neuron would
reflect increased attention modulation due to the feature component. For the
entire population of 71 V4 neurons recorded from, McAdams and Maunsell
(2000) found a 31 % increase in activity during the spatial attention task, but a
54% increase in activity during the space and feature task. They conclude that
"spatial attention and feature attention coexist in a relatively early stage of
visual processing, cortical area V4," that "the same neuron can receive
multiple types of attentional inputs," and that "directing attention to a
stimulus feature might modulate the responses of neurons throughout the
visual field ," not just to the ones whose receptive fields contain that stimulus
(2000, 1754). Notice that in their "space and feature" task, the color stimulus
always appears at the location outside of the receptive field of the cell being
recorded from.
What we have in the experimental work of Maunsell and his
colleagues, coupled with that of Gilbert and his, is an expl anation of explicit
selective visual attention emerging from work at the "single-cell" level.
Again, what some philosophers insist can't be done turns out to be a booming
research program in mainstream current cellular neuroscience. But this
research and its results carry an even more telling philosophical consequence.
Some "consciophiles" have reconciled themselves with psycho -physicalism.
They maintain this uneasy truce by holding out for some special, unique
nature of consciousness's neural realization and effects. Even if consciousness
is neural -cum-physical, it must be a special type of neural-cum-physical
event, cause, or effect. But the upshot of the research discussed in this section
denies even this much, for at least one prominent feature of consciousness. At
the level of individual neurons, explicit con scious selective attention
accomplishes nothing more than increasing external stimulus salience and
CONSCIOUSNESS
189
contrast accomplish. It simply "turns up the gain" on neuron action potential

frequency by way of endogenous activity across horizontal connections within
a neural region and feedback projections from processing regions further
upstream. This troubling consequence for consciophiles is by itself philosophically interesting. At the level of individual neurons, this feature of
consciousness is nothing special. That this consequence was garnered by
"single-cell neurophysiology" shows further the potential of "reductionistic"
neuroscience, even for issues in the philosophy of consciousness.
4 SINGLE-CELL NEUROPHYSIOLOGY AND THE

"HARD PROBLEM"
4.1 Chalmers on Easy versus Hard Problems of Consciousness
Now we can expect consciophiles to shed the kid gloves. So far we've
seen how "single-cell" reductionistic neuroscience sheds light on two features
of conscious experience: working memory and explicit selective attention.
Sympathizers of David Chalmers, however, will point out that scientific
explanations of these features of consciousness were never at issue or in
serious doubt. Chalmers explicitly includes on his list of the "easy" problems
of consciousness the following three : "the integration of information by a
cognitive system; the ability of a system to access its own internal states; the
focus of attention" (1995, 200-201). The first two clearly include the kind of
working memory addressed by Goldrnan-Rakic and her colleagues; the third
clearly includes the explanatory target of Maunsell, Gilbert, and their
colleagues. Perhaps it is surprising that reductionistic neuroscience has
already made such headway on these problems. Chalmers did predict eight
years ago that "getting the details right will probably take a century or two of
hard empirical work" (1995, 201). But hey, science is full of surprises of this
sort. The key point is, we knew going in that those problems of consciousness
were solvable by science.
This contrasts with "the really hard problem of consciousness .. . the
problem of experience. When we think and perceive, there is a whir of
information processing, but there is also a subjective aspect. ... This subjective
aspect is experience. ... What unites all of these states is that there is
something it is like to be in them. All of them are states of experience"
(Chalmers 1995,201). What separates the easy problems from the hard one is
that subjective experience isn't exhausted by its functional properties and so
can't be "reductionistically explained.t" But reductive explanation is the
broad method of the special sciences (sciences other than basic physics),
190
including cogmtive science, neuroscience, molecular biology, molecular

genetics, and biochemistry. As these sciences stand, "they are only equipped
to explain the performance of functions. When it comes to the hard problem,
the standard approach has nothing to say " (1995 , 204) .
I leave it to readers to decide if what I will offer in the rest of this
chapter is a step toward solving Chalmer's "hard problem." My guess is that
committed mysterians about consciousness will deny that it is; but such
denials will reveal the "pragmatic fruitlessness" of the hard problem, and the
neo-Carnapian attitude I expressed in Chapter One bids us to ignore those.
Over this section and the next three I will describe empirical progress that has
been made in "inducing phenomenology" by manipulating tiny clusters of
sensory neurons using a standard technique from traditional neurophysiology;
suggest that this is a start toward explaining the neural basis of (sensory)
qualia; draw out lessons from this research for an increasingly prominent
position in the philosophy of consciousness, "phenomenal externalism"; and
end by arguing that philosophers aren't the sole proprietors of "the problems
of qualia and subjective experience"-that mainstream reductionistic neuroscientists have already staked a claim on them. I hope that readers not yet
committed to the inexplicable nature of qualia and subjective experience will
give this scientific evidence a fair reading.
There are two projects being pursued in this chapter. The positive
project presents current scientific hypotheses and experimental results from
cellular neuroscience about properties thought by many to be connected with
conscious experience. The negative project is an implicit response to
mysterians. The first is the project that most excites me, since it describes
ongoing attempts to address features of consciousness scientifically. The
empirical evidence shows that this scientific endeavor is progressing with
demonstrated results. That this evidence can now be drawn from ruthlessly
reductive cellular neuroscience is all the more intriguing, since few
philosophers and cognitive scientists are even aware that these results exist.
This final point is where the positive project of this chapter hooks up with the
implicit jab at mysterians. Mystery thrives in ignorance; only now the
ignorance is self-imposed because relevant empirical data is out there, in the
cellular and molecular neuroscientific journals."
4.2 Neuroscientific background: Wilder Penfield's pioneering

use of cortical stimulation
Four decades ago , neurosurgeon Wilder Penfield published a
comprehensive review of "experiential responses" elicited by electrical
stimulation of the cortex in awake humans (Penfield and Perot 1963). He
CONSCIOUSNESS
191
perfected this technique as part of a surgical procedure for treating otherwise

intractable epilepsy. Since brains lack pain receptors, patients whose scalps,
skulls and underlying connective tissue had been deadened with local
anesthetics could comfortably remain conscious while surgeons ablated
(removed) the site(s) of their seizure origins . Penfield's procedure was a
clinical breakthrough. If electrical stimulation at a specific site evoked
epileptic symptoms, this was evidence that the site is one of seizure origin .
And by probing respon ses of conscious patients during announced and
unannounced stimulations, the surgeon could explore the functional
significance of tissue he was considering removing.
Penfield and Perot's comprehensive review reports that from 1938 to
1963, Penfield and his associates at the Montreal Neurological Institute
performed 1,288 surgeries on 1,132 patients. 520 cases involved exposing and
exploring the temporal lobe; 612 involved other neural region s. Electrical
stimulation produced "experiential responses " in none of the latter 612 cases ,
while it did so in 40 of the former 520 cases (7.7%). Experiential responses
were more complex than sensory experiences like whirring or buzzing sounds
and color flashes or motor phenomena like involuntary limb movements. The
latter were elicited routinely in many patients by electrical stimulation to
appropriate sensory or motor cortical regions (Penfield and Perot 1963, 597).
True experiential responses instead resembled the spontaneous "experiential
hallucinations" and "dreamy states" characteri stic of temporal lobe epileptic
seizures. The ones surgeons induced by electrical stimulation were
"sometimes extensive and elaborate, sometimes fragmentary ," and often
included "the sights and sounds and the accompanying emotions of a period
of time, and the patient usually recognizes it spontaneously as coming from
his past" (1963 , 596). Auditory responses were most frequent, including a
voice or voice s, music, or other meaningful sounds. Experiences of music
were surprisingly prominent. Visual responses were also frequent , often of a
person or group of persons, a scene, or other recognizable objects. Auditoryexperiential and visual-experiential responses sometimes occurred in combination, usually as scenes with appropriate sound s or a person or people
singing or talking. In patients who commonly suffered from spontaneous
experiential hallucinations during their seizures, electrically invoked
experiential respon ses often resembled their spontaneous hallucinations.
Experiential responses elicited from one site were often identical or similar to
responses elicited from nearby sites.
Examples from the published transcripts of their forty case historie s
illustrate these features .lO After removing the anterior tip of D.F .' s right temporal lobe (Case 5, 619-620), the surgeon stimulated a site on the cut surface
of the superior and medial region of the first temporal convolution . On the
second stimulation D.F. reported , "I hear some music ." When the stimulation
192
was repeated without warning, D.F. reported, "I hear music again. It is like
the radio." She was unable to name the tune, but claimed it was familiar.
Upon a later stimulation to this same site, D.F. reported, "I hear it." The
electrode was kept in place and D.F. was asked to describe her experience.
She hummed the tune. The operating room nurse named the tune and D.F.
agreed with her judgment. The nurse agreed that D.F.'s humming captured the
tune's proper timing and tempo. On further inquiry D.F. claimed that the
experience was not that of "being made to think about" the tune , but that she
"actually heard it."
E.c. had a history of "psychic precipitations" that always ended in
seizure (Case 19, 632-633). Each of his attacks began after he saw someone
grab an object from another person. The visual perception would produce a
vivid memory of a time when he was thirteen, playing with his dog by
grabbing a stick from the dog 's mouth and throwing it. This association would
confuse him and produce a seizure. During stimulation of a site just superior
to the first temporal convolution in his left hemisphere, while E.C. was
naming pictured objects, he reported, "There he is ... It was like a spell. He
was doing that thing : grabbing something from somebody. It was somebody
else doing the grabbing." When asked what he was grabbing, E.c. replied, "A
stick, or something." When asked where he was, he replied "Up the street . ..
That was like an attack, doing that thing." When the surgeon returned to
stimulate this site again ten minutes later without warning , E.C. reported
"There it is." The stimulating electrode was kept in place a short time longer
and a major seizure ensued.
R.Re. , a native South African, began suffering seizures eight years
after recovering from a severe case of meningitis (Case 14, 628-629). During
surgery a number of anatomically proximal sites were stimulated above the
first temporal convolution in his left hemisphere, from the anterior tip to a
point adjacent to the central sulcus . R.Re . offered the following sequence of
reports : "Yes, something that someone has said ... Not here, in Johannesburg."
"Yes, something that was said , also something that was said in Johannesburg ,
and it was said by somebody that had been put out." "Yes, I was hearing at
Johannesburg, it came and went very clearly." "Yes, that same sort of
sensation, somebody was speaking to me in Johannesburg." He mentioned
that the speaking voice was different each time.
Based on their case histories and analysis, Penfield and Perot (1963)
offered a number of clinical and neuropsychological conclusions. Some speak
directly to the issue of phenomenology induced by dire ct cortical electric
stimulation. For example, they write
The conclusion is inescapable that some, if not all, of these
evoked responses represent activation of a neural mechani sm
CONSCIOUSNESS
193
that keeps the record of current experience. There is

activation too of the emotional tone or feeling that belonged
to the original experience. The responses have that basic
element of reference to the past that one associates with
memory. But their vividness or wealth of detail and the sense
of immediacy that goes with them serves to set them apart
from the ordinary process of recollection which rarely displays such qualities. (1963 , 679)
They emphasize the connection between evoked memories and the "stream of
consciousness" during past experience:
The true nature of [epileptic experiential] hallucinations
becomes quite clear when the records of the stimulation
responses are studied . ... They are reproductions of past
experience. ... At operation it is usually quite clear that the
evoked experiential response is a random reproduction of
whatever composed the stream of consciousness during some
interval of the patient's past waking life. (1963, 686-687)
Near the manuscript's end they conclude: "There is within the adult human
brain a remarkable record of the stream of each individual 's awareness or
consciousness. Stimulation of certain areas of cortex, lying in the temporal
lobe between the auditory sensory and the visual sensory areas , causes
previous experience to return to the mind of the conscious patient" (1963,
692).
That documented cases of phenomenology with sensory qualia and
subjective feel induced by cortical stimulation exist at all is interesting and
relevant for questions about the neural basis of phenomenal consciousness.
But philosophical caution is appropriate here . Penfield and his associates were
able to elicit experiential responses in only a minority of temporal lobe
epileptics. And these limited result s were in epileptic brains, near sites of
seizure origin , where electrical activation was admittedly "facilitated" by the
organic damage. The evoked experiences were limited not only to (long-term)
memory items, but also only to certain types of memories. In the paper' s final
section , Penfield and Perot (1963) run through a list of memory experiences
that were never invoked. Finally, their evidence does not even support the
localization of these memory experiences to the site of stimulation. As
Penfield and Perot note explicitly, during subsequent interviews days or
weeks after their surgeries, patients could recall the experiential responses
evoked, even when the site of stimulation had been ablated at a later stage of
the surgical procedure! In terms of the neural "location" of subjectively
194
accessible "engrams," the most that these results show is that regions in the
temporal cortices, especially ones superior to the first temporal convolution,
"play in adult life some role in the subconscious recall of past experience,
making it available for present interpretation" by "activating connections with
that part of the record of the stream of consciousness in which hearing and
seeing are the prominent components" (Penfield and Perot 1963, 689). But
Penfield's results were only the beginning. Recent results from three primate
research labs carry us further toward induced phenomenology and lessons
about the neural basis of sensory qualia and subjective experience.
5 INDUCING PHENOMENOLOGY FROM VISUAL

