when epistemic closure does and does not fail 35

References
Beall, Jc. Forthcoming. True and false – as if. In New Essays on Non-Contradiction, ed.
G. Priest, Jc. Beall and B. Armour-Garb. Oxford: Oxford University Press.
Dummett, M. 1978. Truth. In his Truth and Other Enigmas. London: Duckworth.
Dunn, J. M. 2000. Partiality and its dual. Studia Logica 65: 5–40.
Field, H. Forthcoming. Is the liar sentence both true and false? In Deflationism and
Paradox, ed. Jc Beall and B. Armour-Garb. Oxford: Oxford University Press.
Parsons, T. 1990. True contradictions. Canadian Journal of Philosophy 20: 335–54.
Priest, G. 1987. In Contradiction: A Study of the Transconsistent. The Hague:
Martinus Nijhoff.
Sainsbury, R. M. 1995. Paradoxes. Cambridge: Cambridge University Press.
Varzi, A. 1999. An Essay in Universal Semantics. Boston: Kluwer Academic Publishers.

When epistemic closure does and does not fail:

a lesson from the history of epistemology
Ted A. Warfield

Does the failure of a necessary condition for knowledge to be closed under

known entailment imply that knowledge itself is not closed under known
entailment? More generally, does the failure of a necessary condition for
knowledge to be closed under some relation R imply that knowledge itself
is not closed under relation R? An examination of the recent history of epis-
temology would lead one to think that the answer to these questions is
‘yes’. I show in this note that the correct answer to these questions is ‘no’.
Those who answer ‘yes’ are committing the fallacy of composition. I also
show that the mistake merits correction because it concerns important
matters and occurs quite frequently.
Near the beginning of his impressive early overview of issues concerning
epistemic closure principles and their possible use in sceptical argumenta-
tion, Anthony Brueckner asserts that ‘Knowledge is closed under known
logical implication only if each necessary condition is so closed’ (1985: 91).
It is likely that in making this remark Brueckner was taking his cue from
earlier work on epistemic closure by Robert Nozick. Nozick, (1981/1998),
had defended his well known ‘tracking’ account of knowledge, which in its
simplest form says that:
S knows that P if and only if
(1) P is true.
(2) S believes that P.

Analysis 64.1, January 2004, pp. 35–41. © Ted A. Warfield

36 ted a. warfield

(3) If P weren’t true, S wouldn’t believe that P.

(4) If P were true, S would believe that P (159–63).1
Nozick famously moved on to argue that given this account of knowledge,
knowledge fails to be closed under known implication. Nozick traced
this failure to a failure of closure for one of his necessary conditions on
knowledge:
This failure of knowledge to be closed under known logical implica-
tion stems from the fact that condition 3 is not closed under known
logical implication; condition 3 can hold of one statement believed
while not of another known to be entailed by the first. It is clear that
any account that includes as a necessary condition for knowledge the
subjunctive condition 3 … will have the consequence that knowledge
is not closed under known implication. (1981/1998: 171–72)
Brueckner is not the only philosopher to have followed Nozick in (appar-
ently) claiming that if a necessary condition on knowledge fails to be closed
under known entailment then knowledge is not closed under known entail-
ment. To give only one more example, Ernest Sosa seems to have recently
endorsed this same inference (Sosa 1999: 149).2
Sosa compares Nozick’s condition (3), which Sosa calls ‘Sensitivity’, to
a necessary condition for knowledge that Sosa prefers to it, his much-
discussed ‘Safety’ condition, where Safety requires for knowledge that ‘S
would believe that P only if it were so that P’.3 After arguing persuasively
that Safety has advantages over Sensitivity, Sosa considers the following
hypothetical objection: ‘Doesn’t the requirement of safety share with the
requirement of sensitivity the drawback that it makes knowledge not
closed under deduction?’ (149). Sosa responds to this question in the affir-
mative, strongly suggesting that he agrees that the failure of a necessary
condition on knowledge to be closed implies the failure of knowledge to be
closed.4
1
Here and elsewhere in this essay I refer to the page numbers from the reprinted selec-
tions from Nozick referenced in the bibliography.
2
I point to Nozick, Brueckner and Sosa as representative prominent epistemologists
who I believe have made the mistake I am correcting. Many others have also done so.
The frequency with which I have encountered the error in recent discussions with
colleagues about epistemic closure prompted this paper.
3
Sosa’s Safety is not equivalent to Nozick’s Sensitivity because counterfactuals do not
validly contrapose. Like Nozick, Sosa ultimately offers refinements to his ‘tracking’
necessary condition on knowledge. In neither case do the refinements matter for pur-
poses of evaluating the inference under discussion.
4
Sosa goes on in this article (and elsewhere in his recent work) to refine his under-
standing of the safety condition and to refine his overall account of knowledge. An
evaluation of Sosa’s full views on knowledge is beyond the scope of this paper.
 when epistemic closure does and does not fail 37

