2019 UNESCO AI SustDev

Principles for
RTIFICIAL INTELLIGENCE
Towards a Humanistic Approach ?
ARTIFICIAL
INTELLIGENCE
FOR SUSTAINABLE
DEVELOPMENT:
challenges and opportunities
for UNESCO’s science and
engineering programmes
Coordination, review and and editing: Susan Schneegans
Graphic design: Aurélia Mazoyer
Cover design: Aurélia Mazoyer
Ref: SC/PCB/WP/2019/AI
ARTIFICIAL
INTELLIGENCE
FOR SUSTAINABLE
DEVELOPMENT:
Working paper
August 2019 (Revised)
Contents 3.3. The implications of AI for freshwater
management 17
3.3.1. Machine-learning algorithms already in use
in water science 17
3.3.2. The advantages of a three-pronged approach
to urban water management 17
1
3.4. The implications of AI for disaster risk
reduction 18
3.4.1. Extending AI beyond disaster response to
prevention 18
3.4.2. AI already being used to manage water-
related hazards 18
INTRODUCTION4 3.4.3. AI can improve climate change assessments 18
2 4
FOSTERING QUALITY DATA FOR
THE USE OF AI IN SCIENTIFIC RESEARCH OPTIMUM RESEARCH 22
AND EXPERIMENTAL DEVELOPMENT 8
4.1. Ensuring quality data and data management 22

2.1. The contribution of AI to scientific discovery 8
4.2. The Internet of Things can help bridge the
2.2. What might the future hold? 10 scientific divide 23
5
2.2.1. Could major scientific discoveries come
from AI systems? 10
2.2.2. Quantum computing may be the next frontier 11
3
THE USE OF AI FOR ENVIRONMENTAL
ISSUES AT THE SCIENCE–POLICY NEXUS
5.1. Smart solutions are already here

26
26
MANAGEMENT AND DISASTER RISK 5.2. How neutral is technology? 26
REDUCTION14
5.3. AI may widen the technological divide 27
3.1. The implications of AI for ecosystem and 5.3.1. The genie is already out of the bottle 28
environmental management 14
5.3.2. Open source and open science movements
can make a difference 31
3.2. The implications of AI for geosciences 16
3.2.1. Reducing subjectivity in interpreting image logs 16 5.4. AI: the new space race? 32
3.2.2. Big data, cloud computing and AI in the
service of sustainable development 16 5.5. What will greater automation mean
for low-skilled jobs? 32
3.2.3. Enhancing the visitor’s experience 16
2 I Artificial intelligence for sustainable development: challenges and opportunities for UNESCO’s science and engineering programmes
7
5.6. Growing competition for skills 35
5.7. The need for skills to keep pace

with technology36
5.7.1. A trend towards greater investment in R&D 36
5.7.2. An underexploited pool of talent 38
5.7.3. The case for investing in cross-disciplinary
CONCLUSION 52
education and new knowledge creation 39
5.7.4. An integrated approach to science teaching 41
5.7.5. Building awareness of AI and critical thinking 43 BIBLIOGRAPHY 54
6
GLOSSARY OF TERMS 56
THE IMPLICATIONS OF AI FOR POLICY-

MAKING AND FORESIGHT 48
6.1. A growing number of national AI strategies 48
6.2. Existing policy tools can frame AI strategies 48
6.3. AI systems can contribute to policy-making 49
6.4. Regional economic communities can help

to frame AI strategies 49
6.5. Global strategies will also be necessary to

ensure collective choices 49
Artificial intelligence for sustainable development: challenges and opportunities for UNESCO’s science and engineering programmes I 3
Introduction
1. Introduction
We are living through what has been termed the Fourth

Industrial Revolution, a fast-paced socio-economic
transformation driven by rapid technological progress that
is disrupting entire industries and governance systems. Also
termed Industry 4.0, it follows on the First Industrial Revolution
in the 18th century (mechanization), driven by the invention of
the steam engine; the Second Industrial Revolution (mass
production) in the 19th century, driven by the invention of
electricity; and the Third Industrial Revolution (automation)
in the 20th century, driven by the invention of electronics and
information technology.
Since the second half of the 20th century, biotechnology,

nanotechnology, cognitive science and information technology
have converged, blurring the frontier between the physical,
the digital and the biological as they spawned new research
fields such as nanobiotechnology informatics. The advent of
the World Wide Web in 1993 heralded the dawn of the digital
economy, which has revolutionized commercial transactions
and business models in fields ranging from media and
entertainment to retail and healthcare.
Digital technologies include big data and analytics, robots

and other cyberphysical systems, the Internet of Things,
Blockchain technology and artificial intelligence. The term
‘artificial intelligence’ was actually coined back in 1956, at a
workshop at Dartmouth College, New Hampshire, in the USA,
three years before the first microchip was patented.
Progress in artificial intelligence was initially slow, hampered

by an insufficient volume of data and computer power. Artificial
intelligence first caught the public eye in 1997, when IBM’s
Deep Blue became the first computer to beat the grandmaster
of chess, Russian legend Garry Kasparov. Twenty years on,
Sophia, a humanoid robot developed by Hanson Robotics,
gave her first interview to CNBC television’s Andrew Ross
Sorkina at the Future Investment Institute in Saudi Arabia in
October 2017, prompting the host country to grant ‘her’ Saudi
citizenship1.
UNESCO’s own International Hydrological Programme began

developing an application of artificial intelligence in 2005, in
partnership with a team at the University of California, Irvine
(USA). The UNESCO G-WADI Geoserver application uses an
artificial neural network algorithm to estimate real-time
1 See: https://www.youtube.com/watch?v=S5t6K9iwcdw
Introduction
precipitation worldwide. The application is being used to The Apple iPad2 tablet computer is more powerful than the
inform emergency planning and management of hydrological Cray-2 supercomputer, which was the fastest on the planet
risks, such as floods, droughts and storms. It tracked the at the time the Human Genome Project got under way in
Haiyan Super Typhoon as it approached the Philippines 1990. The new research frontier is quantum computing, a
in 2013, for example. The Namibian Drought Hydrological field which promises a future of unprecedented computing
Services use the geoserver to prepare daily bulletins with speed. Scientists from the Abdus Salam International Centre
up-to-date information on flood and drought conditions for for Theoretical Physics are participating in this adventure and
local communities. training their peers from developing countries through the
new Trieste Institute for the Theory of Quantum Technologies.
Given the speed with which artificial intelligence is evolving

and its potential to disrupt the economy – and, thus, peoples’
livelihoods –, it is imperative that UNESCO position itself
rapidly to ensure that it can provide all of its Member
States with the necessary foresight to take advantage of
this technological revolution while ensuring respect for
human dignity and security. Artificial intelligence holds
great promise for building inclusive knowledge societies and
helping countries reach their targets under the 2030 Agenda
for Sustainable Development but it also poses acute ethical
challenges.
Developing countries will only be able to reap the benefits of

the Fourth Industrial Revolution, if they can overcome the skills
shortage that hamstrings their national innovation systems.
They will need to introduce new, out-of-the box teaching and
training techniques that value cross-disciplinary skills and
creative thinking, in order to produce a technically savvy,
innovative workforce. They will also need to adopt policies
and programmes which nurture quality basic research and
an innovation culture.
© UNESCO
The current paper sets out to identify emerging opportunities

and challenges arising from the rapid development of artificial
intelligence and their implications for UNESCO’s programmes
Sophia and friend
in natural sciences. As part of this foresight exercise, the
Natural Sciences Sector surveyed relevant staff and UNESCO
chairholders and centres. Respondents were asked to identify
Scientists working with UNESCO use a variety of digital tools to
opportunities and challenges for UNESCO’s work in the
analyse vast quantities of data in fields ranging from theoretical
following areas:
physics to ecological research, water management and
climate change assessments. Scientists are using the Internet ◼◼ Scientific research and experimental development
of Things for instance, to improve urban water management
◼◼ Ecosystem and environmental management
(smart water systems) and, in biosphere reserves, to monitor
biodiversity with drones and motion sensors. AI components ◼◼ Science governance and foresight
can be incorporated in these and other connected objects.
The findings of this survey have informed the current paper.
Artificial intelligence is now evolving at breakneck It is one of several working papers prepared by UNESCO’s
speed, thanks to increasingly powerful supercomputers. programme sectors in 2018–2019.
The use of AI in
scientific research
and experimental
development
iStock/Getty Images Plus