MOTION TO SOMATOSENSORY FLUTTER... AND
BEYOND?
5.1 Results from William Newsome's lab
Area MT (Middle Temporal cortex) in primates (including humans) is
the gateway to the "dorsal" visual processing stream (Figures 4.2, 4.3 above).
As discussed earlier in this chapter, this stream proces ses information about
objects' locations and motion leading to actions guided by vision. Both lesion
studies and electrophysiological recordings in nonhuman primates have
revealed MT' s specific role in visual judgments about motion direction. Most
MT neurons are direction selective, spiking at highest frequency to a visual
stimulus moving in a single direction in their receptive field s, a bit less
frequently to related directions, and not at all (above baseline spiking rates) to
motions unrelated to their preferred direction.!! Like many cortical regions,
MT has a columnar organization. Neurons in a given vertical MT column
share similar receptive fields and motion selectivity. These features vary in
neurons from column to column, and MT in its entirety realizes a "map" that
represents all motion directions at all regions of the visual field (Albright et
al. 1984).
Will iam Newsome and his collaborators developed a measure that
quantifies the strength of a motion stimulus (Salzman et al. 1992) (Figure 4.5
below) . A pattern of dots appears on a computer screen . The strength of a
motion stimulus, expressed as a "percentage correlation," reflects the percen tage of dots that are re-plotted on subsequent screens at a fixed spatial interval
and direction from their original position. All other dots are re-plotted at the
same spatial and temporal intervals but in random directions from their
original positions. This re-plotting and the temporal interval between the
screens gives the illusion of a motion stimulus, with some percentage of the
CONSCIOUSNESS
195
dots appearing to move in one direction and the rest appearing to move in
random directions. For example, in a "50 % correlation vertical stimulus," half
of the dots on the original screen are re-plotted on later screens at a fixed
upward interval, providing the illusion of vertical motion, while the other half
are re-plotted randomly.
No Correlation
50% Correlation
100% Correlation
Figure 4.5. Newsome and his colleagues ' measure of motion stimulus strength in terms of
percent correl ation of dots appearing to move in a particular direction (the rest appearing to
move in random directions). Reprinted with permiss ion from Salzman et al ., figure I, 2333,
copyright 1992 by the Society for Neuroscience.
Newsome's lab also developed a behavioral paradigm in which rhesus

monkey s express judgments about motion direction. Their full litany of
controls is elaborate but the basic idea is straightforward. The monkey fixates
on a central point on a computer screen display and maintains fixation while a
visual motion stimulus of a particular strength is presented (a particular
percentage correlation in some direction). Both the fixation point and the
motion stimulus are extinguished and target lights (LEDs) appear at the
screen's peripheries. The "preferred" (Pref) LED is located in the direction
(from the original fixation point) of the motion stimulus; the "null" LED is
located in the opposite periphery. The monkey indicates its judgment of
stimulus motion direction by saccading (moving his eyes quickly) to one of
the LEDs . His saccade is his report of apparent (perceived) motion direction.
The monkey is only rewarded when he saccades correctly: to the Pref LED in
the direction of the percentage correlation motion stimulu s. Using standard
single-cell electrophysiological recording procedures, Newsome's group first
locates an MT neuron's receptive field and preferred motion selectivity. A
percentage correlation motion stimulus is then presented only to that neuron's
receptive field (as the monkey maintains fixation on the central point) . They
can then compare the monkey's report about stimulus motion direction acro ss
196
differing strengths (percentage correlation) when electrical stimulation is

applied to that neuron through a stimulating electrode and when it is not.
Penfield and his associates induced electrical stimulation through a
monopolar silver ball electrode with an area of cortical contact approximately
1.5 square mm. Their typical stimulus was a square wave pulse, 2-5 milliseconds in duration, at 1 to 5 volts and a frequency of 40-100 Hz (cycles per
second). The resistance was 10,000-20 ,000 ohms, yielding a stimulus current
that varied between 50-500 milliamperes (Penfield and Perot 1963, 602).
Thus vast numbers of neurons were stimulated on a single trial. Newsome's
lab microstimulates MT neurons using tungsten microelectrodes with an
exposed tip length of 20-30 microns. Stimulating pulses are biphasic, each
with an 0.2 millisecond duration, with frequency of either 200 Hz or 500 Hz,
producing a current of 10 microamperes in amplitude. In one of their
publications, Newsome's group reports data from primate motor cortex that a
single cathodal 10IlA current pulse directly activates neurons within 85
microns of the electrode tip. The number of neurons directly stimulated by the
electric current is thus many orders of magnitude smaller than the number
directly activated by Penfield's electrodes and pulses .
The percentage correlation measure of stimulus motion strength and
their behavioral paradigm permit Newsome's group to plot the proportion of
monkeys' reports of apparent motion in stimulated MT neurons' preferred
direction as a function of stimulus motion strength. (With MT microstimulation , the "Pref" response is defined as the monkey's judgment that the
external stimulus is moving in the MT neuron cluster's preferred direction.)
Figure 4.6 represents a monkey 's performance with a choice bias slightly in
this neuron cluster's preferred direction of motion stimuli. Dots and the
sigmoid regression line drawn through them represent the monkey's
performance in the absence of electrical microstimulation. When even a small
percentage of the dots appear to be moving in this cluster's preferred direction
(e.g., > 20% correlation), the monkey correctly judges motion in the preferred
direction on nearly every trial (1.0 Proportion Preferred Direction (PD)
judgment). When a moderate percentage of the dots appear to be moving
opposite this cluster's preferred direction (e.g. , < -50% correlation), the
monkey correctly judges motion in the null direction on nearly every trial (0.0
Proportion PD). If microstimulation to this direction-selective MT neuron
cluster adds signal to the neural processes underlying visual judgments of
motion direction, then it will bias the monkeys' reports toward the stimulated
neurons' preferred direction. When graphed, this would produce a leftward
shift of the psychometric function (Figure 4.6, line A). The monkey then will
be more prone to judge motion in the Pref direction, even when fewer of the
dots actually appear to be moving that way. If microstimulation adds noise to
the neural processes underlying motion judgment, this will exacerbate the
197
CONSCIOUSNESS
monkey 's choice bias. When graphed, this would produce nearly constant
judgments around the y-intercept of the orig inal function at 0% correlation,
with only a slight increase for highly correlated preferred stimu lus direction
and a slight decrease for highly correlated null stimulus direction (Figure 4.6,
line B) .
0...
A /
co
0
to
a
0aL
0...
___ _ -I-f
-4-
/'
-.:t
N
0
I
I
/
0
0
;'
-60 - 40 -20
20
40
60
Correl a tion (%)

Figure 4.6. Psychomet ric function relating the proportion of monkey's responses to the MT
neuron cluster's preferred motion direction to the percent correlation of the visual motion
stimulus. A. If microstimulation to the cluster adds signal to neuronal activity. psychometric
curve will be displaced to the left, indicating a higher propo rtion of responses in the neurons'
preferred motion direction to the same strength of visual motion stimulus. B. If
microstimul ation adds noise, the curve will be replaced by a line nearly perpend icular to the xaxis near the y-intercept value at 0% correlation, with a slightly higher percentag e of
"preferred" respon ses to exceptionally strong visual stimuli in the preferred direction and a
slightly lower percenta ge of "preferred" respon ses to exceptionally strong visual stimuli in the
"null " direction . See text for full explanation. Reprinted with permission from Salzman et al .,
figure 3, 2335 , copyright 1992 by the Society.for Neuro science.
Newsome and his colleagues continually ob served the "adds signal"

results of microstimulation to direction -selective MT neuron clusters, under a
variety of percentage correlations (stimulus strengths) and microstimulation
198
frequencies (Salzman et al. 1992, especially Figures 4 and 8; Murasugi et al.

1993, especially Figures 2 and 5). At nearly every percentage correlation,
microstimulation of a direction-selective MT neuron cluster biased the
monkeys' saccades significantly to the Pref LED. This bias occurred even in
the presence of strong motion stimuli in the other (null) direction (e.g., > 50% correlation). Recall that monkeys are only rewarded when they report
stimulus motion direction (percentage correlation) correctly. They never
receive a reward for their continually incorrect choices under conditions of
actual motion stimuli in the null direction and applied microstimulation.
Increasing microstimulation frequency (up to 500 Hz) increases the proportion of motion reports in the neurons' preferred direction.
These results lead naturally to the question of what the monkeys are
seeing-experiencing-during microstimulation trials? Are monkeys conscious of motion in the microstimulated neurons' preferred direction, even
when the actual motion stimulus is strongly in the opposite direction?
Newsome and his colleagues admit that their experiments with monkeys
cannot answer this question conclusively. But they don't shrink from offering
suggestions, writing: "A plausible hypothesis is that microstimulation evokes
a subjective sensation of motion like that experienced during the motion
aftereffect, or waterfall illusion .... Motion therefore appears to be a quality
that can be computed independently within the brain and "assigned" to
patterned objects in the environment" (Salzman et al. 1992, 2352) . They are
suggesting that visual motion qualia are generated in the brain and attached to
internal visual representations of external objects. Happily, in ordinary
circumstances, our "internal assignments" of features to representations tend
to correlate with features and relations of the objects represented. Natural
selection was crueler to creatures whose "internal assignments" were more
haphazard or skewed. But under appropriate conditions, our internally generated and assigned qualia and the external environmental features can be
dissociated. That's what happens , apparently, in Newsome's MT mincrostimulation-motion studies.
Rodolfo Llinas and Patricia Churchland call the general idea behind
this suggestion endogenesis. They explain: "sensory experience is not created
by incoming signals from the world but by intrinsic, continuing processes of
the brain" (Llinas and Churchland 1996, x). Incoming signals from sensory
receptors keyed to external physical parameters serve only to "trellis, shape,
and otherwise sculpt the intrinsic activity to yield a survival-facilitating, mein-the-world representational scheme" (ibid.) . Natural selection-and hence
adequacy for exploiting an available environmental niche, not truthdetermines a scheme's success.
Microstimulation motion effects are not specific to nonhuman
primates. Newsome and his colleagues remark that "it has recently been
CONSCIOUSNESS
199
reported that crude motion percepts can be elicited with electrical stimulation
of the human parietal-occipital cortex" (Salzmann et al. 1992, 2354) . Their
results are continuous with Penfield's pioneering studies (reported in the
previous section). Nor are microstimulation effects specific to visual motion .
Newsome and his collaborators have also succeeded in affecting judgments of
stereoscopic depth. One important depth cue for creatures with overlapping
visual fields (like primates) is binocular disparity, small differences in
position between images of an object formed on the two retinas. Neurons
responding selectivity to binocular disparity have been found throughout
primate visual cortex, including in area MT. There they are found in clusters,
typically 200-300llm in diameter, surrounded by neurons with little or no
disparity selectivity. Each neuron in a cluster responds optimally to a common
range of preferred binocular disparities. By finding such clusters using
standard microelectrode search techniques and then inducing electric
microstimulation into their centers, Newsome and his collaborators have been
able to affect judgments of binocular disparity similar to those they elicited on
judgments of motion direction (DeAngelis et al . 1998).
Monkeys begin the depth-discrimination task by fixing their gaze on a
point in the center of a computer screen display. A random-dot pattern is
presented within a circular aperture roughly the size and location of the
multiunit receptive field of the MT neurons to be microstimulated. The dots,
arranged as red/green anaglyphs viewed at the appropriate distance through
red/green optical filters, give the illusion of depth at particular degrees of
horizontal binocular disparity.l'' Similar to their measure of motion stimulus
strength, Newsome and his collaborators measure depth stimulus strength in
terms of percentage of binocular disparity correlation. Some percentage of the
dots (the "signal") shown to the right eye (through the colored filter) are
displayed with a dot shown to the left eye to produce identical degrees of
horizontal disparity and the illusion of constant depth (either closer or further
away than the fixation point); the rest of the dots C'noise") in the receptive
field aperture are presented to produce random disparity. Monkeys report
judgments of near or far depth by saccading to an appropriate LED. A trial
begins when the fixation point appears and the monkey fixes his gaze. Dots
then appear in the receptive field aperture to reflect some strength of depth
stimulus. On half the trials, selected at random, microstimulation to the
disparity-selective MT neuron cluster begins when the visual stimulus
appears. After one second the visual stimulus is extinguished (along with
microstimulation, if it occurs on the given trial) and the LED target lights
appear. The monkey saccades quickly to the appropriate target to reflect his
judgment of the stimulus's near or far disparity. Rewards are given on all and
only the trials when the monkeys' response correctly indicates the depth
stimulus ("signal").
200
Just as they found with MT microstimulation and motion stimuli,

Newsome and his collaborators found that microstimulation adds signal to
monkeys' judgments about stereoscopic depth . In one MT cluster tuned to
"near" disparity displays, at ~ 50% binocular correlation-where one-half or
more of the "signal" dots are at the neurons' preferred degree of horizontal
disparity (i.e., the degree that generates maximal spiking frequency in these
neuron s under normal viewing conditions)-and no microstimulation,
monkeys gave a correct judgment (saccaded toward the "Near target " LED)
on roughly 80% of the trials. Similarly, at ~ -15 % correlation, when 15
percent or more of the signal dots were opposite these neurons' preferred
disparity, and no microstimulation, monkeys correctly saccaded away from
the "Near target" LED on every trial. However, with microstimulation to this
"near-preferring" cluster, monkeys gave "Pref" judgments on 80% of the
trials with -30% binocular correlation-when thirty percent of the dots
displayed disparity opposite to the microstimulated neuron s' preferred
disparity and the rest displayed random disparity. "Preferred" judgments
reached 100% with microstimulation on trials with 0% horizontal disparity
(i.e., when every dot displayed random disparity). This was a robust and not
atypical result (DeAngelis et al . 1998, Figure 3b, d, p. 679). New some and his
collaborators report 43 statistically significant differences on microstimulation
versus no microstimulation trials in 65 depth disparity experiments on two
monkeys. On 42 of these 43 case s, microstimulation biased monkeys' depth
disparity judgment toward the stimulated cluster's preferred depth disparity
(DeAngelis et al. 1998). Statistically significant results were limited to
clusters that displayed moderate to strong disparity selectivity. As before, the
intriguing suggestion is that activity in MT depth -selective neurons generates
visual depth qualia that get "assigned" to representations of external visual
stimuli . Subjective phenomenal conscious experience with depth qualia seems
to be induced by highly specific electric rnicrostimulation to tiny clusters of
depth-selective neurons in area MT.
5.2 Results from Kenneth Britten's lab

Nor are microstimulation effects specific to only one lab. Changing
patterns of visual motion on the retinas as we move through space, called
"optic flow," prov ides a rich source of information about direction of selfmovement or "heading." Earli er work with rhesus monkeys indicated that the
medial superior temporal area (MST) (Figure s 4.2 and 4.3 above) contains
neuron s selective for optic flow information and for stimuli that simulate the
visual effects of self-motion. Neurons tuned to leftward or rightward heading
direct ions are arranged in clusters spanning region s up to 500llm. Britten and
CONSCIOUSNESS
201
van Wezel (1998) presented rhesus monkeys with visual displays that simulated a cloud of dots at a visual depth from I-10m. All dots were re-plotted at
a particular angle and distance to provide the illusion of self-motion through
space at a particular angle and direction ("heading") (Figure 4.7A, B). A trial
begins when a red fixation point in the center of the visual display is
illuminated and the monkey fixes its gaze upon it. Throughout the trial's
duration, the fixation point either remains at that location or moves at a
constant velocity to the left or right. The dot field appears shortly after the
fixation point. Dot re-plotting begins immediately to simulate optic flow in a
specific leftward or rightward horizontal heading angle. 0 corresponds to
"directly ahead," negative degrees correspond to leftward heading, and
positive degrees correspond to rightward heading. On half of the trials
microstimulation at 20flA amplitude, 200 Hz frequency , begins at the same
time that the dot field appears. Stimulating electrodes have been inserted into
the center of MST clusters of at least 250flm in which all neurons are tuned to
a similar leftward or rightward heading direction and angle . One second later
the dot field is extinguished, microstimulation ceases (if it occurs on the trial) ,
and target lights are illuminated, one in the angle and direction of the heading
stimulus, the other at the same angle but opposite heading direction. Monkeys
are rewarded if they saccade to the target in the direction of the heading
stimulus. The mov ing fixation point s on some trials force s the monkey to
make smooth pursuit eye movements during the stimulus period . MST
neurons are known to display activity correlated with smooth pursuit eye
movements and appear to compensate their heading tuning for distortions in
optic flow produced by smooth pursuit.
Once again, results indicate that microstimulation adds signal to the
neural processes underlying judgments of heading direction. In one experiment , the proportion of rightward heading choices made by the monkey was
plotted against degree of horizontal heading presented in the dot field . The
neurons in the region of the stimulating electrode tip were selective for
rightward optic flow and leftward heading. This monkey exhibited a baselin e
bias, perceiving "directly ahead" at roughly _5 (to the left) of the display 's
geometric center. When this cluster of leftward heading -tuned MST neuron s
was microstimulated and the visual display indicated leftward heading at _4
(which to this monkey correlated to +1 rightward heading due to his baseline
bias), the monkey judged that the heading was rightward only slightly more
than 1 in 4 trials; without microstimulation to the leftward heading-tuned
neuron cluster under these same viewing conditions, the monke y judged
rightward heading in 8 of 10 trials. At _1 (leftward) heading stimulus (which
to this monkey correlated to +4 rightward heading), microstimulation to the
202
Simulated Movement