Epistemic closure principles continue to receive much attention in con-

temporary epistemology. Epistemic closure principles are most frequently
discussed when evaluating a certain style of sceptical argument and they
are occasionally discussed as a topic of independent interest.5 Closure
principles also arise in discussions of epistemic logic, the epistemology
of paradoxes, and elsewhere in epistemology (most notably in recent dis-
cussions of epistemic contextualism). Because epistemic closure principles
are an important part of so many debates within epistemology and because
of their seemingly central role in discussions of scepticism and knowledge,
it is important to be clear about how to evaluate closure principles
properly.
The remarks of Brueckner, Nozick and Sosa suggest that one way to eval-
uate whether knowledge is closed under known entailment is to find out
whether this closure relation holds for some necessary condition for knowl-
edge: if the necessary condition is not closed under known entailment then
neither is knowledge. This appears to be exactly what happens in many dis-
cussions when philosophers point out that the following principle, express-
ing the closure of knowledge under known implication, is false:
Closure:
"p"q ([Kp & K(p implies q)] … Kq)
Philosophers noting that Closure is false will sometimes point out (cor-
rectly) that one could know P and know that (P implies Q) while not even
believing Q. This explanation of the failure of Closure is a proper expla-
nation. A mistake occurs when some explain the failure of Closure by
claiming that it fails ‘because belief [a necessary condition for knowledge]
is apparently not closed under known logical implication’.6 It is true that
the failure of belief to be closed under known entailment is relevant to the
evaluation of Closure.7 It is not correct, however, that the failure of belief
to be closed under known entailment is sufficient for the failure of Closure.
The following argument is invalid:
(P1) Belief is not closed under known entailment.
(P2) Belief is a necessary condition for knowledge.
(C1) So, Knowledge is not closed under known entailment.
Similarly, if Nozick’s argument against the closure of knowledge is under-
stood as follows (using ‘Sensitivity’ as a name for Nozick’s third condition),

5
See David and Warfield (forthcoming a) for a detailed discussion of the use of
epistemic closure principles in sceptical arguments and (forthcoming b) for some key
challenges to the most plausible epistemic closure principles.
6
I here quote Brueckner (1985: 91) as one example (of many available) of this error.
7
If belief were closed under known entailment then the counterexample given
in the proper explanation of the failure of Closure noted above would not be a coun-
terexample.
38 ted a. warfield

(P1) Sensitivity is a necessary condition for knowledge.