The use of AI in scientific research and experimental development
2. The use of AI in scientific research and

experimental development
2.1. The contribution of AI to Even when a person has a mathematical model, such as a set
of accurate equations, they can use AI to achieve comparable
scientific discovery results 10,000 times faster. ‘Say you have a new molecular
structure and you want to know how it’s going to behave in
Today, researchers produce massive quantities of information some environment for pharma exploration’, says Singer.
and data. In the USA, researchers produced 321,846 papers in ‘There are very good predictive models on how it will behave
international journals in 2014 alone, according to the Thomson [that do not use AI]. The problem is that those models take a
Reuters Web of Science, Science Citation Index Expanded tremendous amount of computation and time – it might take
(UNESCO, 2015). Researchers must wade through numerous you weeks to try just one combination’ (Singer, 2018).
papers to keep abreast of developments in their field, including
The predictive capabilities of AI are being used to predict
papers published by their peers abroad. Researchers are also
outbreaks of disease in various countries. The start-up
confronted with a deluge of experimental data that is growing
Artificial Intelligence in Medical Epidemiology (AIME) analyses
at a mindboggling pace. No researcher can digest such an
local government datasets in combination with satellite image
avalanche of information and data – but today’s machines can.
recognition systems. AIME has teamed up with the Brazilian
NGO Viva Rio to provide low-cost predictions of where to expect
Today’s machines can self- a greater incidence of disease in the coming three-month
period. The project has proved so successful that AIME has
learn and self-correct since been deployed in the Dominican Republic (WWWF, 2017).
Today’s machines not only feed on big data. They can also In the past couple of years, explains Singer, ‘new machine
self-learn and self-correct. Through a process known as deep learning methods have emerged for learning how to learn.
learning, AI systems learn from processing large training data These technologies are tackling an almost endless realm of
sets until they can’ see’ patterns and spot anomalies in large, options, like all the possible mutations in human DNA, and are
complex datasets. In an interview last year, Intel Vice President using exploration and meta-learning techniques to identify the
Gadi Singer said he considered using AI to reshape scientific most relevant options to evaluate’ (Singer, 2018).
exploration to be his most important challenge (Singer, 2018). A popular AI method in life science research provides a
Using deep learning, a machine can identify very faint patterns powerful tool for surveying and classifying biological data.
within a large, multidimensional dataset. A typical example Deep-learning algorithms take raw features from an extremely
is the researcher who knows what they are looking for but large, annotated dataset, such as a collection of images or
cannot define the exact mathematical equation. The dataset genomes, and use them to create a predictive tool based on
is too large for trial and error and there are not enough patterns buried inside. Once trained, the algorithms can apply
known features to justify using big-data analytical techniques that training to analyse other data, sometimes from vastly
to search for a pattern. The researcher has tagged several different sources (Nature, 2018). In medical diagnostics,
examples. He or she feeds them into the deep learning system a machine can learn to identify bone fractures if it is fed a
then leaves it to learn what to look for, until it ultimately sufficient number of x-ray images, a skill which could help to
identifies a pattern (Singer, 2018). compensate for the shortage of radiologists in many countries.
Figure 1.
Key components of artificial intelligence
The accuracy of machine
learning techniques has been
called into question… An AI
algorithm will always produce
Algorithm an answer but not always the
same answer to the same
question
AI The accuracy of machine learning techniques has also been
Computer Big Data called into question. Statistician Dr Genevera Allen has
power observed that these techniques have no way of saying ‘I don’t

know’ or ‘it’s not clear’. This means that an AI algorithm will
always produce an answer but not always the same answer to
the same question. She told BBC news that ‘there is general
recognition of a reproducibility crisis in science right now’ and
Source: design by S. Schneegans/UNESCO urged researchers in the field of machine learning to develop
systems which can assess the accuracy of their own findings
AI is already playing an important role in research in data- (Torbert, 2019). Dr Antonello Scardicchio from UNESCO’s
intensive fields such as medicine, genetics, cell and systems Abdus Salam International Centre for Theoretical Physics
biology and neuroscience. For instance, IBM’s Watson (ICTP) concurs. ‘This is, indeed, a problem for reproducibility
computer achieved what may be a world first at the University in science and an even bigger problem when it comes to
of Tokyo in 2015. It diagnosed a Japanese patient with a rare decisions regarding health, legal and financial issues, etc.,
type of leukemia and identified a life-saving therapy much as there are times when nobody can tell how an AI algorithm
faster than if doctors had examined the genetic data manually. has arrived at a given decision’.
The computer compared 25,000 clinical tests, 1.2 million
Innovation in medicine, genetics, cell and systems biology and
patient files and two million scientific articles (Otake, 2016).
neuroscience is increasingly being coupled with information
technology. This convergence raises profound philosophical
The private sector is driving questions, such as: what does it mean to be human? Should AI
be used only to correct a medical condition or handicap (such as
much of research in AI paralysis), or should it be used to enhance human beings? There
are start-ups which are already striving to create human hybrids
The private sector is driving much of research in AI. For instance,
or ‘cyborgs’. Elon Musk, for instance, believes that human
Cognetivity Neurosciences Ltd in the UK is developing the Integrated
beings must merge with AI to become hybrids, lest we become
Cognitive Assessment (ICA), a platform which uses AI and machine
the equivalent of pets to AI systems, or even irrelevant. His start-
learning techniques to test cognitive function, potentially allowing
up, Neuralink, is striving to integrate AI with the human brain
early diagnosis of dementia. ICA eliminates learning bias, since it
using a ‘neural interface’. The latter is a mesh which would
tests cognitive function rather than memory. A person can take the
serve as a wireless brain–computer interface, circulating
test repeatedly over time, including on their Ipad2.
through the host’s veins and arteries (Hinchcliffe, 2018).
One impediment to the use of AI in medical diagnostics is the
difficulty encountered by developers in obtaining the reams of AI brain implants raise a host of ethical issues. What if a
data necessary to establish effective algorithms. Data privacy brain implant became capable of programming a person’s
rules governing patient–doctor confidentiality may prohibit thoughts and, thereby, of controlling them? Up until now, it has
hospitals from sharing medical images, for instance, such been possible to limit a person’s freedom of movement and
as patient scans and x-rays, with a company interested in freedom of speech. Brain implants could extend this power
developing a diagnostic tool. to controlling the mind. Potentially, a person’s thoughts could
even be controlled by a machine, rather than by a human being
2 See the video at: https://www.cognetivity.com/ (Hinchcliffe, 2018).
2.2. What might the future hold? hypothesis generation and automated experimental processes’
(Kitano, 2016).
Could AI systems be designed to ask the ‘right’ questions