...
:

.
.:.
.
.
..
. e._.
.. .
...
.. . . .
..
.. .- ... .....- .
:0\:/:: :
Observer View
Figur e 4.7. Heading direction visua l stimulus in Britten and Wezc l's MST micro stimulation
studies . A. The simu lated "virtual" depth stimulus and heading direction of the monkey with
resp ect to it, as seen from above. B. Appe arance of left- and right -head ing stimu lus, as seen
from the observer's perspective . Reprinted from Nature Neuro scien ce 1, K. Britten and R. van
Wezel, copyright 1998, pages 59-63, with perm ission from Nature Publishing Grou p.
leftward head ing-tun ed neuro n cluster dropped the mon key 's j udgment of
right ward heading direction from 100% (without microstimulation) to j ust
below 60%, only slight ly above chance . Res ults were eve n more dram at ic
whe n the fixation point moved leftward du ring the dot field display, forc ing
the monke y into leftward smoo th pursuit eye movements. Eve n whe n the dot
fie ld indicated a right ward head ing of +4, when without microstimulat ion the
monkey indicated right ward headi ng on 75 % of the trials, micro stimulation of
this leftward head ing-tun ed MST c luster biased the monkey to choose the
leftwa rd headin g target dot in 90% of the trials (Bri tten and van Wesel 1998,
Figure 2, p. 60). As in Newsome's studies, mo nkeys are never rewarded for
incorrec t choices of act ual headin g stimulus. Here we have the visual
CONSCIOUSNESS
203
phenomenology of self-motion through space seemingly induced by highly

specific electrical microstimulation. Heading direction in a cloud of moving
dots giving the illusion of 10m in depth is complex visual phenomenology
indeed to induce by stimulating a cluster of neurons between 250-500/-lm,
especially under conditions of opposite visual presentation ."
5.3 Results from Ranulfo Romo's lab

Nor are microstimulation effects limited to visual stimuli. Ranulfo
Romo and his collaborators trained rhesus monkeys to distinguish differences
in frequency between two flutter stimuli delivered to a fingertip site. Humans
report sensations of "flutter" when mechanical vibrations between 5-50 Hz
are applied to the skin. Such stimuli activate neurons in primary somatosensory cortex (area 3b of Sl , see Figure 4.2 above) whose receptive fields
include the stimulation site. "Quickly Adapting" (QA) neurons are strongly
activated by periodic flutter vibrations and fire with a probability that
oscillates exactly at the input frequency. In other words, their mean firing rate
correlates directly with the frequency of the mechanical vibration applied to
their receptive fields (appropriate portions of the skin) (Mountcastle et al.
1990). These neurons are also arranged in columnar clusters that share similar
receptive fields . All these properties make them a convenient target for
microstimulation studies.
In their first study (Romo et al. 1998), a trial began when a
mechanical probe was lowered onto a monkey's restrained hand , indenting
slightly the glaborous skin of one fingertip (in lay terms, the fingerpad). The
monkey then places its unrestrained right hand onto an immoveable key
within 1 second. After a brief delay period (1.5-3 seconds) the mechanical
probe oscillates at the "base" frequency for 500 milliseconds. This is followed
by a brief delay (1-3 seconds), after which either a second mechanical
stimulus (the "comparison") at either a higher or lower oscillatory frequency
than the base stimulus is delivered to the restrained hand or microstimulation
at a frequency that corresponded to a higher or lower mechanical stimulus is
delivered to the QA neuron cluster in S 1 whose multi -unit receptive field
includes the stimulus site. The monkey indicates detection of the end of the
"comparison" frequency (mechanical or cortical microstimulation) by
releasing the key within 600 milliseconds, and whether the "comparison"
frequency was higher or lower than the "base" by pushing one of two buttons
with its free hand. Monkeys are rewarded on all and only trials in which their
comparative judgment about whether the "comparison" stimulation was of
higher or lower frequency than the "base" is correct. By comparing performances on mechanical versus microstimulation "comparison" stimuli, Romo
204
and his collaborators sought to discover "whether the animals could interpret
the artificial signals [microstimulation "comparison" frequencies] as flutter"
(Romo et al. 1998, 388) .
As long as "base" and "comparison" stimuli consist of two current
pulses with amplitude> 651lA, monkeys achieve over 75% correct for both
mechanical and microstimulation "comparison" frequencies . This result
obtains even when the "comparison" frequency differs from the "base" by
only 8 Hz. They show no statistically significant differences at these frequencies whether the comparison frequency is actual mechanical stimu-Iation
to the fingertip or cortical electrical microstimulation. When the base
frequency is held constant over trials at 20 Hz, monkeys make correct
judgments about comparison frequencies better than 75% of the time when
these frequencies are s, 15 Hz or 2 25 Hz, with no statistically significant
difference between mechanical stimuli and microstimulation (Romo et al.
1998, Figure 2, p. 388). Romo and his collaborators conclude: "The monkeys
were consistently able to extract the comparison frequency from the
artificially inducedsensation" (Romo et al. 1998,388).
What about our issue of "induced phenomenology"? Romo et al. are
not timid here, writing:
Animals continuously switched between purely mechanical
and microstimulation conditions with almost identical performance levels . Such high accuracy, based on the interaction
between natural and artificially evoked activity, is consistent
with the induction of a sensory percept. ... Thus the
microstimulation pattern s used may elicit flutter sensations
referred to the fingertips that are not unlike those felt with
mechanical vibrations. (Romo et al. 1998,399-390)
They also report that on seven experimental runs during this study, they
induced a combined mechanical and microstimulation comparison stimulus.
The mechanical component had a lower frequency than the base stimulation
while the microstimulation was higher. Despite the actual stimulus induced at
the fingertip and no reward given for "incorrect" mechanical judgments,
monkeys judged the (combined) comparison stimulus as higher frequency in
348 of 400 such trials .
In a subsequent study, Romo and his collaborators reversed the base
and comparison stimuli from their original study (Romo , et al. 2000) . All
other parameters remained as described above , but now on half the trials the
base stimulu s was mechanical vibration at the fingertip while on the other half
it was instead cortical microstimulation to area 3b in S 1 (again to QA neuron
clust ers whose receptive fields contained the fingertip stimulation site). The
CONSCIOUSNESS
205
comparison stimulus was mechanical stimulation at either a higher or lower

frequency . In their previous study, monkeys had to be capable of comparing
the result of microstimulation with the base frequency, with the latter
represented and stored in working memory through normal means (beginning
with actual mechanical stimulation of the fingertip). In this second study,
however, if the monkeys are to succeed with microstimulation "base" stimuli
comparable to their performance with actual mechanical "base" stimuli, then
cortical microstimulation alone must engage the entire range of cognitive
processes involved, from sensation through working memory and comparative
decision-making. So besides its intrinsic scientific interest, this study is
doubly relevant for our question of induced conscious experience because of
the tight connection many recognize (and I stressed in sections 1 and 2 of this
chapter) between working memory and consciousness.
Even when base and comparison frequencies differed by as little as 4
Hz, monkeys were able to respond correctly on 75% of the trials about which
frequency was lower. There were no statistical differences between performances on trials with mechanical stimulus or microstimulation base. As a
control, Romo and his collaborators tested monkeys when both base and
comparison frequencies were microstimulations alone , comparing results to
cases of similar base and comparison frequencies where both were actual
mechanical stimuli. In the former cases, there were no actual mechanical
stimulations to induce sensory, working memory, or comparative decisionmaking processes in the normal fashion. Yet monkeys performed nearly
identically in the two types of cases (although there was more variance within
sessions with the purely artificial base and comparison stimuli).
Is there a foreseeable limit on this latest wave of cortical
microstimulation studies? Newsome and his collaborators have long sought
similar results with color stimuli . In their first paper reporting microstimulation results with visual motion stimuli, they write:
A natural extension of this work is to apply the same basic
approach to the study of circuits that mediate aspects of
visual perception other than motion. In principle, the
microstimulation technique is applicable to the analysis of
function in any circuit in which neurons with similar
physiological properties are segregated into columns or large
clusters. ... Given present physiological knowledge, appropriate candidates for future investigation are circuits that
encode orientation, color, and disparity. (Salaman et al. 1992,
2353; my emphasis)
206
However, color continues to pose formidable technical difficulties (Newsome,

personal correspondence). Going beyond sensory stimuli , Liu and Newsome
(2000) recently have raised the possibility of microstimulating the appropriate
neurons involved in the working memory and comparative decision-making
aspects of tasks like Rome's. We saw in sections I and 2 above that cells with
"working memory fields," that fire during short delay periods (up to 12
second) , have been found in prefrontal cortex. Might microstrimulation to
clusters of cells sharing working memory fields and properties induce
causally efficacious "working memories"? As Liu and Newsome put this
question, "might it be possible to influence or change the monkey's memory
by electrically stimulating such neurons?" (2000, R600) . Current physiological knowledge has not yet established that these neurons are grouped
anatomically into columns or clusters with others sharing similar activation
properties. However, as Liu and Newsome note, "only a few years ago the
complexity of the cerebral cortex would have led most sensory physiologists
to declare Romo and colleagues' current microstimulation experiments a
fantasy" (2000, R660) . They insist that "for now, all bets are off until the
experiments are actually tried" (ibid .).
There is an additional reason for why these microstimulation results
from Penfield's through Rome's are central to a ruthlessly reductive cellular
neuroscience of conscious experience. I stressed throughout Chapters Two
and Three that the capacity of experimenters to intervene directly on lower
level mechanisms to generate specific, measurable behavioral effects is
crucial for claiming an explanation, and hence a reduction. Microstimulation
procedures accomplish exactly this result for "induced phenomenology."
Experimenters directly activate tiny clusters of stimulus-specific sensory
neurons and generate observable behavior assumed to be guided normally by
visual or somatosensory experiences. These direct cellular manipulations that
induce sensory experience are consistent with the general explanatory demand
recognized in current mainstream neuroscience. Don't simply find neuronal or
intra-neuronal activity correlated with a cognitive task; invoke the specific
behavior by directly manipulating the hypothesized cellular or molecular
mechanisms. 14
6 THE STRANGE CASE OF PHENOMENAL

EXTERNALISM 15
It has been more than twenty-five years since Hilary Putnam first took
philosophers to "Twin Earth." Twin Earth is just like Earth , down to molecule-for-molecule duplicate beings, except for one feature in the linguistic or
physical environment. In Putnam's (l975b) original fable, the single
CONSCIOUSNESS
207
difference was water's molecular structure. On Earth, water is H 20 ; on Twin

Earth, the clear, tasteless liquid in rivers, streams, swimming pools , rain
showers, drinking fountains , etc., is XYZ. Putnam sought to show that
"meanings ain't in the head." Since meaning determines reference, your
utterances of "water" refer to H 20 , and Twin-your utterances refer to XYZ,
the homonyms must have different meanings. But nothing is different inside
your and Tw in-your heads , so whatever individuates meanings ain't there."
Twin Earth's legacy in philosophy is equally legendary. Tyler Burge
(1979) developed footnote 2 of Putnam (1975b) into a full-blown theory of
mental content, and philosophical orthodoxy decided that whatever
individuates mental content "ain' t in the head," either. Content externalism,
advocating wide content, became the rage. Features external to cognizers
"individuate" their mental contents. Content "extends out into the world." In
the wake of the consciousness craze that swept philosophy throughout the
1990s, it was only a matter of time until prominent philosophers of mind
extended content externalism and the Twin Earth fanta sy to phenomenal or
qualitative content.
Phenomenal externali sm holds that the environment external to an
individual's receptor surface "individuates" the qualitative content (qualia) of
his or her sensory experiences. Features of the external environment, literally,
distinguish qualitative states from one another (Dretske 1996; Lycan 1996). A
common intuition pump for this view is ... Inverted Earth! First suggested by
Gilbert Harman (1982), the Inverted Earth fantasy reached fruition in Ned
Block's ([1990] 1997) essay. I? Block writes:
Inverted Earth differs from Earth in two respects. First ,
everything has the complementary color of the color on
Earth. The sky is yellow, grass is red, fire hydrants are green,
and so on. I mean everything really has these oddball colors .
... Secondly, the vocabulary of the residents on Inverted Earth
is also inverted: If you ask what color the (yellow) sky is,
they truthfully say "Blue !" ... If you brought a speaker of the
Inverted Earth dialect to a neutral place (with unknown sky
color, unfamiliar vegetation , and the like) and employed a
team of linguists using any reasonable methods to plumb his
language, you would have to come to the conclusion that he
uses 'red ' to mean what we mean by 'green,' 'blue' to mean
what we mean by ' yellow,' and so on. ([ 1990] 1997, 682)
Of course, no saga worthy of Putnam's legacy could stop here. The
next step is ... Secret Transplanetary Relocation! Block cont inues:
208

A team of mad scientists knocks you out. While you are out
cold, they insert color-inverting lenses in your eyes , and
change your body pigments so you don 't have a nasty shock
when you wake up and look at your feet. They transport you
to Inverted Earth, where you are substituted for a counterpart
who has occupied a niche on Inverted Earth that corre sponds
exactly (except for colors of things) with your niche at home .
You wake up, and since the inverting lenses cancel out the
inverted colors, you notice no difference at all. ([ 1990] 1997,
683)
"What it is like" for you on Inverted Earth, with the lenses and body pigment
alterations, seemingly doesn't change from "what it is like" for you on Earth.
When you look up at the yellow sky, you have a brill iant "Carolina blue "
quale (as they say in Lycan' s and Dretske's current abode). This experience is
just like the quale you would have minu s the lenses if you looked up at the
blue sky from Earth.
However, the phenomenal externalist must balk at the intuition just
expressed. He or she must insist that the qualitative contents of your visual
experiences on Earth and Inverted Earth (with the implanted lenses) must
differ. Dretske (1996) is explicit on this point. He claims that nothing prevents
one who uses Twin Earth intuitions to defend externalism about linguistic
meaning and mental content from using them to defend phenomenal externalism. "Just as we distinguish and identify beliefs by what they are beliefs
about, and what they are beliefs about in terms of what they stand in the
appropriate relation to, so we must distinguish and identify experiences in
term s of what they are experiences of' (1996, 145).1 8 The radical natu re of
Dretske's proposal is apparent in his slogan : "The experiences themselves are
in the head ... but nothing in the head ... need have the qualities that
distinguish these experiences" (1996, 144-145). Earth-you and altered-andtransported-to-Inverted-Earth-you have conscious visual states with different
phenomenal content-different qualia-as you (plural) gaze up into your
(plural) respective skies. Earth-yours is blue; it is an experience of a blue sky.
Altered-and-transported-to-Inverted-Earth-yours is yellow ; it is an experience
of a yellow sky.
Like many philosophical fantasi es that began with good intentions,
Inverted Earth has spawn ed a host of confusing philosophical exoti ca. Block
([1990] 1997) and Lycan (1996) invented "Strange Qualia" and "New Strange
Qualia," to name just two, and I'm sure that more are on offer in this growing
philosophical literature. In their review of Lycan (1996), Tom Polger and
Owen Flanagan (2001) liken the qualia literature spawn ed by Block's and
Lycan 's exchanges over Inverted Earth to a massive crash during an car race
CONSCIOUSNESS
209
that continues on at full speed. They warn: "For those who can identify
working bits among the mangled wreckage, there are some excellent parts for
the taking. But try not to get hurt" (2001, 120-121) . Lycan doe sn't balk at this
assessment, writing in his reply that "Polger and Flanagan ' s vivid description
of the Block-Lycan imbroglio over Strange Qualia and New Strange Qualia
recalls exactly what it felt like to be embroiled in it, especially the part about
trying to describe the accident as we saw it on all the monitors and, of course,
fixing blame" (2001, 130-131 , footnote 1).19 At the risk of oversimplifying a
complicated philosophical matter, my diagnosis for why this discussion
imploded has to do with the fanta stical nature of the Inverted Earth fantasy.
Garbage in, garbage out, as the saying goes, and when the input is as
unconstrained by reality as Body Inversion and Secret Transport to Inverted
Earth scenarios, small wonder that confusing philosophical exotica and
esoteric discussions soon arose .
Happily, the microstimulation results surveyed in section 5 of this
chapter give us real (Earthly) scientific analogs to the Inverted Earth fanta sy.
Consider the key feature s of an Inverted Earth scenario, in which the
Earthling individual has been transported with the lenses and body pigment
changes:
Same (internal) brain state (because of the inverted lenses);