(P2) Sensitivity is not closed under known entailment.
(C1) So, Knowledge is not closed under known entailment.
then it too is invalid.8
The problem I am pointing to is not simply a problem about argu-
ments concerning closure under known entailment. Just as epistemologists
working on the many problems in which issues concerning epistemic
closure arise are interested in a variety of purported necessary conditions
on knowledge (examples: belief, Safety, Sensitivity, justification), they are
also interested in evaluating epistemic closure principles distinct from
Closure. The remarks of Brueckner and Nozick about failures of closure
suggest the following general argument form which would, if valid, be of
great use in evaluating proposed epistemic closure principles:9
Where X is a proposed necessary condition on knowledge and R is a
specified closure relation,
(P1) X is a necessary condition for knowledge.
(P2) X is not closed under R.
(C1) So, Knowledge is not closed under R.
This type of argument, however, is not of any use to epistemologists.
This argument form (of which the two previously displayed arguments
are instances) is invalid. Arguments of this form commit the fallacy of
composition.
The failure of some necessary condition on knowledge to have some
property (example: closure under R) does not imply that knowledge itself
does not have the property. Knowledge could have the property in ques-
tion, consistent with some necessary condition on knowledge not having
the property, in a variety of ways. In one simple scenario, knowledge could
have the property while some necessary condition for knowledge lacks the
property, if knowledge has the property in virtue of its possession by some
distinct necessary condition for knowledge.10 In an alternative scenario,
knowledge could have the property without any (proper) necessary
condition on knowledge having the property if the property attaches to
knowledge because of the interaction of two or more necessary conditions

8
The same is true, though I will not display the argument, concerning Sosa’s remarks
about his Safety condition.
9
Alternatively, for those who take it as an adequacy condition on proposed accounts
of knowledge that they imply (or at least be consistent with) the truth of some key
epistemic closure principle, the argument form would, if valid, be of use in evaluat-
ing proposed accounts of knowledge.
10
Of course, the possession of the property by a necessary condition for know-
ledge would not alone imply that knowledge has the property. This inference is also
fallacious.
 when epistemic closure does and does not fail 39

on knowledge. These possible ways for knowledge to have a property

while some necessary condition for knowledge lacks the property are
obvious and do not exhaust the ways that this situation could obtain. I
suspect that once the composition fallacy involved in the fallacious argu-
ments displayed above is noted, these further points are obvious and
uncontroversial.
For concreteness, I now show how the failure of two specific alleged nec-
essary conditions on knowledge to be closed under known entailment does
not suffice for the conclusion that Closure, as formulated above, is false.
This illustration is somewhat artificial because, as noted earlier, Closure is
false. The illustration shows that specific arguments for this conclusion are
fallacious: the illustration does not show that the conclusion of such argu-
ments is incorrect.
Assume (as is certainly the case) that belief is a necessary condition on
knowledge and that belief is not closed under known entailment. These
claims alone do not imply that Closure is false because it could be that a
further necessary condition on knowledge is that one believe that which is
known to be entailed by that which is known. The point here, more for-
mally, is that the following claims are consistent:

(5)
"p (Kp … Bp)

– knowledge implies belief
(6) ~
"p"q ([Bp & K(p implies q)] … Bq)
– belief not closed under known entailment
(7)
"p"q (Kp … [K(p implies q) … Bq])
– knowledge requires believing known conse-
quences of propositions known

No one would seriously assert that (7) expresses a necessary condition on

knowledge, but this is irrelevant. The point is that the conjunction of (5)
and (6) does not imply that Closure is false. (7) is not precluded by the con-
junction of (5) and (6), and (7) blocks counterexamples to Closure that one
might hope to find in (6). Similarly, assume (controversially) that Nozick’s
Sensitivity is a necessary condition for knowledge and that Sensitivity is not
closed under known implication. These claims alone do not imply that
Closure is false because the following claims are consistent (using Æ for
the counterfactual conditional):
(8)
"p (Kp … (~p Æ ~Bp))
– Sensitivity is necessary for knowledge
(9) ~
"p"q ([K(p implies q) & (~p Æ ~Bp)] … (~q Æ ~Bq))
– failure of Sensitivity closure
(10)
"p"q (Kp … [K(p implies q) … (~q Æ ~Bq)])
– knowledge requires Sensitivity to known conse-
quences of propositions known
40 ted a. warfield