that may lead to major scientific discoveries? Kitano (2016)
2.2.1. Could major scientific discoveries come
questions the importance of this step in basic research. He
from AI systems? argues that budget, competition and resource constraints
In a mid- to long-term perspective, AI could fundamentally such as the length of a person’s active research career already
limit efficiency in basic research. When time and resources
alter the way in which research is conducted, structured and
are abundant, he suggests, asking the right questions is of
understood. This evolution will require scientists and engineers
lesser importance. He also recalls the role played by chance
to learn new skills to keep up and could eventually result in AI
in scientific discovery, suggesting that ‘the critical aspect of
outcompeting research organized by, and for, humans.
scientific discovery is how many hypotheses can be generated
and tested, including examples that may seem highly unlikely’.
AI is already mimicking the Kitano (2016) speculates that AI systems could be placed at the
scientific method centre of a network of intelligent agents comprising both other
AI systems and humans to coordinate large-scale biomedical
research. He takes this hypothesis a step further, suggesting
AI is already mimicking the scientific method by generating
that scientific discovery could not only combine AI and qualified
hypotheses and verifying them through simple experiments.
researchers but also involve citizen scientists who would
A decade ago, Ross King and his colleagues developed a undertake distinct tasks to produce a collaborative form
systematic robot scientist that can infer possible biological of intelligence, or crowd intelligence. He cites the example
hypotheses and design simple experiments using a defined of Patientslikeme, a patient-powered research network.
protocol automated system to analyse orphan genes in ‘Whether this path would ultimately make our civilization more
budding yeast. ‘While this brought only a moderate level of robust (by facilitating a series of major scientific discoveries)
discovery within the defined context of budding yeast genes, or more fragile (due to extensive and excessive dependence
the study represented an integration of bioinformatics-driven on AI systems) is yet to be seen’ (Kitano, 2016).
2.2.2. Quantum computing may be the next

frontier
Researchers in computer science and physics see quantum
computing as being the next frontier. Quantum computing
promises a future of unprecedented computing speed. This
is slowly becoming a reality, with IBM and Google already
producing quantum computing chips that will someday drive a
quantum machine. The creation of the first quantum computer
will certainly have an impact on AI by enabling us to solve more
complex problems.
Quantum computing promises

a future of unprecedented
computing speed
To streamline its research and training activities on this topic,
the ICTP inaugurated the Trieste Institute for the Theory of
Quantum Technologies (TQT) in March 2019. The institute has
been established in partnership with the University of Trieste
and the International School for Advanced Studies (SISSA) in
Italy, with support from IBM and Google.
The TQT provides a hub for the study of the future of AI on

quantum devices, offering in parallel a convenient link to
private actors (Google and IBM to begin with). The TQT’s
first conferences on quantum technology in 2019 were the
Advanced School and Workshop on Ubiquitous Quantum
Physics: the New Quantum Revolution from 18 to 22 February
and the School on Advances in Condensed Matter Physics:
New Trends and Materials in Quantum Technologies, from
7 to 15 May.
In November 2018, a researcher from the ICTP’s Condensed

Matter and Statistical Physics Section was among the
recipients of a €1 billion European Commission initiative
called the Quantum Flagship that is funding the development
of quantum technology over the next 10 years.
For the first three-year phase of the Flagship, ICTP’s Marcello

Dalmonte will be acting as co-principal investigator of a
research project that will focus on quantum simulation, one
of five of the Quantum Flagship’s research themes. Quantum
simulators study quantum systems, such as the atomic make-
up of materials or chemical reactions, performing tasks that
are too complex for even the most powerful, transistor-based
classical computers3.
3 https://en.unesco.org/news/international-centre-theoretical-physics-
contributes-second-quantum-revolution
The use of AI for
environmental
management and
disaster risk reduction

The use of AI for environmental management and disaster risk reduction
3. The use of AI for environmental

management and disaster risk reduction
3.1. The implications of AI for Another example is the use of drones equipped with AI
technologies to combat deforestation or poaching. In Kenya, for
ecosystem and environmental instance, the World Wildlife Fund has received a US$5 million
management grant from Google to use an AI device with drones to track
poachers in the Maasai Mara (WWWF, 2017). Although such
AI and related technologies such as the Internet of Things systems can successfully detect and deter poachers, it would
are expected to foster progress in most, if not all, areas be naïve to assume that the illegal trade in wildlife will not
of ecological and biodiversity research, together with adopt AI to its own advantage. For example, poachers could
environmental and ecosystem management, in general. use drones to identify and possibly kill valuable wild animal
species.
Motion-sensor cameras can …

gather vast quantities of data Drones equipped with AI
on biodiversity technologies [can] combat
deforestation or poaching
Motion-sensor cameras can be used in natural habitats to
gather vast quantities of data on biodiversity inexpensively AI can also provide tools for forecasting and analysing changes
and unobtrusively. These are already being used in the World to ecosystems. Sensors are being used in agriculture, for
Network of Biosphere Reserves. It used to be time-consuming instance, to help farmers monitor their crops and livestock.
to analyse such images but a recent article in the Proceedings Agriculture is a key economic activity in many biosphere
of the National Academy of Sciences has shown that AI can reserves. The tool Mcrops has been developed in Uganda to
automate animal identification for 99.3% of the 3.2 million flag diseases affecting cassava plants in real time (WWWF,
images in the Snapshot Serengeti dataset. Moreover, it can 2017). Another example is Sequoia, developed by Parrot, a
perform at the same level of accuracy (96.6%) as crowdsourced French company. Sequoia is a multispectral sensor attached
teams of human volunteers. The authors of the article affirm to a drone which combines a camera with a GPS sun sensor
that ‘automatically, accurately, and inexpensively collecting (ultraviolet and infrared rays) to analyse crop vitality by
such data could help catalyze the transformation of many capturing the amount of light plants absorb and reflect.4
fields of ecology, wildlife biology, zoology, conservation biology, AI can be integrated with other technologies to amplify the
and animal behavior into “big data” sciences’ (Norouzzadeh, spatio-temporal scales at which ecological patterns and
2018). processes can be studied. For example, remote sensing
This is a good example of the clear benefits and power of AI based on satellite imagery can provide an impression of
but it also illustrates the possible impact on the status and habitat heterogeneity over a large area. By incorporating this
employment opportunities of present ecological knowledge- information, a study can be designed to monitor biodiversity
holders. If field observations are performed by cameras and by deploying camera traps in a combination of locations
image recognition is undertaken by machines, field ecologists representative of the entire area. One recent study combined
could find reduced demand for their services and, ultimately, citizen science with deep learning algorithms (artificial
for their knowledge. This could have profound implications for intelligence), and produced accurate abundance counts of
the community of researchers, who will be required to learn
new skills to adapt to changes within their profession, as well
as for holders of indigenous and traditional knowledge.
4 See: https://www.imnovation-hub.com/society/types-of-smart-
sensors-applied-to-agriculture/#1
©UNESCO/Susan Schneegans
Elephants in the United Republic of Tanzania; drones and motion-sensors can help to monitor biodiversity but could also fall into the hands of poachers.
wildebeest migration within hours, compared to weeks for 2019 on Monitoring using New Technology, including drones.
traditional counting techniques (Torney et al., 2019). This meeting will also be an asset for the capacity-building
component of the Biosphere and Heritage of the Lake Chad
The Man and the Biosphere (MAB) programme has some
Basin (BIOPALT) project targeting managers of protected
experience of using drones in the Wudalianchi Biosphere
sites. The BIOPALT project is helping states to prepare their
Reserve in China. In 2015, a leading Chinese company for
application for the creation of a transboundary biosphere
the manufacture of civil drones, Dji Technology Co. Ltd,
reserve in the basin and their nomination for Lake Chad to be
demonstrated how drones could be utilized for land use
designated both a world heritage site and biosphere reserve.
planning and monitoring to biosphere reserve managers
attending the sixth international training workshop of the East In parallel, MAB is working with Kyoto University in the
Asian Biosphere Reserve Network in Harbin City, Heilongjiang Mont Nimba Biosphere Reserve in Guinea to improve the
Province. Using field data acquired by drones, the team conservation of chimpanzees in this biosphere reserve by
from Dji Technology Co. Ltd had developed 3D models of using drones to monitor land use and change in vegetation
Wudalianchi Biosphere Reserve. The workshop was organized cover. The drones are not yet equipped with AI but there is
jointly by DJI Technology Co. Ltd and the International Center obvious potential for doing so.
on Space Technology for Natural and Cultural Heritage, a
MAB and colleagues from UNESCO’s Global Geopark
UNESCO category 2 centre.
programme are planning to survey biosphere reserve
UNESCO and the International Research Centre on Space managers and geopark managers jointly by 2020 to ascertain
Technologies for Natural and Cultural Heritage (a category 2 the extent to which these sites are using AI technologies, or
centre) are organizing a capacity-building meeting in Africa in plan to do so.
3.2. The implications of AI The IGCP Council also identified nine priority topics5 for its
2019 call for project proposals. One of these topics is Big data,
for geosciences Cloud Computing and Artificial Intelligence in Geosciences,
which will involve close collaboration with research institutions
and the private sector. The projects selected will be sponsored
3.2.1. Reducing subjectivity in interpreting jointly by the Jeju Province Development Corporation of the
image logs Republic of Korea, UNESCO and the International Union of
Geological Sciences for five years.
The main advantage of using AI in geology is to interpret image logs
related to exploration data. Today, exploration geologists have to
deal with vast quantities of collected data. AI helps manage these Several UNESCO Global
data. The system uses heuristic techniques (experimentation, Geoparks already use
trial and error) and has the potential to help geologists search for augmented and virtual reality to
mineral resources and precious metals. For example, it can help
geologists interpret borehole images in glaciated terrain.
add to the visitor’s experience
The main advantage of using AI 3.2.3. Enhancing the visitor’s experience