Same perceptual judgment ("Still blue");
Different external stimulus (blue sky versus yellow sky).
The issue between the phenomenal internalist and externalist is whether the
contents of the qualitative states-the qualia-count as the same or different
across these features . But these are exactly the f eatures of the normal
presentation without microstimulation versus the microstimulation plus opposite stimulu s in the studies surveyed above. Consider the Newsome lab's
motion direction study (although any of the others would illustrate this point
equally well). The "Earth" analogy is the case where the monkey is presented
with a stimulus with a percent correlation of dots moving in a particular
direction, say 90 left, and no microstimulation. The "translocated Inverted
Earth " analogy is the case where the monkey is presented with the opposite
stimulus, e.g., the same percent correlation of dots moving 90 right, plus
microstimulation to a left-preferring cluster of MT motion selective neurons.
The microstimulation is analogous to the inverting lenses . Acro ss these two
experimental cases we have, analogous to the above :
the "same" brain states , one induced by the external leftward motion
stimulu s, the other induced by micro stimulation;
210
the same judgment about motion direction, since the monkey in both
. cases will saccade consistently to the leftward or "Pref" target light ;
different external environmental stimuli, namely, percent correlation
motion to the left in the former case and to the right in the latter.
We thus don't need the fantastic fictional example. Microstimulation studies

being pursued in real laboratories right here on Earth capture exactly the
features that Inverted Earth with Secret Transplanetary Relocation were
designed to illustrate. We can thus ask the phenomenal internalist and externalist to reflect on real science. Behold, a methodological gain in the
philosophy of consciousness!
Block ([ 1990] 1998) , incidentally, anticipated the possibility of an
empirical case analogous in the important ways to Inverted Earth, although he
has nowhere suggested that current microstimulation research provides this.
When considering an objection to his Inverted Earth criticism of intentionalist/functionalist accounts of qualia, namely that "the boundary between the
inside and the outside should be moved inward, and inputs should be thought
of in terms of the color of the light hitting the retina, not in terms of colored
objects," Block replies: "I mentioned an alternative to the inverting lenses,
one that seems to me more physically plausible than the lenses , namely, a
neural inverter behind the retina.... I don't know any reason why it shouldn't
be in principle possible for a miniaturized silicon chip to register these
impulses and substitute transformed impulses for them" ([1990] 1998, 688).
Substitute "tungsten electrode" for Block's fantasized silicon chip, implant it
well back "behind the retina" (into the appropriate cortical region for the
stimulus at issue) and you have the actual scientific method of current
microstimulation studies.
However, more than just a gain in philosophical method seems to be
on offer here . The neuroscientists doing the microstimulation studies surveyed
above commonly suggest that the monkeys' subjective consciou s experiences
were similar across the microstimulation and no-rnicrostimulation cases,
despite the different external stimuli. These scientists don't use philosophical
jargon like "qualia," "phenomenal content," and "individuation," but we can
easily translate the suggestive comments surveyed in the previous section into
these terms. According to these neuroscientists, their results show that what
matters for qualitative sensory experience is what goes on in the brain, not
what goes on in the external environment. Normally, external stimuli correlate
systematically with internally generated qualia. But as the microstimulation
studies consistently suggest, the two can be dissociated. When they are ,
monkeys consistently judge stimulus qualities based on the preferred stimuli
of the neurons being microstimulated, not the actual features of the external
stimuli. The natural interpretation is that the internally generated subjective
CONSCIOUSNESS
211
qualia across microstimulation and no-microstimulation cases share the same

features. Any generalized skepticism about "actual phenomenology" being
induced electrically in higher primates is softened by the variety of stimulus
types and modalities for which these microstimulation results have been
gathered and, most importantly, from linguistic reports from one type of
higher primate-humans-undergoing the Penfield procedure. The science
points in favor of internalist intuitions about phenomenal, qualitative content.
I round off this discussion by returning to Lycan's comment about the
"car crash" quality of the Block-Lycan debate. After agreeing with Polger and
Flanagan's assessment (see footnote 19 of this Chapter), Lycan ends his
discussion with the following suggestion:
I would point out, though, that the issue between Block and
me was clear: I maintain that mental states and events (as
such) have no metaphysically interesting properties save their
functional properties and their representational properties.
Block disagrees, insisting that phenomenal states have special
properties of a third sort. The mayhem that followed was a
matter of my trying to understand exactly what sorts of
property those are supposed to be, and (as you might guess)
of the two of us simultaneously competing to distribute the
burden of proof. (2001 , 131, footnote 1)
But in at least one passage Block offers a suggestion about this "third sort" of
property, and the scientific lessons I surveyed in the previous two sections
help fill it out. He writes: "I take the view that thought is functional, but that
the best bet for a theory of qualia lies in neurophysiology" ([ 1990] 1998, 693,
footnote 29) . Block's qualia need not be some metaphysical "third property,"
as Lycan suggests, because even functionalists about intentional states like
Lycan and Block presume (at least token) physiological realizers for all types
of mental states. But Block is suggesting that the physiological properties of
phenomenal states exhaust their qualitative contents. The microstimulation
studies suggest such an account. What is the motion quale of my visual
experience as the lobbed football moves across my visual field? It is the
activity in motion-selective MT neurons, especially those spiking most
frequently to actual motion in my visual field at that speed, direction, and
location. Dissociate the internal and external events-lob the football across
my visual field in a different direction while microstimulating a small cluster
of MT neurons selective for motion stimuli in the first direction-and I'll
have the same motion quale as before . Only now, I'll be a lot less effective as
a receiver on your flag-football team. A similar account can now be offered
for visual depth qualia, visual heading qualia, and somatosensory flutter
212
qualia, with the promise by the neurophysiologists carrying out these studies
of more coming in the future.
And yet we'll still hear the clamor: "But action potential frequency
patterns in cortical neurons don't seem like sensory qualia! How can
neuroscience bridge that gap?" Although the available evidence-i.e., the
philosophical literature on consciousness-suggests that nothing changes
opinions on either side of this clamor, I'll end this chapter with evidence that
prompts me to dismiss this clamor. This evidence reflects an attitude toward
"the hard problem" from the hard-core Society for Neuroscience crowd.
7 THE "HARD PROBLEM" AND THE SOCIETY FOR

NEUROSCIENCE CROWD
There are neuroscientists who think of the brain as "just another
organ" or 'just another piece of biological tissue." However, many pursue the
discipline for reasons that historically have motivated humanists, and are not
afraid to express these motives in print. A nice example is this passage from
the introductory chapter of Gordon Shepherd's influential textbook: "As we
grow older, we experience the full richness of human behavior-the ability to
think and feel, to remember and create-and we wonder, if we have any
wonder at all, how the brain makes this possible" (1994, 3). This is not the
ranting of some left-field crank, but rather from the current editor of the
Journal of Neuroscience . Similar passages can be cited many-fold. Many
bench neuroscientists aren't philosophical philistines.
These passages won't satisfy some philosophers, however, who
remain jealous guardians of the "qualitative" and "subjective" aspects of
Mind. They assume that only they, and perhaps a handful of theoretically
minded psychologists, grapple seriously with "what it is like" to be a
conscious, mindful human being. They assume that these features are beyond
neuroscientists' professional grasp and serious interest. But they are wrong.
Consider the following passage from William Newsome, commenting on
microstimulation studies from his lab (surveyed above in section 5.1 of this
chapter):
I believe the nature of internal experience matters for our
understanding of nervous system function ... Even if I could
explain a monkey 's behavior on our task in its entirety (in
neural terms), 1 would not be satisfied unless I knew whether
microstimulation in MT actually causes the monkey to see
motion. If we close up shop before answering this question
and understanding its implications, we have mined silver and
CONSCIOUSNESS
213
left the gold lying in the tailings. (Gazzaniga 1997,65-66; my

emphases)
Yet Newsome asks for no special discipline or methodology to address this
question. He sees no shortcuts around a broadly empi ricist, reductionist path,
writing: "For the time being ... I suspect we must feel our way towards these
ambitious goals from the bottom up, letting the new light obtained at each
level of inquiry hint at the questions to be asked at the next level" (Gazzaniga
1997,67)?O
Zealous guardians of "the hard problem" in the philosophy of
consciousness should loosen up. Philosophers aren't the only ones respectful
or in pursuit of the full glory of Mind. If we are to trust neuroscientists
pursuing "mainstream" research on cellular and molecular mechanisms, this
problem is not beyond their professional interest or training. And any
philosophical position that insists on a priori grounds that " it must be, based
on the very nature of reductionistic science ," would profit from taking a look
at actual, mainstream, professional-scientific-journal-publication research .
The lesson I've urged throughout the last three chapters is that the
philosophical lessons lie there, even for "the hard problem" of consciousness.
And to my lights , pursuing those lessons there sure beats chiming in with one
more intuition about worn philosophers' fantasies like Twin Earth , inverted
spectra, dancing qualia, and Mary the colorless visual neuroscientist.
NOTES
lOne set of results I will not discu ss is object- speci fic activity in neuron s comp rising regions of
the "ventral" visual stream (see Figures 4.2 and 4.3 below). My ignoring these exper iments and
results is not due to their lack of scientific interest or philo sophical import, but rather becau se
they have already been presented to philosopher s of consciousness in an admirable paper by
Jesse Prin z (2000).
2 For example, Dennett ( 199 1) gives an extend ed argument against theate r metaphors for the
conscious mind .
With one exception : Funahash i et al. (1989) report some differences in PS neuron response
propertie s compared with frontal eye field (FEF) neurons during the ODR task. Since the
output of a preponderant number of FEF neurons is known to code for oculomotor dimensions
of an upcoming saccade, these differences are suggestive against the altern ative explanat ion . I
discussed FEF pre-saccadic activity briefly , with references, Chapter Three, section 3.2 above.
3
4 On the Stroop Task, subjects are presented with a list of color words ('blue,' 'g reen,' ' red,'
etc.) that appear in color , but never the same color as the colo r word's reference . For example,
the word 'red ' appears in blue , 'green' in red, and so on. One task is to name the color in which
the word appe ars as quickly as possible, which requires subje cts to inhibit reading the color
name. The Wisconsin card sorting task presents subjects with four stimulu s cards , each with
214
designs that differ in color, form, and number of elements . There are numerous variations, but
in one the first card might have one pink heart , the second two yellow moons , the third three
orange stars, and the fourth four green clove rs. The rest of the cards in the deck have alternative
combinations of these colors , forms, and numbers. Subjects must sort each remaining card in
the deck by placing it in front of one of the four stimulus cards . The only hint subjects are given
is an indication of whether a given sort is correct or incorrect. Unannounced to the subject,
"correct" and "incorrect" are determined in the following way. Color is the first solution. As
soon as the subject indicates that he or she has figured this out , the solution suddenly changes
(without warning or explicit announcement) to form. To succeed , subjects must inhibit sorting
the cards on the basis of color and switch to form . Once the subject has switched successfully
to form, the solution changes (without warning or explicit announcement) to number of
elements . Later it will become color again, and so on. Both tasks are used extensively in
neuropsychological investigations of frontal lobe damage, as patients with certain types of
frontal damage perform very poorly on them. Incidentally, nonpaticnt controls regularly
express subjective difficulty with these tasks and the need for explicit, conscious concentration
to perform well. For a "textbook" description of these tasks and their neuropsychological use,
see Kolb and Whishaw 1996, chapter 14.
Pyramidal cells are a type of neuron, so named because of the shape of their cell bodies. They
are the primary type of "working memory " neurons in prefrontal cortex .
In keeping with the theme in reductionist neuroscientific research emphasized in Chapter

Two, section 5.2 above, they note that this result is consi stent with some early results using D I
mutant knock-out mice.
For those worried that my talk of the causal effects of explicit conscious attention on single
neuron activity borders on dualism or mystery , be comforted. Toward the end of this section
I'll discuss a neurobiological explanation of these effects that is under active development and
invest igation .
8 This isn't just an arbitrary intuition on Chalmers' part. He has a much-discussed theory of
reductive explanation, most fully developed in the early chapters of his (1996) book .
Thanks to John Symons for separating these two projects and emphasizing the importance of
the first.
10 Case and page numbers cited in the next three paragraphs refer to the case histories in
Penfield and Perot (1963).
11 Thus their responses likewise are fit by Gaussian curves (Figure 4.4A above), where values
on the x-axis represent motion direction parameters.
12 Fans of cheesy 1950s 3-D horror and science fiction movies are familiar with anaglyphs.
Remember the cardboard glasses with red and green plastic lenses? Anaglyphs are stereoscopic
motion or still pictures whose right component, usually red in color , is superimposed on the left
component, usually green, to produce a three-dimensional effect when viewed through
differently-colored optical filters over the two eyes.
Britten and van Wezel report a fair amount of heterogeneity in their microstimulation effects .
In a few cases , microstirnulation of MST clusters biased the monkeys ' choices opposite of the
heading -direction tuning of the neurons in the cluster, i.e., stimulating a cluster tuned to left
heading direction biased the monkey choice toward the right heading target. They offer
explanations for this puzzling effect , one being that the visual display far exceeds the receptive
field of the stimulated neurons. The monkeys probably use cues from the entire stimulus, not
13
CONSCIOUSNESS
215
just from within the receptive fields of the stimulated neurons . Lateral interactions within MST
could then on occasions override the microstimulation effects . They also note that heading
computations might be exceedingly complex, keyed to more dimensions of the visual stimulus
than the ones controlled for in this study. Given the obvious complexity of this type of visual
stimulus and the neuronal computations proeessing it (even within a single region like MST),
that Britten and van Wezel (1998) achieved results as robust as they did is fascinating from our
perspective of phenomenology induced by neuronal microstimulation.
14 Thanks to Huib Looren de long and Maurice Schouten for continually impressing this worry
about explanation concerning the scientific evidence presented in Chapters Two through Four.
15 Thanks to Tom Polger for discussions that helped me clari fy the views and arguments of this
section.
16 Twin-you is your molecule-far-molecule doppelganger-okay, ignore the fact that so much
of you consists of water!
17However, Block adapted Inverted Earth from Harman to criticize "functionalist!