Once again, no one would seriously propose (10) as a necessary condition

on knowledge. The condition is put forward only to show that the con-
junction of (8) and (9) does not suffice for showing that Closure is false. As
in the previous example, the conjunction of (8) and (9) does not imply that
(10) is false, and (10), if true, would block counterexamples to Closure one
might hope to find in (9).
The complexity of these examples might suggest that showing that
knowledge is not closed under some relation is a rather difficult matter, but
this is certainly not the case. The complexity reveals only that one should
not evaluate closure claims about knowledge by examining closure claims
about necessary conditions for knowledge. Earlier we saw one example of
how simple a proper explanation of the failure of a closure principle can
be. We saw that Closure fails because one can know that P, know that P
entails Q, and yet fail to believe that Q thereby failing to satisfy a neces-
sary condition for knowing that Q.
A proper explanation of the failure of any proposed epistemic closure
principle will resemble this explanation in the following way. The expla-
nation will display that the antecedent of the principle can be satisfied while
simultaneously some necessary condition for the truth of the consequent of
the principle is not satisfied.11 This is what is needed to show that knowl-
edge is not closed under some relation. Claiming that knowledge is not
closed under some relation because a necessary condition for knowledge is
not closed under the relation is a mistake of the past that future episte-
mologists should not repeat.12

University of Notre Dame

Notre Dame, IN 46556, USA
[email protected]

References
Brueckner, A. 1985. Skepticism and epistemic closure. Philosophical Topics 13: 89–117.
David, M. and T. A. Warfield. Forthcoming a. Knowledge-closure and scepticism. In
Epistemology: New Philosophical Essays, ed. Q. Smith. Oxford: Oxford University
Press.
David, M. and T. A. Warfield. Forthcoming b. Six possible counterexamples to one or
two epistemic closure principles.

11
To his credit, Nozick on one occasion explained the failure of Closure (given his
account of knowledge) correctly. He explained that one could satisfy the antecedent
of the closure principle for some P while not satisfying the ‘sensitivity’ condition nec-
essary for satisfying the principle’s consequent for some Q entailed by P (see the top
of Nozick 1981/1998: 171).
12
I thank Tom Crisp, Marian David and Leopold Stubenberg for helpful discussion. My
employer generously provided a computer and printer but provided no other support.
 brewer on the mckinsey problem 41

Nozick, R. 1981/1998. Philosophical Explanations. Cambridge, Mass.: Harvard

University Press. Parts reprinted as: Selections from Philosophical Explanations.
In Scepticism: A Contemporary Reader, ed. K. DeRose and T. A. Warfield. 1998.
Oxford: Oxford University Press.
Sosa, E. 1999. How to defeat opposition to Moore. Philosophical Perspectives 13:
141–53.

Brewer on the McKinsey problem

Anthony Brueckner

Bill Brewer (in Brewer 1999 and Brewer 2000) presents a solution to the
problem for externalism about mental content that was first set by Michael
McKinsey in McKinsey 1991.1 To see the problem, consider this argument,
regarding belief of propositions involving concepts purporting to apply to
natural kinds:
(e1) I believe that p.
(e2) If x believes that p, then x’s environment contains (or did contain)
C.
Therefore
(e3) My environment contains (or did contain) C. (Brewer 2000: 428)
It would seem that ‘an adequate account of self-knowledge’ would entail
that my knowledge of (e1) is non-empirical, in the sense that ‘neither its
acquisition nor its status as knowledge necessarily involves any specific
empirical investigation’ (Brewer 2000: 416). Content externalism would
seem to entail that for some p and C, knowledge of (e2) ‘can be derived
from non-empirical philosophical reflection upon the necessary conditions
upon determinate empirical belief possession’ (Brewer 2000: 416).2 The
upshot is that ‘the truth of content externalism – in the presence of an ade-
quate account of self-knowledge – enables a person to derive … [(e3)]
without any empirical investigation whatsoever’ (Brewer 2000: 416). But
‘such non-empirical knowledge of empirical facts is intuitively intolerable’
(Brewer 2000: 416).
Brewer’s solution to the problem is to hold on to content externalism
while rejecting the view that my knowledge of the premise (e1) is ‘wholly

1
Strangely enough, Brewer refers neither to McKinsey’s original paper, nor to any of
the numerous papers in this journal spurred by McKinsey’s argument.
2
Brewer considers the case where p = Water is wet and C = water.
Analysis 64.1, January 2004, pp. 41–43. © Anthony Brueckner