in geology is to interpret image Several UNESCO Global Geoparks already use augmented and
logs related to exploration data virtual reality to add to the visitor’s experience by enabling them
to visualize geological and/or environmental change more easily.
Image logs hold important information about subsurface The integration of AI to personalize the visitor’s experience even
sequences and the spatial distribution and characteristics of further would not be a big step forward from this point.
faults/fractures and bedding (well-defined planes horizontal
Through AI, the education packages provided by geoparks
or parallel to the land surface in massive rock formations). will be tailored to the level of education and interest of the
Image logs can provide insights into the rock texture, textural individual. In museums, visitor centres and at sites of interest
organization and type and distribution of porosity. To reduce in the geopark, AI chatbots could increase the visitor’s
the subjectivity of the interpretation and cut the time it takes experience. Through AI-driven data mining and analysis of
to interpret images, AI can provide a new semi-automatic visitor behaviour, AI could also be used to develop customized
process for image log interpretation. programmes or routes matching the individual desires, habits
AI can automatically extract rock properties information and and preferences of each tourist.
has two advantages: it avoids subjectivity in interpretation and UNESCO geoscientists and MAB colleagues are planning to
reduces the interpretation time. However, some level of human survey managers of UNESCO Global Geoparks and managers
interaction and correction is still necessary in this process. of biosphere reserves in 2019, to ascertain the extent to
which they are using AI technologies or plan to do so in these
designated sites.
3.2.2. Big data, cloud computing and AI in the
service of sustainable development
When it met at UNESCO headquarters in Paris from 18 to
21 February 2019, the Council of UNESCO’s International
Geoscience Programme (IGCP) stressed the urgency of
integrating sustainability into the geosciences, to help countries
reach their Sustainable Development Goals. The Council
identified three priority areas for research in geosciences:
◼◼ Efficient, safe and sustainable Earth resources exploration

and extraction, iStock/Getty Images Plus
◼◼ Innovative renewable energy production and CO2 mitigation,

5 See here for a list of the other eight special topics:
◼◼ Better understanding and predicting climate change and http://www.unesco.org/new/en/natural-sciences/environment/earth-
geohazards. sciences/international-geoscience-programme/proposal-submission/
3.3. The implications of AI for for an explanation of these terms.) Blockchain approaches might
be useful for detecting the manipulation of data and algorithms
freshwater management and, thereby, protecting the integrity of data.
3.3.1. Machine-learning algorithms already in 3.3.2. The advantages of a three-pronged