intentionalist" theories of qualitative experience. His targets were early prototypes of representationalist theories prominent these days, and closely allied with phenomenal externalism.
18 The necessity modality in the last clause of this quotation must be viewed in larger context.
Throughout his (1996), Dretske is careful to point out that he is urging the availability, not the
truth, of phenomenal externalism. Lycan (1996) is a bit bolder, albeit less pithy.
Lycan doesn 't end the footnote here, however. Despite the admitted car crash quality of the
Bleck-Lycan exchange, he still thinks that he can put his finger on the issue that really
separates the two views of qualia. I'll say more about his additional remarks at the end of this
section .
19
Notice that Newsome's reductionism doesn't collapse levels , at least not methodologically,
as many anti-rcductionist philosophers fear.
20
REFERENCES
Abel , T., P . Ng uyen, M. Barad, T . Deuel, E.R. Kandel , and R. Bourtchouladze ( 1997 ).
"Genetic demonstratio n of a role for PKA in the late phase of LTP and in
hippocampus-based lon g-term me mory ." Cell 88 : 615-626.
Alb er ini, C.M ., M. Ghirardi, R. Metz, and E.R . Kand el (1994). "C/ EBP is an immedi ate early
gene require d for the con solidation of long-term facilitation in Aplysia." Cell 76 :
1099 -11 14.
Alb right, T .D ., R. Desimone , and C. Gros s (1994). "Columnar or ganization of dire ctionally
selective cell s in visual area MT of mac aques." Journal of Neurophysiology 51 : 163 1.
Aranci o, 0 , M . Kiebl er, C. Lee, V. Lev -Ram, R. Tsi en , E.R. Kandel , and R. Hawk ins (1996).
"N itric ox ide act s dir ectly in the pre synaptic neuron to produce long-term
potentiati on in cultured hippocampal neuron s." Cell 87 : 1025-1035.
Arancio, 0., I. Antonova, S. Gambaryan , S. Lohmann, 1. Wood, D. Lawrence, and R. Haw kin s
(2000). "Presynaptic rol e of cGMP-dep end ent protein kin ase during lon g-lasting
pot entiation." Journal of Neuroscience 21 : 143-149.
Arthurs, O. and S. Bon iface (2002). "How well do we und erstand the neur al origins of the
fMRI-BOLD signal ?" Trends in Neurosciences 25 : 27- 31.
Baars, B. (1997). In the Theater of Consciousness. Oxfo rd : Oxford University Pres s.
Baddeley, A. ( 1986). Working Memory. Oxford: Clarendon Press.
Baddeley,
A. (1994) . "Working
memory."
In M. Ga zzaniga
(cd .), The
Cognitive
Neurosciences. Cambridge, MA : MIT Press, 755-764 .

Bah r, B., U. Staubli , P. Xiao , D. C hun, Z . Ji., E. Esteb an , and G. Lynch (199 7). "Arg-Gly- AspSer-Selective Adhesion and the stabili za tion of lon g-term potentiati on :
Pharmacolog ical studies and the characteri zat ion o f a candidate ma trix recepto r."
Journal of Neuroscience 17 : 1320-1329.
Bailey, C ., D. Bartsch , and E.R. Kand el (1996). "Toward a mol ecular definition o f lon g-term
potentiation." Proceedings of the National Academy of Sciences 93 : 13445-13452.
Bailey , C.H . and M . Chen. (1989). 'T ime course o f structural changes at identifi ed sensory
neuron synapses during lon g-term sensi tization in Apl ysia." Journal of Neuroscience
9 : 1774-1 780 .
Baker , L.R. ( 1993) . "Metaphysics and mental cau sation ." In J. \-leil and A. Mele (cds .), op. cit.
Balz er , W. and C .U. Moulines (1996). Structuralist Theory of Science. Walter de Gru yter ,
Berlin .
Balze r, W., C.U . Moulines, and J.D. Sneed (1987). An Architectonic f or Science. Reidel ,
Do rdrecht.
Bartsch , D., A. Casadio, K.A. Karl, P. Serodio, and E.R. Kandel (1998). "CREB I encodes a
nuclear activator, a repr essor, and a cytopl asmic modul ator that form a regulatory
unit cri tical for long-term fac ilita tion ." Cell 95 : 211-223 .
Bartsch, D., M. Ghirardi , A. Casadio, M. Giustetto, K.A. Karl , \-l. Zhu, and E.R. Kandel (2000) .
" Enhancement of memory-rel ated lo ng-t erm facili tation by ApAF, a novel
218

transcription factor that acts downstre am from both CREB I and CREB2." Cell 103 :
595-608 .
Bartsch, D., M. Ghirardi, P. Skehel , K. Karl, S. Herder, M. Chen, C. Bailey, and E.R. Kandel
( 1995). " Aplysia CR EB2 repre sses long-term facilitation : Relief of repre ssion
converts transient facilitation into long-te rm functional and structural chan ge." Cell
83: 979-992 .
Bechtel, W. and J. Munda le (1999) . "Multiple realizability revis ited : Linking cognitive and
neural states ." Philo soph y of Science 66: 175-207.
Bechtel , W. and R. Richardson (1993). Disco vering Compl exity: Decomp osition and Localization as Strate gies in Scientific Research . Princeton : Princeton University Press, 1993.
Beckermann, A. (200 1). "Phy sicalism and new wave reductionism." Grazer Phi losophische
Stud ien 61, 257-261.
Bergold . P., 1. Sweatt, I. Winicov, K. Weiss, E.R. Kandel, and J.M. Schwartz (1990). "Protein
synthesis during acq uisition of long-term facilitation is needed for the persistent loss
of regulatory subunits of the Ap lysia cAMP -dependent protein kinase." Proceedings
of the National Acad emy ofScien ces USA 87 : 3788-379 1.
Bernabeu , R., L. Bevilaqya, P. Ardenghi , E. Bromberg, P. Schmitz , M. Bianchin, I. Izquierdo ,
and J. Medina, 1. (1997) . "Involvement of hippocampal cAMP/cAMP-dependent
protein kinase signaling pathways in a late memory consolidation phase of aversive ly
motivated learning in rats." Proceeding s of the National Academy of Sciences 94 :
704 1-7046.
Bernstein, M., S. Stiehl , and J. Bickle (2000) . 'The effect of motivation on the stream of consciousness: Generalizing from a neurocomputational model of cingulo-frontal circuits
controlling saccadic eye movemen ts." In R. Ellis and N. Newton (eds .), The Caldron
of Consciousness: Motivation, Affect, and Self-Organi zation . Amsterdam: John
Benjamins ,133-161.
Bickle, 1. (1989) . Towards a Scientific Reformulation of the Mind-Body Problem . Doctoral
dissertat ion, University of California, Irvine.
Bickle, J. (I 992a) . " Multiple realizability and psychophysical reduction. " Beha vior and
Philosophy 20: 47-58 .
Bickle , J. (l992b). "Revisionary physicalism ." Biology and Philosophy 7: 411 -430.
Bickle, J. (1993) . "Connectionism, elirninativism, and the semantic view of theorie s."
Erkenntnis 39 : 359-382 .
Bickle, 1. ( 1995a). "Connectionism, reduction, and multiple realizability ." Behavior and
Philosophy 23 : 29-39 .
Bickle, J. (I 995b ). "Psychoneural reduction of the genuinely cognitive: Some acco mplished
facts." Philosophical Psychology 8: 265-285 .
Bickle, J. (1998) . Psychon eural Reduction: The New Wave. Cambridge, MA: MIT Press.
Bickle, J. (1999) . " Multiple realizability." In E. Zalta (cd.) Stanford Encyclopedia of
Philosophy (http://plato .stanfor d.edu/entries/multipl e-realizabi lity/)
Bickle, J. (2000). "Concepts of intertheoretic reduction in current philosophy of mind. " In A
Field Guid e to Philosophy of Mind (Societa Italiana Filosofia Analitica). http :
//www .uniroma3.itlkantlfi eld/
Bickle, 1. (2001). "New wave metascience: Replies to Beckerman, Malon ey, and Stephen."
Grazer Philosophische Studien 61: 285-29 3.
REFERENCES
219
Bickle, J. (2002). "Concepts structured through reduction : A structuralist resource illuminates

the consolidation-long-term potentiation (LTP) link ." Synthese 130: 123-133.
Bickle, J., C. Worley, and M . Bernstein (2000). "Vecto r subtraction implemented neurally: A
neurocomputational model of some sequential cognitive and conscious processes."
Consciousness and Cognition 9 : 117-144.
Bickle, J., S. Holland, V . Schmithorst, and M. Avison (2001). "Cellular mechanisms of sequential processing in frontal regions revealed using a combined computational-fMRI
methodology. " Proceedings of the First World Congress in Neuroinformatics . Technical University Vienna: 581-597.
Bliss, T. and A. Gardner-Medwin (1973). " Lo ng-lasting potentiation of synaptic transmission
in the dentate area of the unanaesthetized rabbit following stimulation of the
perforant path ." Journal of Physiology (London) 232 : 357-374.
Bliss , T. and T . Lerno (1973) . " Lo ng-lasting potentiation of synaptic transmission in the
dentate area of the anaesthetized rabbit following stimulation of the perforant path. "
Journal ofPhysiology (London) 232 : 331-356.
Block, N. ([1990] 1997). "Inverted earth. " Reprinted in N. Block, O. Flanagan, and G.
Guzeldere (eds .), The Nature of Concs ciousness. Cambridge, MA : MIT Press, 677693.
Block, N . (1978). "Troubles with functionalism." In C. Savage (ed .), Perception and
Cognition : Issues in the Foundations of Psychology. Minnesota Studies in the
Philosophy of Science, volume 9. Minneapolis: University of Minnesota Press, 261325.
Bentley, T. (2000) . "Review of Bickle, Psychoneural Reduction: The New Wave." British
Journal for the Philosophy of Science 51: 901-905.
Bourtchouladze, R., B. Frenguelli, J. Blendy, D. Cioffi, G. Schutz, and Silva, A. (1994) .
"Deficient long-term memory in mice with a targeted mutation of the cAMPresponsive cl ement binding protein.' Cell 79 : 59-68.
Britten, K. and R. van We zel (1998). "Electrical microstimulation of cortical area MST biases
heading perception in monkeys." Nature Neuroscience 1: 59-63.
Bruce, C. and M . Goldberg (1985) . " Prim ate frontal eye fields . I. S ingle neurons discharging
before saccades. " Journal of Neurophysiology 53 : 603-635.
Bunsey, M. and H. Eichenbaum (1996). "Co nservation of hippocampal memory function in
rats and humans. Nature 379: 255-257 .
Burge, T. (1979). "Individualism and the mental ." In P. French, T. Euhling, and H. Wettstein
(eds .), Midwest Studies in Philosophy, Volume 4: Stud ies in Epistemology.
Minneapolis: University of Minnesota Press.
Burge, T. (1993). "Mind-body causation and explanatory practice." In J. Heil and A. Mele
(eds .), op. cit.
Burman, D. and C. Bruce (1997). "Suppressio n of task -related saccades by electrical stimulation in the primate 's frontal eye fields ." Journal of Neurophysiology 77 : 2252-
Cammarota:~. ,
L. Bevilaqua, P. Ardenghi, G. Paratcha, M. Levi de Stein, I. Izquierdo, and J.

Medina (2000) . "Learning-associated activation of nuclear MAPK, CREB and Elk- I,
along with fos production, in the rat hippocampus after one-trial avoidance learn ing :
Abolition by NMDA receptor blockade." Molecular Brain Research 76 : 36-46.
Campbell, N. and J. Reece (2001). Biology, 6 th Ed. New York : Benjamin/Cummings.
220
Carlson, N.R. (1994) . Physiology of Behavior, 5 th Ed . Boston : Allyn and Bacon.

Carnap, R. ([1932] 1958) . "The elimination of metaphysics through logical analysis of
language" (trans. Arthur Pap) . Reprinted in A.J . Ayer (cd .), Logical Positivism .
Glencoe, IL: Free Press, 1959, 60-81 .
Carnap, R. ([1950], 1956) . "Empiricism, semantics, ontology." Reprinted in R. Carnap,
Meaning and Necessity, 2 11d Ed. Chicago: University of Chicago Press, 205-221.
Camap, R. (1934). The Unity of Science (trans. M. Black) . London: Kegan, Paul, Trench,
Trubner and Company.
Chafee, M. and P.S . Goldman-Rakic (1998). " Matching patterns of activity in primate prefrontal area 8a and and parietal area 7ip neurons during a spatial working memory
task ." Journal of Neurophysiology 79 : 2919-2940.
Chain, D., A. Casadio, S . Schacher, A. Hegde, M. Valbrun, N. Yamamoto, A. Goldberg, D.
Bartsch, E.R . Kandel, and J. Schwartz (1999). " Mechanisms for generating the
autonomous cAMP-dependent protein kinase required for long-term facilitation in
Aplysia.' Neuron 22 : 147-156.
Chalmers, D. (1995). "Facing up to the problem of consciousness." Journal of Consciousness
Studies 2: 200-219.
Chalmers. D. (1996). The Conscious Mind . Oxford: Oxford University Press.
Churchland, P.M . (1982). "Is thinker a natural kind?" Dialogue 21: 223-238 .
Churchland, P.M . (1985). " Reduction, qualia, and the direct introspection of brain states."
Journal of Philosophy 82: 1-22 .
Churchland , P .M . (1987). Matter and Consciousness, Revised Ed . Cambridge, MA : MIT Press.
Churchland, P.M . (1995). The Engine of Reason , The Seat of the Soul. Cambridge, MA : MIT
Press.
Churchland, P.S . (1986). Neurophilosophy: Towards a Unified Science of the Mind/Brain.
Cambridge, MA : MIT Press.
Churchland, P.S . and T.J . Sejnowski (1992). The Computational Brain . Cambridge, MA : MIT
Press.
Clayton, D.F . (2000). "The genomic action potential. " Neurobiology of Learning and Memory
74 : 185-216.
Cohen, N . and H. Eichenbaum (1993). Memory , Amnesia, and the Hippocampal System . Cambridge, MA: MIT Press.
Compte, A., N . Brunei, P.S . Goldman-Rakic, and X. Wang (2000). "Synaptic mechanisms and
network dynamics underlying spatial working memory in a cortical network model. "
Cerebral Cortex 10: 910-923 .
Connolly, J.B . and T. Tully (1997). "Behaviour, learning and memory." In D. Roberts (cd .),
Drosophila: A Practical Approach . Oxford: Oxford University Press .
Courtney, S., Petit, L., Ma isong, J., Ungerlcider, L., and Haxby, J. (1998). "An area specialized
for spatial working memory in the human frontal cortex." Science 279: 1347-1351.
Craver, C. (2003). "The making of a memory mechanism." Forthcoming in Journal of the
History of Biology .
REFERENCES
221
Craver, C. and L. Darden (200 I) . "Discovering mechanisms in neurobiology : The case of