use in water science approach to urban water management
Although AI is probably only being used in exceptional cases The Internet of Things, machine learning and Blockchain
in the operational water sector, machine learning algorithms technology can be combined to facilitate urban water
are increasingly being used in water science. In Serbia, for management. By combining these three technologies,
instance, the Centre for Water for Sustainable Development UNESCO’s aim is to generate and disseminate knowledge with
and Adaptation to Climate Change, which operates under the a view to improving service provision and quality, while ensuring
auspices of UNESCO (category 2 centre), has been using AI and the sustainability of the water resource itself.
statistical modelling for years to control the quality of time-series
data in structural and environmental monitoring.
The Internet of Things, machine
One of the most important techniques is deep learning, part of
learning and Blockchain
the so-called weak AI. Deep learning can complement modelling
forecasts as a predictive tool to detect patterns, classify and technology can be combined
correct remote sensing products, or for mitigation by anticipating to facilitate urban water
the future. management
Deep learning has great potential for supporting a wide range
of applications, such as water demand projections, the water– Smart water systems use an approach based on the Internet
food–energy nexus or climate change. For example, in Germany, of Things. These smart systems are comprised of a network of
the Federal Institute of Hydrology and Fraunhofer Institute for physical devices (like a flow meter). Devices include a sensor
Intelligent Analysis and Information Systems applied Echo State which records data (such as the level of water quantity and quality,
Networks (ESN) to produce discharge forecasts and water-level images etc.) and a communication device that transmits data in
simulations on the Rhine and Danube Rivers. The results of the real time to a cloud-based server. These smart water systems
ESN models are better than the existing traditional hydrological are gaining momentum in urban water resources management,
model, Hydrologiska Byråns Vattenbalansavdelning (HBV). as they increase the efficiency and effectiveness of the system
Nevertheless, the HBV models provide added value if the while providing cost-savings.
outflows calculated by them are used as input data for an ESN
The following example demonstrates how the three technologies
model. The simulations show that ESN models are also suitable
(Internet of Things, machine learning and Blockchain) work
for cases in which non-physical factors (such as the operation of
together to ensure smarter water management:
locks) play a major role.
◼◼ Firstly, IoT devices collect water data through sensors and
The application of deep learning might also accelerate the
transmit the data to a Blockchain-based server. For example,
standardization of data and metadata formats in the hydrological
while monitoring the environmental impact of fracking, water
field. This would lead to a much improved global monitoring
quantity and quality data can be transmitted through Internet
system of environmental systems and would directly support open
of Things devices embedded in the water distribution system.
data initiatives to make data Findable, Accessible, Interoperable
and Reusable (FAIR). Today, there are shortcomings in this area ◼◼ The data in the Blockchain server are transparent and
resulting from the high cost and complexity of managing water protected through a distributed peer to peer network. For
data, combined with non-existent or outdated policies for the example, the water quantity and quality data can be stored in
collection, storage, dissemination, sharing and use of water data. the Blockchain registry in a tamper-proof manner.
Deep learning also comes with caveats. It tends to deliver ‘black ◼◼ The machine learning algorithm then uses the data to create
box’ solutions that discriminate between the signal and background predictive analytics and aid decision-making capabilities.
without providing the operator with an understanding of the Smart contract protocols embedded in the Blockchain
relationship between hydrological input and output. This makes the technology can be used to execute a decision. For example, the
interpretation of the output difficult from a physical, mechanistic water quantity and quality data collected over a period of time
point of view and can lead to unwanted bias. An example of bias can be used by machine learning algorithms to predict future
are what have been termed ‘filter bubbles’ (or ‘echo chambers’). water quality and quantity and see when they reach dangerous
Deep learning is also highly sensitive to overfitting. (See the Glossary levels. If and when they do, smart contract protocols can be
configured to send a warning message to citizens, media earthquake. However, seismology is not yet capable of predicting
personal and relevant authorities to ensure swift action. an earthquake hours, days or weeks in advance. A panel set up
by the Government of Japan to discuss the contours of an AI
When well integrated with the water distribution network, a
strategy stated in its final report in 2017 that ‘no methods have
combination of the Internet of Things, machine learning and
been established, so far, to predict earthquakes using AI’. This
Blockchain technology can be used to:
is because the field of seismology has not yet determined which
◼◼ pay water bills to reduce transaction costs; type of data can best follow the processes signaling a build-up to
◼◼ set up a peer to peer trading system thus incentivizing an earthquake. Without the right kind of data, machine-learning
consumers to use less water; algorithms will be unable to develop an appropriate model for
predicting earthquakes.
◼◼ account for water transfers in water markets and, thus, help
to avoid bureaucratic turf battles and overlapping jurisdictions;
and 3.4.2. AI already being used to manage water-
◼◼ track assets in supply chain management, thus helping to related hazards
determine the carbon footprint better.
AI can be useful for producing hydrological runoff time series,
which are essential for reservoir management, environmental
protection and hydropower management.
3.4. The implications of AI for A number of new modelling systems are being tested for
disaster risk reduction their ability to forecast drought events with precision: Artificial
Neural Networks (ANN), Adaptive Neural-based Fuzzy Inference
Systems (ANFIS), Genetic Programming (GP) and Support Vector
Machines (Wang et al., 2009; Jalalkamali and Moradi, 2015). The
3.4.1. Extending AI beyond disaster response to current drawback in using AI for drought management is the
prevention lack of ‘big data’ necessary to produce models that can provide
accurate forecasting. The G-Wadi Geoserver tackles this problem
Many ideas and prototypes have already been tested in disaster
(see Box 1).
risk reduction. So far, they have tended to focus on the response
and rescue phase. For example, Sendai city in Japan has tested
a prototype with private companies for a tsunami alert using AI 3.4.3. AI can improve climate change
and Blockchain technology, whereby the AI system automatically
assessments
launched a drone, sending an alert through mobile phones and
radios and using facial recognition software to identify victims, Climate scientists have to juggle huge amounts of observational
such as people being swept away in a car by a tsunami wave. (Earth and satellite) and simulated data to monitor the current
climate, make future projections and detect high-risk zones.
UNESCO is a part of a consortium participating in a project
funded by the European Union. I-REACT is using a kit of new
technological tools to reduce risks linked to natural disasters,
including drones to improve mapping, wearables to improve
By improving the accuracy
geographical positioning and glasses with augmented reality of global climate models and
to facilitate reporting and information visualization by first climate forecasts, AI and
responders. The kit comes with a mobile app and a social media
machine learning algorithms
analysis tool to account for real-time crowdsourced information.
The automatic analysis of social media tweets by means of will help to mitigate and
machine learning algorithms and advanced semantic techniques manage climate change-related
makes it possible to pinpoint ongoing floods and provide real- risk
time data to enhance situational awareness; such information
can be useful to both citizens and first responders faced with a
Clouds are the single biggest source of uncertainty in global
natural hazard.
climate models. New studies (e.g. Rasp et al., 2018) suggest that
It is expected that future applications of AI will focus not only on AI and artificial neural networks can successfully resolve more
disaster response but also on the prevention phase. In the field complicated and smaller-scale atmospheric processes like the
of earthquake risk reduction, there are sensors available which ones involved in convective cloud formation and, thus, reduce the
can provide about 10 seconds’ advance warning of an impending uncertainties inherent to current climate models.
By improving the accuracy of global climate models and both more frequent and more severe (McGovern et al., 2017);
climate forecasts, AI and machine learning algorithms will AI and machine learning algorithms will also help to increase
help to mitigate and manage climate change-related risk, preparedness and response to environmental risk, fields where
including catastrophic weather events such as tornadoes, fast, smart solutions are of vital importance.
hurricanes and thunderstorms, which are projected to become
Box 1.
Using AI to manage drought and other hydrological risks
U N E S C O ’s G - W A D I G e o s e r v e r
application (Water and Development
Information for Arid Lands – a Global
Network) uses an artificial neural
network (ANN) algorithm to estimate
real-time precipitation worldwide.
This product is called the Precipitation
Estimation from Remotely Sensed
Information using Artificial Neural
Networks – Cloud Classification System
(G-WADI PERSIANN-CCS6).
The G-WADI PERSIANN-CCS GeoServer

has been under development since 2005,
through collaboration between the Center
for Hydrometeorology and Remote Sensing
at the University of California, Irvine, and
UNESCO’s International Hydrological
Programme. The core algorithm of
this system, supported by NASA and Daily Flood Bulletin prepared by the Namibian Hydrological Services using the G-WADI PERSIANN-
CCS Geoserver; slide shown by Dr Will Logan at an awareness-building workshop organized by
NOAA, extracts local and regional cloud UNESCO in March 2019, see Box 5.
features (coldness, geometric structure
uses it to prepare daily bulletins with up- precipitation monitoring.7 iRain allows
and texture) from the international
to-date information on flood and drought users to visualize real-time global
constellation of GEO satellites capturing conditions for local communities (see satellite precipitation observations,
infrared imagery and estimates rainfall image). It is being widely used globally track extreme precipitation events
at 0.04° × 0.04° spatial resolution (roughly to track storms, such as in the case worldwide and report local rainfall
4 km2) every 30 minutes. Information from of the Haiyan Super Typhoon, where information using a crowd-sourcing
LEO satellites is then used to adjust the the Geoserver captured the maximum functionality to supplement the
initial precipitation estimation from the precipitation intensity of approximately data. This provides an opportunity to
ANN algorithm. 361 mm per day on 7 November 2013, improve remotely sensed estimations
The G-WADI PERSIANN-CCS Geoserver is while the storm was approaching the of precipitation. Moreover, the use
being used to inform emergency planning Philippines. of a crowdsourcing functionality in
iRain to supplement the data opens
and management of hydrological risks, The system is now available through
up opportunities for engaging citizen
such as floods, droughts and extreme the iRain mobile application, devoted
scientists.
weather events. For example, the to facilitating citizen involvement
Namibian Drought Hydrological Services in collecting local data for global
7 See: http://en.unesco.org/news/irain-new-
mobile-app-promote-citizen-science-and-
6 It can be accessed at http://hydis.eng.uci.edu. support-water-management
Fostering quality
data for optimum
research

Fostering quality data for optimum research
4. Fostering quality data for

optimum research
4.1. Ensuring quality data and data Box 2.

UNESCO’s Water Information Network System
management
The International Hydrological Programme’s Water
Information Network System (WINS) was launched in January
The Natural Sciences Sector supports scientific research and
2017. It is an open-access, participatory platform for sharing,
experimental development (R&D) as a means of advancing
accessing and visualizing water‑related information, as well
human knowledge and sustainable development. Since R&D
as for connecting water stakeholders.
are largely data driven, these UNESCO programmes can
benefit greatly from the predictive and prescriptive capabilities Based on a tool equipped with a geographical information
of AI to find innovative solutions in fields ranging from health system, WINS allows people to store, access and create
to the environment and industry. tailored maps on water at all levels.
Despite the increasing gathering and availability of data in Its online library offers a wide range of additional information,
many parts of the world, the use of information (i.e. processed in the form of reports, policy briefs, graphs, charts, videos, etc.
data) to inform policies and decisions remains limited. Reasons
Contributors are empowered to share information by the
include a shortage of financial and human resources, a lack of
system’s transparency and respect for authorship: indeed, all
commitment and investment from political leadership, gaps in
information shared on the platform benefits from standardized
technical skills and an absence of clearly defined strategies
metadata and from a Digital Object Identifier (DOI), which
and mechanisms to support the sharing and use of data and
accurately identifies and credits the contribution, thereby
information. Since the quality of the data fed into the AI system
facilitating sharing down the line.
will be critical to determining the success of research, it is
imperative that UNESCO pursue its support of member states Cooperation is also fostered between service providers:
in ensuring quality data and data management. standardized protocols allow WINS to display information from
external platforms and vice versa. This helps to create a one-
In freshwater management, for instance, standards need to
stop-shop online catalogue of information, as recommended
be adopted for water data and metadata, to foster a decision-
by the UN Administrative Committee on Coordination in its
making process informed by science and ensure sound
report ACC/2000/18.
management and governance of water. Countries also need
to embrace an open data approach to water data access In this sense, the platform is helping to bridge the gap
and licensing and a quality system needs putting in place to between North and South in terms of access to, and sharing of,
manage data products derived from AI. knowledge, while fostering proactive cooperation in this area.
So far, 40 UNESCO member states have joined the platform, as
The data generated and shared through UNESCO online
well as five centres operating under the auspices of UNESCO
platforms must be of the highest quality to safeguard the
(category 2 centres), and seven water-related Chairs. Overall,
credibility of UNESCO as an intellectual organization. These
340 contributors are sharing information.
platforms currently include the Global Observatory of Science,
Technology and Innovation Policy Instruments (GO-SPIN),
For details, see: https://en.unesco.org/ihp-wins
managed by the Section for Science Policy and Partnerships,
the Water Information Network System, managed by the
International Hydrological Programme (see Box 2), and the
indicators and statistics in science and education platform,
managed by the UNESCO Institute for Statistics.
Fostering quality data for optimum research
Applications based on the Internet of Things exist in many

More data are needed to map fields. Drones and sensors can be used to monitor water
quantity and quality, soil moisture and air quality, to track
the global research effort animals and map disease. Given the wide range of potential
applications that can benefit communities (Biggs et al., 2015).
More data are needed to map the global research effort. Could UNESCO has been advocating the use of the Internet of Things
AI help to develop better indicators in science, technology and for development. Between 2010 and 2018, ICTP organized
innovation (STI), to ensure that policy-makers have better 26 training activities on the Internet of Things in 20 developing
information at their disposal? UNESCO’s project in STEM and countries stretching from Nepal to Benin.
Gender Advancement (SAGA) has identified the need for better
indicators to improve monitoring of gender trends in research
and innovation, for instance. Could AI eventually do away with UNESCO has been advocating
the need for data surveys?
the use of the Internet of
Things for development
Could AI help to develop better
indicators and ensure that The Internet of Things can help scientists from developing
countries to bridge the so-called scientific divide. If we define
policy-makers have better the digital divide as being the gap between those with access to
information at their disposal? digital technologies and those without, the scientific divide can
be defined as the gap between those with access to scientific
As early as 2010, UNESCO’s Abdus Salam International Centre data and those without. The collection of empirical data has
for Theoretical Physics (ICTP) identified AI and data science as enabled advances in science and contributed to improvements
critical areas for its new strategic plan. ICTP is strategically in the quality of life.
placed not only because of its connections with the scientific
community worldwide but also because the theoretical basis
of AI is, in many cases, familiar to theoretical physicists. In The Internet of Things is
addition, theoretical physicists are playing a leading role in a low-cost, low-power
the development of AI for quantum computers.
technology [that] can be
Last but not least, AI is already playing an important role in deployed in remote areas
research, especially in data-intensive fields such as medicine,
genetics, cell and systems biology and neuroscience, as we
Until recently, data collection has been based mainly on
saw earlier.
a limited range of expensive equipment that uses wired
ICTP is fostering excellence in research by its in-house team infrastructure, particularly when it comes to environmental
of scientists and is currently recruiting a staff scientist in AI applications. Data collection has been a costly, difficult task
and data science. restricted to a relatively small number of fixed, sparsely
distributed locations and maintained by organizations with
large budgets. As a result, the data gathered are often
incomplete, especially in developing countries and remote
4.2. The Internet of Things can help areas.
bridge the scientific divide The Internet of Things could radically change that; this low-
cost, low-power technology does not require any pre-existing
AI helps to organize and make sense of huge amounts of data. infrastructure and can be deployed in most remote areas. The
In the near future, the Internet of Things will be one of the vast range of sensors that can be connected to the nodes
sources of real-time data. There may even come a day when support many different scientific applications, including those
the Internet of things will not exist without AI components. outlined above.
These components can be incorporated in connected objects
such as drones, sensors and robots. Imagine an AI algorithm
capable of collecting data from sensors in a factory and
performing predictive maintenance; such a system would save
time and money by preventing unplanned shutdowns.
Issues at the
science–policy nexus

Issues at the science–policy nexus
5. Issues at the science–policy nexus
5.1. Smart solutions are ‘Smart solutions’ will make urban life more pleasant, such
as by reducing traffic jams and pollution, but they will come
already here at a price. People will also have to accept the omnipresence
of surveillance cameras and the inherent invasion of privacy.
The revolution in artificial intelligence is already under way. Motion sensors could control the movements of employees,
How prepared is society for this revolution and in what ways such as by monitoring the amount of time they are absent
can public policies ensure that this disruptive technology has from their workstation or the time they spend on a specific
a beneficial effect on human societies and the environment task. There is, thus, a risk that machines will become our
and does not widen the technological divide? masters. Today’s smart phones already automatically count
the number of steps a person takes in a day. Were a person
to remain beneath what is considered a healthy threshold,
Economists predict that AI machines could hypothetically apply pressure for the person
products will account for up to to walk more that day, such as by emitting shrill sounds or
by repeating instructions at regular intervals. Today’s cars
15% of the total production of already emit a shrill sound if an occupant forgets to attach
goods within a decade their safety belt.
Yoshua Bengio
In July 2018, Yoshua Bengio, an expert in deep learning, told

the UNESCO Courier that ‘economists predict that AI products 5.2. How neutral is technology?
will account for up to 15% of the total production of goods
within a decade’ (UNESCO, 2018a). Over time, technology may come to be used in ways that its
creator never imagined. The Global Positioning System (GPS),
Applications of AI are already offering consumers ‘smart for instance, was originally designed by the US Department
solutions’. These solutions may target the individual (smart of Defense for military use and only later adapted to civil use,
heart stimulators like pacemakers, robotic house-cleaners, in recognition of the technology’s potential for commercial
smart washing machines and refrigerators, drone deliveries, spin-offs.
see photo), the group (smart ventilation, lighting and heating
AI systems already include dual-use technologies, such as
systems for buildings) or urban populations (traffic light
sensors, drones and robots. The myriad possibilities offered by
control). The systems tend to be situation-specific, so it may
AI systems begs the question: which applications are desirable
not be straightforward to transfer the technology elsewhere.
and which are not? For instance, robots can be used to improve
The Fourth Industrial Revolution (Industry 4.0) currently human safety, such as by defusing bombs and identifying the
under way will see much practical engineering work shift to location of anti-personnel mines. They can also be turned into
the Internet and the Internet of things. In the near future, killers. In July this year, the Belgian parliament voted to ban
home refrigerators will be able to communicate with individual killer robots. The resolution calls on the Belgian government
stores to place orders for food then inspect the delivered goods to forbid the Belgian military from using lethal autonomous
before accepting them. Over the next ten years, we shall use weapons and to work towards an international ban (Pax, 2018).
many AI applications without even realizing that the task was Is technology neutral? Opinion is divided on the question. For
accomplished by a computer rather than by a human being. American historian Melvin Kranzberg, ‘technology is neither
The human being will still be needed to define the parameters good nor bad; nor is it neutral’8.
and interpret the results, though, since AI systems are, after
all, algorithms. 8 See: https://en.wikipedia.org/wiki/Melvin_Kranzberg
Getty Images / Baranozdemir Delivery by drone
If technology is not neutral, is it a reflection of its creator’s 5.3. AI may widen the
mind? And if so, can the creator of a technology be held
responsible for applications of the technology that he or she technological divide
never imagined? For example, machines trained through
deep learning can interpret 360° camera views, making it There is concern that the potential commercial rewards of AI
possible for them to identify a face in the crowd by connecting could lead to a monopolization of research in this burgeoning
the image to a database – although they have been known field. Yoshua Bengio, a Canadian expert in deep learning, has
to make mistakes. This face recognition software could be raised this concern. For him, by broadening their customer
used to identify missing children on city streets but also to base, AI companies ‘will increase the amount of data they have
ensure surveillance of the general population. To take another access to – and that data is a goldmine that makes the system
example, imagine a machine connected to a large database even more powerful. Such a concentration of power can have
of genetic syndromes which has learned to analyse cases of a negative impact on both democracy and the economy. He
human malformation when presented with images of patients. warns that ‘it favours large companies and slows down the
This knowledge could help doctors reach a rapid diagnosis but ability of small new companies to enter the market, even if
could also be used by insurance companies to discriminate they have better products to offer’ (UNESCO, 2018a).
against certain customers.
Both national and international public policies are needed There is concern that
to establish the parameters of AI and avoid abuses of this
technology which could have far-reaching consequences for
the potential commercial
human society. rewards of AI could lead to a
monopolization of research
Cockburn et al. (2017) have raised similar concerns. They argue Figure 2.
that the potential commercial reward from AI-related research Patents in artificial intelligence held by top R&D companies, by
is likely to give individual companies powerful incentives to headquarters’ location, 2012–2014
acquire and control critical large datasets and application- Top 5 economies
specific algorithms. They suggest that the adoption of policies
Japan 32,8%
encouraging transparency and sharing of core datasets across
both public and private actors could stimulate innovation- Rep.Korea 20%
oriented competition and enhance research productivity. USA 17,9%
Taiwan
Province of China 9,4%
Policies encouraging China