spatial memory ." In P. K. Machamer, R. Grush, and P. McLaughlin (cds.), Theory
and Method in Neuroscience . Pittsburgh , PA: University of Pittsburgh Press.
Davis, H. and L. Squire (1984) . "Protein synthesis and memory : A review ." Psychological
Bulletin 96: 518-559.
DeAngelis, G., B. Cumming, and W. Newsome (1998). "Cortical area MT and the perception
of stereoscopic depth ." Nature 394: 677-680.
Dennett , D.C. (1991) . Consciousness Explained. Boston : Little Brown.
Downing, C. (1988) . " Expectancy and visual-spatial attention: Effects on perceptual quality. "
Journal of Experimental Psychology: Human Perception and Performance 14: 188202 .
Dretske , F. (1996) . "Phenomenal externalism: If meanings ain't in the head, where are qualia?"
In E. Villaneuva (ed.), Philosophical Issues 7. Atascadero, CA: Ridgeview, 143- 158.
Dubnau, J. and T. Tully (1998). "Gene discovery in Drosophila : New insights for learning and
memory ." Annual Review of Neuroscience 21: 407-444.
Dudai, Y. and W.G. Quinn (1976). "Memory phases in Drosophila." Nature 262 : 576-577 .
Duncan, C. (1949) . "The retroactive effect of electroshock on learning ." Journal of
Comparative and Physiological Psychology 42 : 32-44 .
Eichenbaum, H. (1997). "Declarative memory: Insights from cognitive neurobiology." Annual
Review of Psychology 48 : 547-572.
Enc, B. (1983). "In defense of the identity theory." Journal of Philosophy 80: 279-298.
Endicott, R (2001) . "Post-structuralist angst: Critical notice of Bickle's Psychoneural
Reduction : The New Wave ." Philosophy ofScience 68: 377-393 .
Feigl, H. (1967) . The 'Mental' and the 'Physical': The Essay and a Postscript. Minneapolis :
University of Minnesota Press.
Feyerabend, P.K. (1962) . "Explanation, reduction , and empiricism. " Minnesota Studies in the
Philosophy of Science, vol. 3. Reprinted in P.K . Feyerabend, Realism, Ratonalism
and Scientific Method: Philosophical Papers, vol. I. Cambridge: Cambridge
University Press , 44-96.
Fodor, J. (1975). The Language of Thought. New York: Thomas Crowell.
Fodor, lA. ([ 1974] 1981). "Special sciences ." Reprinted in J.Fodor, Rel'resentations.
Cambridge, MA: MIT Press, 127-145..
Freeman , W. (2000) . Neurodynamics: An Exploration in Mesoscopic Brain Dynamics. New
York : Springer.
Frost, W.N., V.F. Castellucci , R.D. Hawkins , and E.R. Kandel (1985) . "Monosynaptic
connections from the sensory neurons of the gill- and siphon -withdrawal reflex in
Aplysia participate in the storage of long-term memory for sensitization ."
Proceedings of the National Academy of Sciences USA 82 : 8266-8269 .
Funahashi, S., C. Bruce, and P.S. Goldman-Rakic (1989). " Mnemonic coding of visual space in
the monkey's dorsolateral prefrontal cortex ." Journal of Neurophysiology 61: 1-19.
Funahashi, S., C. Bruce, and P.S. Goldman-Rakic (1991). "Neuronal activity related to
saccadic eye movem ents in the monkey 's dorsolateral prefrontal cortex ." Journal of
Neurophysiology 65 :1464-1483 .
222
Funahashi, S., M. Chafee, and P.S. Goldman-Rakic (1993 ). "Prefrontal neuronal activity in
rhesu s monk eys performing an anti -saccade task ." Natur e 365 : 753-756.
Garcia, 1. and R. Koelling (1966). "Relation of cue to consequence in avoidance learning."
Psychonomic Science 4: 123-124.
Gaymard, B., S. Rivaud , 1. Cassarini, T. Dubard, G. Rancure l, Y. Agid , and C. PierrotDeseilligny, C. (1998). "EFfects of anterior cingulate lesions on ocular saccades in
humans." Experimental Brain Research 120: 173-183 .
Gilbert, C ., M . Ito, M. Kapadia, and G. Westheimer (2000). "In teractions between attention,
context and learning in primary visual cortex." Vision Research 40 : 1217 -1226.
Goldberg, M . and Bruce, C. (1990). "Primate frontal eye Fields. III. Maintenance of spatially
accurate saccade signal." Journal of Physiology 64 : 489-508.
Goldberg, M ., H. Eggers, and P. Gouras (199 1). "The oculomotor system." In E.R . Kandel, J.
Schwartz, and T. Jessell (eds .), op. cit.
Goldrnan-Rakic, P.S. (1995). "Cellular basis of working memory." Neuron 14 : 477-485 .
Goldman-Rakic, P.S. (1996) . "Regional and cellular fractionation of working memory."
Proceedings of the Nationa l Academy of Sciences (USA) 93 : 13473- 13480 .
Goldman-Rakic, P.S., S. O'Scalaidhe, and M . Chafee (2000). " Domain specificity in cognitive
systems." In M. Ga zzaniga (ed .), The New Cognitiv e Neurosciences. Cambridge,
MA : MIT Press , 733 -742 .
Goodale, M . and A. Milner (1992). "Separate visual pathways for perception and action."
Trends in Neuroscien ces 15. 20 -25.
Hannan, B. (2000). "Review of Bickle, Psychoneural Reduction : 771e New Wave." Philosophical Books 41: 53-54.
Hardcastle, V.G. (1995 ). Locatin g Consciousness . Ams terdam: John Benjamins .
Harman, G. (1982 ). "Conceptual role semantics." Notre Dame Journal of Formal Logic 23.
Haus-Seuffert, P. and M. Meisterernst (2000). "Mechanisms of transcriptional activation of
cAMP-responsive clement-binding protein CREB ." Molecular and Cellular
Biochemistry 212 : 5-9.
Hawkins , R.D. and E.R. Kandel ( 1984a) . "Is there a cell -bio logical alphabet For simple Forms
of learning?" Psychological Review 91 : 375-39 1.
Hawkins, R.D. and E.R. Kandel (1984b). "Steps toward a cell-biological alph abet For
elementary forms of learning." In G. Lynch, J.L. McGaugh, and N.M . Weinberger
(eds .), Neurobiology of Learning and Memory. New York : Gu ilford Press , 385 -404.
Hebb, D. 1949: The Organization of Behavior. John Wiley : New York .
Hegde, A.N., A.L. Goldberg, and J.H. Schwartz ( 1993) . " Regulatory subunits of cAMPdependent protein kinases are degraded after conjugation to ubiqui tin : A molecular
mechanism underlying long- term synaptic plasticity." Proceedings of the National
Academy of Sciences USA 90 : 7436-7440.
Hegde, A.N., K Inokuchi, W . Pei, A. Casadio, M . Ghirardi , D.G. Chain, KC. Martin, E.R.
Kandel , and J.H. Schwartz ( 1997). "Ubiquitin C-Terminal Hydro lase is an immediate
early gene essential For long-term facilitation in Aplysia ." Cell 89 : 115- 126.
Heil, J. and A.Mele (cds .) (1993). Mental Causation. Oxford: Clarendon Press .
REFERENCES
223
Holland, S., E. Plante, A. Weber-Byers, R. Strawsburg, V. Schmithorst, and W. Ball (2001).

" Normal fMRI bra in activation patterns in children performing a verb generation
task ." Neuroimage 14: 837-843 .
Hooker, C.A . (1981). "Towards a general theory of reduction . Part I: Historical and scientific
setting. Part 11: Identity in reduction. Part lll : Cross-categorial reduction." Dialogue
20 : 38-59, 201-236, 496-529.
Horgan , T. (1993). "Nonreductive materialism and the explanatory autonomy of psychology,"
In S. Wagner and R. Warner (eds.), Naturalism: A Crit ical Appraisal. Notre Dame,
IN: University of Notre Dame Press , 295-320.
Horgan, T. (200 I) . "Causal compatibilism and the exclusion problem." Theoria 16: 95-116.
Hurnmler, E., T. Cole, J. Blendy, R. Ganss , A. Aguzzi, W. Schmid, F. Bccrrnann, and G.
Schutz (1994) . "Targeted mutation of the CREB gene : Compensation within the
CREBIATF family of transcription factors ." Proceedings of the National Academy of
Sciences (USA) 91: 5647-5651.
Izquierdo, 1. and J. Medina (1997). "Me mory format ion: The sequence of biochemical events
in the hippocampus and its connection to activity in other brain structures.'
Neurobiology of Learning and Memory 68 : 285-316.
Jackson, F. (1983). "Epiphenomenal qualia." Philosophical Quarterly 32 : 127-136.
James, W. (1890) . The Principles of Psychology. New York : Henry Holt.
Kandel , E.R. (1979). "Cellular insights into behavior and learning." Harvey Lectures 73 : 19Kandel ,
E~~:, J.R. Schwartz, and T. Jcssell
(cds .) (1991). Principles of Neural Science, 3rd Ed.
New York : McGraw-HilI.
Kandel , E.R., 1.R. Schwartz, and T. Jessell (cds .) (2000). Principles of Neural Science, 4 th Ed.
New York : McGraw-HilI.
Kim , 1. (1990) : 'Explanatory exclusion and the problem of mental causation .' In E. Villanueva
(ed .), Information , Semantics, and Epistemology. Basil Blackwell : Oxford, 36-56.
Kim , 1. (1993). Supervenience and Mind. Cambridge: Cambridge University Press .
Kim, 1. (1996). Philosophy of Mind. Boulder, CO : Westview.
Kim, 1. (1998). Mind in a Physical World . Cambridge, MA : MIT Press .
Kimura, M. (1983). The Neutral Theory of Molecular Evolution . Cambridge: Cambridge University Press .
Kolb , B. and I. Whishaw (1996) . Fundamentals of Human Neuropsychology, 4th Ed. W.H.
Freeman: New York .
Kosslyn, S. (1997). "Mental imagery." In M. Gazzaniga (ed .), Conversations in the Cognitive
Neurosciences. Cambridge, MA : MIT Press, 155-174 .
Kritzer, M. and P.S. Goldman-Rakic (1995). " Intrinsic circuit organi zation of the major layers
and sublayers of the dorsolateral prefrontal cortex of the rhesus monkey." Journal of
Comparative Neurology 359 : 131-143 .
Kuhn, T. (1962). The Structure of Scientific Revolutions. Chicago: University of Chicago Press .
LeDoux, 1.E. (2000). "Emotion circuits in the brain ." Annual Re view of Neuroscien ce 23 : 155184.
LePore , E. and B. Loewer (1989). "More on making mind matter. " Philosophical Topics 17:
175-191.
224
Levin , L., P. Han, P. Hwang, P. Feinstein , R. Davis , and R. Reed (1992). "The Drosophila
learning and memory gene rutabaga encodes a Ci+/calmodulin-responsive adenylyl
cyclase." Cell 68 : 479-489 .
Levine, J. (1983). Materialism and qua lia: The explanatory gap." Pacific Philosophical
Quarterly 64 : 354 -361.
Levitan, I. And L. Kaczmarek (2001). The Neuron, 3 ' d Ed. Oxford: Oxford University Press .
Lewin , B. 1999: Genes VII. Oxford University Press : Oxford.
Lewis, D. (1969). "Review of Art, Mind, and Religion." Journal of Philosophy 66 : 23-35 .
Lewis, D. (1973). "Scorekeeping in a Language Game." Journal of Philosophical Logic 8:
339-59.
Liu, J. and W. Newsome (2000). "Somatosensation: Touching the mind's fingers ." Current
Biology 10: R598-R600.
Llinas , R. and P.S. Churchland (1996). "Introduction." In R. Llinas and P.S. Churchland (eds.),
The Mind -Brain Continuum . Cambridge, MA : MIT Press , ix-xi .
Lodish, H., A. Berk, L. Zipursky, P. Matsudaira, D. Baltimore, and J. Darnell (2000).
Molecular Cell Biology, 4 th Ed . New York : W.H. Freeman.
Loughland, C. L. Williams, and E. Gordon (2002). "Visual scanpaths to positive and negative
facial emotions in an outpatient schizophrenia sample." Sch izophrenia Research 55 :
159-170.
Lycan , W. (1996). Consciousness and Experience. Cambridge, MA : MIT Press.
Lycan, W. (2001). "Response to Polger and Flanagan ." Minds and Machines 11: 127-132 .
Lynch, G. (2000) . "Memory consolidation and long-term potentiation." In M. Gaz zaniga (ed.) ,
771e New Cogn itive Neurosciences. Cambridge, MA : MIT Press , 139-157 .
Lynch , G. and R. Granger (1989). "Simulation and analysis of a simple cortical network."
Psychology of Learning and Motivation 23: 205-241 .
Martin, K., C. Casadio, Y. Zhu , J. Rose, M. Chen , C. Bailey , E.R. Kandel (1997). "Synapsespecific long-term facilitation of Aplysia sensory to motor synapses : A function for
local protein synthesis in memory storage." Cell 91 : 927-938 .
Maunsell, J. (1995). "The brain's visual world : Representation of visual targets in cerebral
cortex." Science 270: 764 -769.
Mayford, M . and E.R. Kandel (1999). "Genetic approaches to memory storage." Trends in
Genetics 15: 463-470.
Mayr , D. (1976). "Investigations of the concept ofreduction, I. " Erkenntnis 10: 275-294.
McAdams, C. and J. Maunsell (1999). "Effects of attention on orientation-tuning functions of
single neurons in macaque cortical area V4 ." Journal of Neuroscience 19 : 431-441.
McAdams, C. and J. Maunsell (2000). "Attention to both space and feature modulates neuronal
responses in macaque area V4 ." Journal of Neuroph ysiology 83 : 1751-1755.
McEachern , J. and C. Shaw (1996). "An alternative to the LTP orthodoxy: A plasticity-pathology continuum model. " Brain Research Reviews 22 : 51-92.
McGaugh, J. (2000). "Memory: A century of consolidation .' Science 287: 248-251.
McGinn, C. (1989). "Can we solve the mind-body problem?" Mind 98 : 349-366.
REFERENCES
225
Mont arolo, P., P. Goe let, V. Castellucci, J. Morgan , E.R. Kandel, and S. Schacher (1986) . "A
critical period for macromo lecula r synthe sis in long-term heterosynaptic Facilitation
in Aplysia ." Science 234 : 1249-1254.
Mouline s, C.U . (1984) . "Ontological reduction in the natural sciences." In W. Balzer, D.
Pearce , and H.J. Schmidt (cds.), Reduct ion in Science . Dordrecht: Reidel , 51-70 .
Mountcastle, V., M. Steinmetz, and R. Romo (1990) . " Frequency discrimi nation in the sense of
flutter : Psychophysical measurements correlated with postcentral events in behaving
monkeys ." Journal of Neuroscience 10: 3032-3044.
Mura sugi, c., C.D. Salzman, and W. Newsome (1993) . "M icrost irnulation in visual area MT:
EFFects of varying pulse amplitude and Frequency." Journal of Neuroscience 13:
17 19-1729.
Nagel, E. (1961) . The Structure of Science. New York: Harcourt, Brace, and World .
Nagel, T. (1989) . The View from Nowhere. Oxfo rd: Oxford Univers ity Press,
Newsome , W. (1997) . " Perceptual processes ." In M. Gazzan iga (ed.), Conversations in the
Cognitiv e Neurosciences. Cambridge, MA: MIT Press, 53-68 .
O'Scalaidhe, S., F. Wilson, and P.S. Goldrnan-Rak ic (1997). "Area segreg ation of Faceprocessing neurons in prefrontal cortex ." Science 278 : 1135-1138.
Pardo, 1., P. Pardo , and M.E. Raichle ( 1993). "Neural correlates of self-induced dysphoria."
American Journal of Psychiat ry 150:713-719.
Parent, A. (1996) . Carpenter' s Human Neuroanatomy, 9 th Ed. Baltimore : Williams and
Wilkins .
Penfield, W. and P. Perot ( 1963). "The brain 's record of auditory and visual experience: A final
summary and discussion ." Brain 86 : 595-696.
Penrose, R. (1994) . Shadows ofthe Mind . Oxford : Oxford University Press.
Petri, H. and M. Mishkin (1994). "Behaviorism, cognitivism, and the neuropsychology of
memory." Am erican Scientist 82: 30-37 .
Polger, T. and O. Flanagan (200 1). "A decad e of teleo functionalism: Lycan 's Consciousness
and Consciousness and Experience." Minds and Machines 11: 113-126.
Popper , K. (1962). "Truth, ration ality, and the growth of scientific knowledge ." In K. Popper,
Conj ectures and Refutati ons. New York: Basic Books , 215-250.
Posner, M. (1980 ). "Orientation of attention ." Quarterly Journal of Experimental Psycholo gy
32 : 3-25 .
Posner, M. and M. Raichle (1994) . Images of Mind. New York : Scientific American Library.
Prinz, J. (2000) . "A neurofunctional theory of visual consciousness." Consciousness and
Cognition 9: 243-259 .
Purves , W., G. Orian s, H. Heller, and D. Sadava (1998) . Life: The Science of Biology , 5 th Ed.
Sunderland, MA: Sinauer Associate s.
Putnam , H. (1967) . "Psychological predic ates." In W. Capitan and D. Merrill (cds.), Art, Mind,
and Relig ion. Pittsburgh : University of Pittsburgh Pre ss. Reprinted in Putnam (1975 ).
Putnam , H. (I 975a) . Mind, Language, and Reality: Philosophical Papers, Vol. 2. Cambridge :
Cambridge University Press.
226
Putnam , H. (1975b) . "The meaning of ' meaning' ." In K. Gunderson (ed.), Language, Mind,
and Knowledge : Minnesota Studies in the Philosophy of Science, volum e 7.
Minneapolis: University of Minnesota Press.
Quine, W.V.O. ([1949] 1953). "Two dogmas of empiricism." Reprinted in W.v.O. Quine,
From a Logical Point of View. New York: Harper and Row, 20-46 .
Quine, W.V.O. ([195 1] 1966). "On Carnap's views on ontology." Reprinted in W.V.O. Quine ,
The Ways of Paradox and Other Essays . New York: Random House, 126-134.
Quinn, W., W. Harris, and S. Benzer (1974) . "Conditioned behavior in Drosophila melanogaster ," Proceedings of the Nationa l Academy ofSciences USA 71: 707-712.
Reichenbach, H. (1957) . The Rise of Scientific Philosophy. Berkeley, CA: University of
California Press.
Reichling, D. and 1. Levine (1999) . "The primary afferent nociceptor as pattern generator." Pain 6 (Supplement) : S 103-S 109.
Richardson, R.C. (1979) . "Functionalism and reduction ." Philosophy of Science 46: 533-558.
Richardson, R.C. (1999). "Cognitive science and neurosc ience : New wave reductionism."
Philosophical Psychology 12: 297-307 .
Ridley, M. (1996) . Evolution , 2"d Ed. Cambridge , MA: Blackwell Science.
Rorno, R., A. Hernande z, A. Zainos, and E. Salinas (1998) . "Somatosensory discrimination
based on microstirnulation." Nature 392 : 387-390 .
Romo, R., A. Hernandez, A. Zainos , C. Brody, and L. Lemus (2000). "Sensing without
touchi ng: Psychophysical performance based on cortical microstimulation. " Neuron
26: 273-278.
Rossi, A. and M. Paradiso (1995). "Feature-specific effects of selective visual attention."
Vision Research 35: 621-634 .
Salzman, C.D., C. Murasugi , K. Britten, and W. Newsome (1992) . "Microstimulation in visual
area MT: Effects on direction discrimination performance." Journal of Neuroscience
12: 2331-2355 .
Saucier, D. and D. Cain (1995). "Spatial learning without NMDA receptor -depe ndent longterm potentiation." Nature 378: 186-189.
Sawaguchi, T. and P.S. Goldman-Rakic (1994) . "The role of D I-dopamine receptor in working
memory: Local injections of dopamine antagonis ts into the prefrontal cortex of
rhesus monkeys performing an oculomotor delayed-response task." Journal of
Neurophysiology 71 : 515-528.
Schacter, D.L., C. Chu, and K. Ochsner (1993). " Implicit memory: A selective review. " Annual
Review of Neuros cience 16: 159-183.
Schaffner, K. (1967) . "Approaches to reductio n." Philosophy ofScience 34: 137- 147.
Schaffner, K. (1992) . "Philosophy of medicine. " In M. Salmon, J. Eatman , C. Glymour, J.
Lennox, P. Machamer , J. McGuire , J. Norton , W. Salmon, and K. Schaffner,
Introduction to the Philosophy of Science . Englewood Cliffs, NJ: Prentice-Hall, 310344.
Schmit horst, V. and B. Dardzinski (2000). "CCHIPS/IDL enables detailed MRI analysis."
http ://www.researchsy stems.com!A ppPro fi le/idl_med_cchips.cfm.
Schouten, M. and H. de Jong (1999). "Reduction, elimination, and levels: The case of the LTPlearning link." Philosophical Psychology 12: 237-262.
REFERENCES
227
Sco ville, W . and B. Mil ner (195 7). " Los s of recent mem ory after bilater al hippocampal
lesions." Journal of Neuro logy, Neurosurgery, and Psychiatry 20 : 11-21 .
Shepherd, G. (1994). Neu robiology , 3'd Ed . Ox ford : Oxford University Press.
Smart, J.J .C. (1959) . "Sensations and brain processes ." Philosophical Review 68 : 141- 156.
Smart, J.J .c. (1963). " Materialis m." Journa l of Philo sophy 60: 65 1-662.
Smart, J.J .C . (1967). "Comments on the papers." In C . Presley (ed .), The Identit y Theory of
Mind. SI. Lucia, Brisbane: University of Queen sland Press .
Sneed, 1. (197 1). The Logical Structure of Mathematical Physics . Dordrecht: Reidel.
Soderling, T. and V. Derkach (2000). "Postsynaptic protein phosphorylation and LTP ." Trends
in Neuroscien ce 23: 75 -80 .
Son , H., R. Hawkins, K. Mart in, M. Kiebler, P. Huan g, M. Fishman, and E.R. Kandel (1996).
" Long-term potentiation is reduced in mice that are doubly mutant in endothelia l and
neuronal nitr ic oxide synthase .' Cell 87: 1015-1023 .
Squire, L. (1 992 ). "Me mory and the hippocampus: A synthesis from findings with rats ,
monkeys, and humans." Psychological Review 99 : 195-23 1.
Squ ire , L. 1987 : Memory and Brain . Oxford University Press , Oxford.
Squire, L. and E.R. Kandel (1999). Memo ry: From Mind to Molecules. Scientific American
Library, New York .
Staubli, U., D. Chu n, and G. Lynch ( 1998). "Time-dependent reversal of long-term poten tiation
by an integrin antagon ist." Journa l of Neuroscience 18 : 3460-3469.
Stegrnuller, W. (1976). The Structure and Dynamics of Theori es. Berlin: Spri nger-Verlag.
Stephen, A. (200 1). "How to lose the mind-body problem." Grazer Philosophische Studien 61 :
279-283 .
Suppes, P. (1956) . Introduction to Logic. Pr inceton, NJ : van Nostrand.
Suppes, P. (1965). "W hat is a scienti fic theory?" In S. Morgenbesser (ed .), Philosophy of
Science Today . New York : Basic Books.
Taubenfield, S., K. Wiig , B. Monti, B. Dolan, G. Pollonini, and C. Alberini (200 1). " Forn ixdependent induction of hip pocampal CCAAT enhancer-binding protein ~ and 0 co localizes with pho sphorylated cAMP response clement-binding protein and
accompanies long -term memory consolidation.' Journal of Neuroscience 21 : 84-91.
Tully, T. and W. Quinn (1985). "Classical conditioning and retention in normal and mut an t
Drosophila melano gaster ,' Journal of Comparative Physiology A 157: 263 -277.
Tulving, E. (1983) . Elements of Episodi c Mel/wry. New York : Ox ford University Press.
Ungerleider , L. and M. Mishkin (1982). "T wo cortical visual syst em s." In D. Ingle, M .
Goodale, and R. Mansfield (eds .), Analysis of Visual Beha vior. Cambridge , MA: MIT
Press, 549-586.
Weaver, R. ( 1999). Mole cular Biology, Boston , MA : WCB/McGraw-H ill.
Williams, G . and P.S . Go ldman-R akic (1995). "Modulation o f memory fields by dopamine DI
receptors in prefrontal co rtex ." Nature 37 6 : 572-575 .
W ilson , F., S . O'Scalaidhe, and P.S . Go ldman-Rakic (1993 ). " Dissociation of object and spatial
processing domains in primate prefrontal cortex." Scien ce 260: 1955-1958.
228
Wilson, F., S. O'Scalaidhe, and P.S. Goldman-Rakic (1994) . "Functional synergism betwe en
putative gamma-aminobutyrate-containing neurons and pyramidal neurons in
prefrontal cortex." Proceedin gs of the National Academ y of Scien ces (USA) 91:
4009 -4013 .
Wimsatt, W.e. (1974). "Reductive explanation: A functional account. " In R.S . Cohen et al.
(eds.) , Philosophy of Science Association Proceedings . Dordrecht: Reidel.
Wittgenstein, L. ([1919]1961). Tractatus Logico-Philosophicus (translated by D. Pears and B.
McGuiness). London: Routledge and Kegan Paul.
Yin , J., J. Wallach, E. Wilder , J. Klingensmith, D. Dang, N. Perrimon, H, Zhou , T. Tully , and
W. Quinn (1995b). "A Drosophila CREB/CREM homolog encodes multiple
isoforrns, including a cyclic AMP-dependent protein kinase-responsive transcriptional activator and antagonist." Molecular and Cellular Biolog y 15: 5123-5130.
Yin, 1., J. Wallach, M. Del Vecchio, E. Wilder, H. Zhou , W. Quinn, and T. Tully (1994).
" Induction of a dominant negative CREB transgene specifically blocks long -term
memory in Drosophila." Cell 79 : 49-58 .
Yin , J., M. Del Vecchio, H. Zhou , and T. Tully (I 995a). "C REB as a memory modulator:
Induced expression of a dCREB2 isoform enhances long -term memory in
Drosophila." Cell 81 : 107-115 .
Zangwell , N. (1992). "Variable reduction not proven ." Philosophical Quarterly 42 : 214-218 .
Zigmond, M., F. Bloom, S. Landis, J. Roberts, and L. Squire (cds .) (1999). Fundamental
Neuroscience. San Diego, CA : Academic Press .
INDEX
alphabet, cell biological (for

learning and memory)
135, 161
amnesia, human global 63, 76-80 ,
III
case H.M. 76, 78

amygdala 78, 90-91, 93,149
analytic-synthetic distinction 3435
anti-reductionism (-ists) 18,2225,39,43,95 ,131,133,
149, 156,215
Aplysia 15,89,94, 135, 141-151,
160
attention, selective (explicit) 178189
autonomy (-ists) 8, 20, 51,103,
Ill , 116-117, 130-131
behavior (-al), data and trials 1-5,
8, 11, 14, 16,20,31,41,
43-45 ,49,51-52,59-60,
62,75-76,78,88-91,93 ,
96,99-100, 107-116,121122, 130-132, 134, 136137,139,141,143-144,
148, 150, 154, 156~157,
160-161, 164, 167, 169175, 179-181, 183-184,
186, 188, 195-196, 206,
212
contextual-conditioning
(memory) task 89-90, 9293
delayed matching-tosample task 180, 186, 188
delayed response (DR)

paradigm 166-168, 170178
percent(age) correlation,
visual stimulus 194-200,
209-210
biochemistry 112, 115, 157, 190
cognition (-tive) (higher) 2, 4-5, 7,
23,30-31,34-35,41,5051,59,76-77,81 ,107,
Ill, 116-119, 128-130,
134-135, 148, 157, 159,
161,163,165-166,171 ,
175,177,189,205-206
science (-tists) 1-6, 9, 24,
28,30-31,35,37,41 ,4344,71,99,103,107,116,
121, 130, 160, 163-164,
169, 171, 178, 180, 190
conductance, neuronal 55-57, 6566, 100
decremental55
salutatory 57
consciophile(s) 179, 184, 188-189
consciousness 2,5, 118, 151, 158159,163-166,171 ,178180, 188-190, 193-194,
205,207,210,212-213
Hard problem of 183, 189190, 212-213
content (representational,
cognitive, propositional,
semantic) 23, 30,41 ,51 ,
71,96, 107-108, 111, 117,
165-166,207-211
cortex 43, 78-79, 91-92,117-120,
126, 128-129, 135, 158,
230

167,171 -173,175-178,
180-182, 187, 190, 193194,196,199,203,206,
214
anterior cingulate (ACC)
126, 159
dorsal visual stream 170171,183,194
dorsolateral prefrontal
(DLPFC) 119-121, 126,
159,167,169-172,174,
177
frontal eye fields (PEFs)
118-123, 126-129, 159,
169, 183,213
inferotemporal (IT) 169,
177,182-183
medial superior
temporal (MST) 182,
200-202 ,214-215
middle temporal (MT)
181-182, 194-200,209,
211-212
movement fields 118-119
posterior parietal (PPC)
169-172,182-183,199
post-saccadic activity
118-123
prefrontal (PFC) 157m
169,171-173,175-178,
206,214
pre-saccadic activity
118-123
primary visual (V 1) 182,
185-187
principal sulcus (PS)
167-171, 174-177, 213
receptive field (sensory
neuron) 168-169, 176,
181-188, 194-195, 199,
203-204,214-215
somatosensory (primary,
S1) 142,203
superior frontal sulcus
120, 126
temporal (lobe) 169, 176177, 182-183, 191-194,