%
8,2%
transparency and sharing of 0 10 20 30 40
core datasets across both Other economies
public and private actors could Germany 3,2%
stimulate innovation-oriented France 2,2%
competition and enhance Netherlands 1,6%
research productivity Finland 1,2%
Canada 0,9%
UK 0,5%
5.3.1. The genie is already out of the bottle
Sweden 0,5%
UNCTAD observes that ‘digitalization will affect all countries,
Switzerland 0,4%
irrespective of whether they actively pursue it. Developing
countries, and especially least developed countries, may risk India 0,2%
increasing dependency on a few global digital multinational
Ireland 0,2%
corporations, or further marginalization from the global
economy’ (UNCTAD, 2017). Denmark 0,1%
AI may already be widening the technological divide, even Belgium 0,1%

among industrial countries. AI patents are currently largely %
concentrated in multinational corporations headquartered 0 1 2 3 4
in just seven economies, according to the Organisation for
Source: OECD (2017), Figure 1.25
Economic Co-operation and Development (OECD, 2017), which
studied the patenting behaviour of the top 2,000 companies
for R&D between 2012 and 2014. Corporations headquartered ‘There are emerging concerns in Europe about the erosion
in just seven economies accounted for 93% of AI patents of its science base through takeover bids from competitors’
registered with the top five patent offices9: Japan (33%), the (UNESCO, 2015). In August 2018, the German government
Republic of Korea (20%), USA (18%), the Taiwan Province of vetoed the purchase of the domestic firm Leifeld by the
China (9%), China (8%), Germany (3%) and France (2%). See Chinese Yantai Taihai Group, citing security reasons. The
Figure 2. German company manufactures specialized machines and
equipment for the automobile, gas, nuclear and aerospace
The current concentration in AI is part of a wider trend. Over
industries, among others (Deutsche Welle, 2018).
the past decade, there has been a dizzying number of mergers
and acquisitions which have created powerful multinational Not all multinational corporations are investing heavily in AI
corporations with a bigger GDP than some national economies. but the OECD (2017) found patenting among the top 2,000
The figures speak for themselves. ‘In the USA, 12,249 companies for R&D to be more dynamic in AI than any other
companies concluded mergers and acquisitions in the twelve technology. According to this study, the number of AI inventions
months to 30 June 2015’. Of these deals, 315 represented an is increasing by 6% per year on average, twice the average
investment of more than US$ 1 billion (UNESCO, 2015). annual growth rate observed for all patents.
9 Patents registered with the top five patent families: European Patent
Office, Japanese Patent Office, Korean Intellectual Property Office,
State Intellectual Property Office of the People’s Republic of China and
the US Patent and Trademark Office.
◼◼ African governments should invest in basic and applied

Patenting among the top research in AI and related technologies, ideally in
partnership with the African private sector, to ensure an
2,000 companies for R&D is endogenous research capacity in AI and endow African
more dynamic in AI than any scientists with the capacity to identify the undesirable
other technology effects of some AI algorithms, such as gender and racial
biases;
◼◼ African stakeholders should invest in the Internet of