200
V4 181-182, 184-186, 188
ventral visual stream
176, 183,213
visual 173, 180, 182, 187,
189
Drosophila melanogaster 136138, 140-144, 146, 148,
150
dunce mutant 136-137,
150
rutabaga mutant 137,
150
eliminative materialism 13-14,
178
embryonic stem (ES) cells 85-86,
88,91
endogenesis 198
epilepsy (-tic) 191, 193
evolution (-ary) 30, 102, 132-133,
136, 148-151, 153, 155,
160,172-173
molecular 151-157, 160
natural selection 152153, 198
neutral theory of
(molecular) 151-152, 160
exclusion, causal-explanatory
108-110
existence questions
external 32-35, 39, 101102
internal 32-35,42, 115
externalism (-ist), phenomenal
190,206-210,215
facilitation, long-lasting
(synaptic) 61,89, 139140, 142-143, 145-147,
149-150
INDEX
fruitless questions
(pragmatically) 34-35, 39,
110,134,190
function(s) (-al) 2, 24, 29-30, 4 1,
43,45-46,71,73-74,8284, 86-87, 90, 94-102,
128-131,138-139,150155,160-161 ,172,175,
180,187,189-191,196197,205,211-212
functional neuroimaging 2, 41,
116, 121-122, 130, 158159
Blood Oxygenation Level
Dependent (BOLD)
signal 122, 124-127
functional magnetic
resonance imaging
(fMRI) 25, 41, 120, 122,
124-126, 129, 159
Region of Interest (ROI)
126
functionalism (-ists) 131-132,
210-211,215
gene expression 51, 54, 58, 60,
63,68-69, 71, 73-74, 88,
92-93,96, 139, 145
CCAAT enhancer
binding protein (e/EBP)
73-74,99, 145, 147-148,
160
immediate early 71-75,
139, 146-148
promoter 71, 87-88, 91,
137
response activators
(enhancers) 71-75, 138139, 155
response repressors 7172, 74-75, 99, 138-139,
148-149
ubiquitin carboxylterminal hydrolase (uch)
73,99, 146-147
231
heuristic 4, 29, 110, 114-115, 130,
157
hippocampus (-pal) 45-46, 53,
61-63,68,70, 76, 78-81,
88-93, 113
CAl region 62-63,66-67,
70,72,90,92
dentate gyrus 46,53 ,61
perforant pathway
(fibers) 52-53, 61-62
Schaffer collateral
pathway (fibers) 62-63,
66-67, 70, 90, 92
identity theory, mind-brain 1314,20,108-109,132,134,
149
intended empirical applications
(of a theory) 96-100
internalism (-ist), phenomenal
209-211
knock-outs, genetic 82-88, 214
CREB- mouse mutant
88-91
logical empiricism 9, 14,27
long term potentiation (LTP) 29,
43-46, 51-52, 62-63, 6667,70-71 ,73-74,81-82,
90-91,94-95,99,102-103,
110, 114, 150, 156
early (E-LTP) 63, 65-70,
73-75 , 82, 88, 90-91, 96,
99, 101, 137, 148
late (L-LTP) 67-75, 82,
88,90-93,96,99-101 ,
112, 137, 148
mechanism(s), cellular and
molecular 3-4, 15-15,21 ,
23,25,29-30,43-46,49,
51-53, 56-60, 62, 65-72,
74-76,82,87-91 ,93-95,
99-102, 104, 110-119,
121-123, 126, 128-130,
232
132, 134-136, 139-144,

146,148-151,154-159,
161,166,168-171 ,175176,178-179,184,186187,192,206,213
memory
consolidation (switch) 43,
46-51,62,68,75 ,82,8891, 93-96, 99-102, 104,
110-116, 136-139, 141,
144-151,155-158,165
declarative 63, 76-79, 8182, 88-93, 102, 111, 114,
137, 148
fields , working memory
165,168,171-175 ,177,
206
flashbulb 148-149
long-term 15,43,45,47,
51,62-63,67,75-76,82,
88-94,96,99, 102-103,
111-113, 136-138, 141142, 146, 148, 150, 157,
193
nondeclarative 77, 81,
90,93, 114, 136, 1M, 148,
150
short-term 15,43,45,47,
51,62-63,75-76,82,8890,92-93,96,99, Ill ,
139, 141, 143, 145-147,
157, 165
working 47,119-124,
126-129, 159, 164-166,
169-176, 178, 180, 189,
205-206,214
mental causation (problem of)
107-111,115-116,158
metaphysics (-ical) 32, 36, 38-39,
108-110, 132, 134, 158,
187,211
metascience (-tific), new wave 1,
31-32,34-37,39-40,4243,95-96,99-101 , 114
mind 1-5,7-8, 13-14,20-21 ,31 32,39-40,76,95 ,107,

109, 131-132, 165-166,
193,212-213
mind-body problem 6-8, 10, 12,
14-15,20,26,28,30,3435,40, 111, 157
intertheoretic reduction
reformulation of 15, 20,
26,28,40
models (of a theory) 12,27-28,
96-100
molecular biology 1,5,16,71,
87,98,137,144,155,157
genetics 24, 71, 92, 96,
98, Ill, 190
multiple realizability (-ation) 2025, 102, 107, 115, 131136, 145, 149, 155-158,
161
mysterian(s) 37, 163, 165, 186,
190
naturalism, philosophical 2, 5, 13,
39,102
neurocomputational model 2,
122-123, 126, 129-130,
179
biological model(ing)
121-122, 128, 159
vector subtraction 119123, 128, 129
neuroscience
cellular and molecular 16,25,31,33,36-40,43,
45-46,60, 75, 95, 99-100,
102,107,110,112,114116,131-132, 134, 141,
154, 156-158,160-161,
163-164, 176, 190,213
cognitive 2-4,37, 107,
115-116, 128-128, 158,
163
mainstream 2-3,31,3739,42-45, 52, 71, 76, 88,
INDEX
95, 100-102, 130, 134,
158,188,190,206,213
neurotransmitter(s) (-ssion) 4344,55 ,58-60,63,66-70,
74,94,99,104,138-140,
142-144, 147-149, 154,
173,178
dopamine (DA) 67, 94,
113,138-139,173-176
glutamate 63-68 , 99, 112
retrograde (nitric oxide ,
NO) 66-68
serotonin (5-HT) 138139, 143-147, 149, 174175
ontology (-ical) 7-8, 10-12, 19-29,
27-28,30-36,40,42, 101,
104
philosophy (-er) 1-11, 13,15, 20,
22,25,28-29,31-32,34,
36-39,41 ,43-44,94-95,
102,107-110,113-117,
121, 128, 130-134, 156159,161,163-165,178179,188-190,193,206210,212-213,215
analytic 7
neuro- 1,40
of biology 2, 104
of mind 1-2,5-7,9-10,
12-14, 16, 19,21,31 ,36,
43,71,94-95,102,107,
109, Ill , 115, 131-132,
157-158, 160, 163, 180,
207
of neuroscience 6, 39, 95,
103
of science 6,9-11 , 13-14,
16,21,27-28,31,36,3839,42,60,95 , 105, 116
orthodoxy in 39, 109,
157,163,207
scientific 37-39
233
structuralist program in
27,29,95 , 104
physicalism (-ists) 7-10 ,13-14,
20-21, 29-30 , 33-34 , 36,
101,109-111,132,157158
nonreductive 21, Ill,
132, 157
revisionary 20, 28
physiological (including cell-,
electro-, and neuro-) 4, 12,
20,30,52-53 ,62,67,9091,100,117-120,122123, 126, 128-129, 163,
170, 179, 181, 184, 194,
196,205 ,211
physiology (including cell- ,
electro-, and neuro-) 2, 12,
16,24,52,54,57,94,116,
121-122, 130, 159, 163164, 171, 178, 189-190,
2 11
multiplicative scaling
184-187
single-cell approach 163,
166-167, 170-171, 173,
175-181 ,186,188-189,
195,206,214
tuning curves 184-186
potential, action (spike) 54-61,
63,66, 70, 74, 113, 118,
138-139, 143, 154, 164,
181-182, 184-186, 188189,212
excitatory postsynaptic
(EPSP) 53, 58-59 , 61, 63,
66-67,70,90-92,143 ,145
membrane 52-60, 64-65
practice, (neuro-) scientific 12,
14, 22-23, 30-32 , 34-39,
44, 102, 110-112, 114115, 117, 130-131, 137,
158
234
protein synthesis 63, 68, 70-71,

74,88,93 ,98-99,139,
145-146, 148
proteins
adenylyl cyclase 64, 67,
69-70,99, 113, 138-139,
143-144
calcium-calmodulin
kinase II (CaMKII) 64,
66,91,147
calmodulin 64, 67, 138,
143, 147
cyclic AMP response
element binding protein
(CREB) 73, 88-93, 99,
113, 138-139, 144-145,
147-150, 155, 160
CREB 1 (isoforms) 71-75,
88, 139, 144-147
CREB 2 (isoforms) 71,
74-75, 137-139, 145-146,
148-149, 160
G proteins 58, 66-68, 72,
138, 143-145
mitogen activated
protein kinase (MAP K)
71, 74
protein kinase A (PKA)
66-67,69-71, 73-75,8889,91-93,99, 112-113,
137-140, 143-148, 160
psychology (-ical, -ists) 2, 4-10,
12-16,20,22-26,28-29,
31-33,38,41 ,43,46-47,
49-52,67,75-76,78,9596,99-101 , 104-105, 107,
110-116, 132-135, 148151,154,156-158,161 ,
163,165-166,171,179,
183,212
folk 8, 40, 178
neuro- 2, 4,52,61 , 76-78,
115,117,159,165,192,
214
qualia 179, 190, 193-194, 198,

200,207-213,215
receptor(s) 46,51,58-59,63-68,
70, 75, 82,94, 112, 138,
143-144, 154-155, 171,
173-176,178,191,198
AMPA 63-68, 99
dopamine 1 (D 1) 70, 173176,214
ionotropic 58, 63
metabotropic 58m 67,
138, 143-144
NMDA 46,64-67,70,79,
178
reduction 3, 5-6, 10,31,43,46,
51,75,95,99,101-102,
107,116, 130, 157-158,
206
new wave 1,6,21,26-31,
41,45
psychoneural3, 7,0,1516,20-21,23,25-26,28,
30-31,33,43,45 ,76,95 ,
101-102,104,131-132,
149, 156-157
ruthless 4-6, 31, 37, 39,
44,94, 131, 134, 156-158,
179
reduction, intertheoretic
(scientifi c) 1,6-22,24-32,
36,40-41 ,43,45 ,97-101 ,
104-105, 117, 160
combinatorial 160
concepts structured
through 95-96, 98-100,
102
in-practice 31, 36, 43, 95,
100-102, 105, 107
links, ontological 27, 104
thermodynamics to
statistical mechanics and
microphysics, of 9, II ,
19-20, 22-23, 27-28
INDEX
reductionism (-ists) 1,5,21 -24,
29-30,36,45 ,60-61 ,94,
100, 102, 107, 112, 114115,131 ,134,139,149,
164,173,175 ,178-179,
189-190,213-215
epiphany 59
rehearsaI47,51 ,62,67,95,111 113,165
repetition, stimulus 47, 51, 62, 67,
94, 112, 141, 144, 156
representation(s) 51, Ill , 1I7,
128, 171, 180, 184, 198,
200,211
retrograde interference 47-49,
41, 75-76, 99, 111-113,
156
revision, conceptual 8, 19
saccade(s) 118-129, 159, 167-168,
170-171,174-176,195 ,
198-201,210,213
sequences (-ing) 119,
121-129, 159
scientific realism (-ists) 37-38
second messenger (signaling
pathways) 51, 58, 64, 67,
91,137,144,146-147,
149, 154-155
cyclic adenosine
monophospate (cAMP)
64-71,89,91-93,99,113,
131, 137, 139, 143-144,
146-148, 150, 155
sensitization 141, 143-146, 148
Society for Neuroscience 4,212
stimulation
cortical 190-193, 196,
199
micro- 164, 196-206,209212,214-215
temperature (classical
thermodynamics) 8-9, 2223
235
transdisciplinary research 116117,128-131
transgenics 51, 82-85, 87-88, 9193,99, 112, 137-138, 161
R(AB) (PKA) mouse
mutants 91-93, 99
Twin Earth 206-208, 213
Inverted Earth 207-210,
215
water maze, Morris 46, 79-80,89,
92

(Studies in Brain and Mind 2) John Bickle (Auth.) - Philosophy and Neuroscience - A Ruthlessly Reductive Account-Springer Netherlands (2003)

Uploaded by

Copyright:

Available Formats

(Studies in Brain and Mind 2) John Bickle (Auth.) - Philosophy and Neuroscience - A Ruthlessly Reductive Account-Springer Netherlands (2003)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Studies in Brain and Mind 2) John Bickle (Auth.) - Philosophy and Neuroscience - A Ruthlessly Reductive Account-Springer Netherlands (2003)

Uploaded by

Copyright:

Available Formats

Philosophy and Neuroscience

Studies in Brain and Mind

Series Editors '

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

Library of Congress Cataloging-in-Publication Data

Printed on acid-free paper

AlI Rights Reserved

"And yet, there are philosophers who refuse to acknowledge scientific

Chapter Two: Reduction-in-Practice in Current Mainstream Neuroscience 43

Chapter Three: Mental Causation, Cognitive Neuroscience, and Multiple

Chapter Four: Consciousness

As a work that combines philosophy and (what was then) current

PHILOSOPHY AND NEUROSCIENCE

fare of science fiction, but of Cell, Journal of Neurophysiology, Science,

PHILOSOPHY AND NEUROSCIENCE

Faucher (University of Quebec at Montreal), Owen Flanagan (Duke

FROM NEW WAVE REDUCTION TO NEW

This book is about contemporary neuroscience. More specifically, it

1 WHY CELLULAR AND MOLECULAR

PHILOSOPHY AND NEUROSCIENCE

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

PHILOSOPHY AND NEUROSCIENCE

"further down" to cellular, synaptic , and ultimately molecular biological

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

PHILOSOPHY AND NEUROSCIENCE

work (including some who describe their research as "philosophy of

2 BACKGROUND: THE INTERTHEORETICREDUCTION REFORMULATION OF THE MIND BODY

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

PHILOSOPHY AND NEUROSCIENCE

the autonomy of a theory's objects and properties (e.g., electrical

It has also been fashionable among some physicalists to claim that

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

PHILOSOPHY AND NEUROSCIENCE

Reformulated in light of Nagel's theory , the traditional philosophical

3 REVOLTS AGAINST NAGEL'S ACCOUNT

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

PHILOSOPHY AND NEUROSCIENCE

the underlying logical empiricist account of theory structure, Patrick Suppes

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

Feyerabend's account of scientific reduction, "eliminative materialism"

PHILOSOPHY AND NEUROSCIENCE

scientific analogy that became prevalent in later eliminativist writings : the

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

3.2 Schaffner's General Reduction (-Replacement) Paradigm

PHILOSOPHY AND NEUROSCIENCE

Reduced and reducing theories do not involve laws akin to those in

It is by virtue of this last feature that Schaffner's GRR conditions of

3.3 Hooker's General Theory of Reduction

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

Tn (& boundary conditions, limiting assumptions, as needed)

e.g. , (x)(Jx -7 Kx), (x)((Kx & Lx) -7 Mx)

PHILOSOPHY AND NEUROSCIENCE

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

PHILOSOPHY AND NEUROSCIENCE

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE

4 EXTENDING HOOKER'S INSIGHTS: NEW WAVE

PHILOSOPHY AND NEUROSCIENCE

principles are consistent with the multiple realizability of T R posits within T B.

NEW WAVE REDUCTION TO NEW WAVE METASCIENCE