What are the implications for start-ups in AI of the Things, given that the quality of the data fed into the
dominance of multinational corporations around the AI system will be critical to determining the success of
world? At a forum on AI in Africa organized by UNESCO research conducted in Africa, that this low-cost, low-power
in Benguerir, Morocco, in December 2018, one participant technology does not require any pre-existing infrastructure
suggested that multinationals stifled the growth of disruptive and can be deployed in most remote areas and that the
start-ups through acquisitions. Speaker Dr Tapiwa Chiwewe wide range of potential applications of this technology
from IBM Research Africa responded that his corporation can benefit rural communities, in particular; moreover,
was actually helping start-ups in South Africa thrive by AI components can be incorporated in connected objects
providing them with access to information and equipment such as drones, sensors and robots;
via the cloud that would not have been available to them in ◼◼ Governments should engage with multinational
the past (see Box 3). corporations to find common ground on how these
Dr Chiwewe was one of three panelists at the breakout corporations can best contribute to achieving the national
session in Benguerir on the theme of Building Capacity in agenda in a rules-based environment that includes the
AI: from basic research to applications. The other panelists adoption of policies and laws that safeguard the national
were Dr Alessandro Scardicchio, a researcher at the ICTP, interest while allowing the multinationals to meet their
and Dr Fernando Vega Redondo from the University of own objectives;
Bocconi in Italy. ◼◼ Multinational corporations should embrace the open
source movement, such as by donating some of their
The following recommendations emerged from the session: intellectual property to open source initiatives or creating
◼◼ The teaching of ethics should be embedded in AI training open access platforms like IBM’s Digital Nation Africa to
courses, including at the ICTP. provide people with equipment (via the cloud) and the
means to develop their skills.
◼◼ Computer science departments in Africa need to engage
UNESCO is participating in a ‘bottom-up’ project led by the
in curricular reform to ensure that they are able to do
Knowledge for All platform which is mapping players in
research on AI and develop related skills.
AI in the Global South. This multipartner project involves
◼◼ The ICTP should develop cooperation with African UNESCO’s Communication and Information Sector and the
governments which have manifested interest in funding newly established UNESCO Chair in Artificial Intelligence,
research chairs or centres in AI at national universities John Shawe-Taylor, Professor of Computational Statistics
to foster quality education and both basic and applied and Machine Learning at University College London. As of
research in AI. April 2019, the project had identified 286 start-ups in Asia
(excluding China), 29 in sub-Saharan Africa and 22 in the
◼◼ Stakeholders should develop open access online learning Middle East and North Africa. The project is also mapping
platforms to facilitate skills development in AI, one AI players in the Global South in academia, accelerators and
possible model being IBM’s Digital Nation Africa platform. investors, corporate entities and the social sector10.
10 See: https://www.k4all.org/project/aiecosystem/
Box 3.
The social compact between IBM and the Government of South Africa
The Government of South Africa has He explained that IBM was focusing on research on AI. ‘There is a need for
passed legislation to ensure that healthcare, education, agriculture and curricular reform in these departments
companies behave as responsible financial services in South Africa, as these to develop skills’.
corporate citizens. Codes of Practice were major challenges for the country. IBM also has an enterprise development
require all entities operating in the ‘Before IBM takes on a new project’, programme that mentors young inventors.
South African economy to contribute he said, ‘it first checks that there is a One participant asked whether the
to the objectives of Broad-based Black pressing problem needing to be solved. It growing concentration of AI in the hands
Economic Empowerment (B-BBEE). Since then brings an external partner on board of a few multinationals was not having a
multinationals may have global practices to understand the context of the problem detrimental effect on South African start-
preventing them from complying with the better and ensure that the solution ups, which could easily be swallowed up
ownership element of B-BBEE through developed is fit for purpose. IBM may by these tech giants.
the traditional sale of shares to black obtain private data (or publicly available
In reply, Dr Chiwewe cited examples of
South Africans, the Codes of Practice data) from that partner and use that to
South African start-ups in AI that had
have made provision for the recognition solve a given problem... To ensure that AI
become viable businesses. JUMO, for
of Equity Equivalent contributions, as an solutions are fair, IBM analyses the source
instance, has launched an AI-powered
alternative contribution to the economy. of data, the output of the AI models and
platform to assess lending risk and
Equity Equivalent programmes are the underlying algorithms to check for
tailor financial products to those living
expected to contribute towards the bias’.
in developing countries where credit
achievement of: Dr Chiwewe gave examples of the type information is scarce; it has received an
## enterprise creation and development; of research work being done by IBM. investment of US$52 million from several
## foreign direct investment ; ‘For instance, there is a four-year lag in investors, including Goldman Sachs.
reporting cancer statistics in South Africa’,
## accelerated empowerment of black Another start-up, NMRQL, has launched
he said. ‘AI can correct this by automating
rural women and youth; an AI-powered unit trust fund.
the process of studying pathology reports,
## sustainable growth and development. ‘Today’s start-ups have the advantage of
meaning that this analysis can now be
## human development with focus on being able to access equipment via the
done in near-real time. To take another
education and skills development; cloud from companies such as IBM’, he
example, in the financial sector, access to
said, ‘where the start-ups could even
## infrastructure investment with credit is a problem. An AI application can
open free accounts. There is a freer flow
emphasis on developing the create a credit score that will reduce the
of information and knowledge nowadays
country’s research and development default rate on repaying loans’.
which gives start-ups an advantage’.
infrastructure. The IBM research lab in South Africa
Dr Chiwewe assured participants that ‘IBM
When IBM decided to set up a research is located in the Tshimologong digital
believes in the open source movement and
lab in South Africa in 2016, it first had innovation precinct, an innovation hub
donates some of its patents to open source
to convince the government that the close to Wits University. In parallel to the
initiatives. To empower African youth
company’s research focus would reflect research lab, IBM has set up an academic
with digital skills and tools to improve
national priorities. One outcome of the programme offering internships and
their daily lives and provide access to a
negotiations was an agreement under scholarships to South African students.
wide range of opportunities, anyone can
which IBM set up an Equity Equivalence IBM has also awarded faculty awards to
log onto IBM’s Digital Nation Africa to
Investment Programme. professors doing work in key technology
learn about technologies such as AI and
Dr Tapiwa Chiwewe from IBM Research areas.
obtain a certificate. For more advanced
in South Africa described how the Dr Chiwewe and his colleagues spend
technologies like quantum computing,
multinational corporation was fulfilling time giving talks at universities to share
people can access a 16-qubit quantum
its social obligations, at a session on their knowledge of AI. They also organize
computer via the cloud through IBM’s Q
capacity-building in basic and applied student competitions on related themes.
Experience’.
research organized on 13 December From his visits to computer science
2018 as part of UNESCO’s Forum on AI departments across South Africa, he For details of the Equity Equivalence programme, see:
for Africa. has noticed that few of them are doing https://www-05.ibm.com/za/aic/programme.html
5.3.2. Open source and open science sector. HOPE contributes to this by providing youth and young
movements can make a difference professionals in the water sector with training in how to use free
open source software (see also p. 40).
The open source and open science movements are striving to
level the playing field in AI. Thanks to the lower cost, free open
Figure 3.
source software can improve access to technologies in the Global expenditure on computer software and services in 2011
developing world, in particular, and provide a sustainable basis
for scientific decision-making. Computer software and services spending, $ billions
According to the World Information Technology and Services

(2005) 253 632
Alliance and Information Handling Services Global Insight Inc.,
global spending on computer software and services (excluding
(2006) 275 674
software embedded in devices) amounted to an estimated
US$1.2 trillion in 2011 (see Figure 3), or almost one-third of global (2007) 296 712
spending on information and communication technologies (ICTs)
the same year, which is equal to about 2% of GDP (UNCTAD, (2008) 312 745
2012).
An estimated 10 million people were employed in the global (2009) 305 715
computer software and services sector in 2011. National shares
of this sector ranged from 0.1% to 2.2% of total employment. The
(2010) 325 757
developing countries with the highest proportion of employment
(2011) 337 844
in this sector, according to available data, were Costa Rica (0.8%),
South Africa (0.7%) and India (0.6%) [UNCTAD, 2012].
● Computer software ● Computer services
There is considerable regional variation in the intensity of policy
initiatives related to free open source software, according
Distribution of ICT spending, in %
to global surveys conducted by the Center for Strategic and
International Studies (CSIS, 2010). Europe is the most active 9%
region, accounting for close to half (46%) of all initiatives, with Computer software
a high approval rate (see Table overleaf). Among developing
regions, Asia is the frontrunner with more than 80 initiatives, 12%
followed by Latin America (57) and Africa (9). Computer Hardware
However, many buyers still attach importance to the reputation

of the brand names of proprietary software vendors and it is not 59%
Communication
uncommon for software users in both the public and private
sectors to associate price with importance, thereby discouraging 20%
Computer services
lower-cost solutions. At the same time, there is a dire need in
low-income and middle-income economies for affordable and
accessible specialized software in engineering.
Source: graphic by UNESCO, using data from UNCTAD, 2011
Hydrologists are constantly creating innovative mathematical
models to manage water in an effective way, to solve local
problems. For these models to be effective, however, they should
There are a number of independent initiatives in ecological
be easy to use and capable of handling massive data sets. This
monitoring that permit open-source information-sharing,
last step requires sophisticated software infrastructure, which is
significantly reducing costs. The following are some examples.
expensive and may take years to build.
Open databases supporting in situ activity sensors include
In 2013, UNESCO launched the Hydro Free and (or) Open-source the Tropical Ecology Assessment and Monitoring (TEAM)
Platform of Experts (HOPE11) initiative. HOPE brings experts Network Education portal, the TRY plant trait database and
together for capacity-building and training using free open the Bio acoustica online repository and analysis platform. DNA
source software. HOPE has also been designed to stimulate sequences are shared as reference sources for DNA barcoding.
cooperation in R&D and enhance the dissemination of free open GEO BON, meanwhile, supports open datasets for Exceptional
source software. Since education is becoming increasingly linked Biodiversity Variables through its own portal (Mulatu et al., 2017).
to technology, it is essential to promote and foster equal access A vast array of types of open source software are available for
to ICTs, in order to improve the quality of education in the water remote sensing and data analyses.
11 https://en.unesco.org/hope
There are also initiatives in the private sector. Open AI is a 5.4. AI: the new space race?
non-profit firm whose 60 full-time researchers and engineers
focus on problems that require the team to make fundamental
It has been suggested that AI may be the new space race.
advances in AI capabilities. They publish their research at
The Centre for a New American Security, a think tank based
machine learning conferences and use open-source software
in Washington DC, has stated that ‘falling behind in AI
tools. Their website states that ‘we will not keep information
development and implementation would present a risk for US
private for private benefit but, in the long term, we expect to
global economic and military leadership. The United States
create formal processes for keeping technologies private when
may very well be in a new space race’ (Horowitz et al., 2018).
there are safety concerns’12.
The President of the Russian Federation stated in 2017 that
‘artificial intelligence is the future, not only for Russia, but
Table. for all humankind. It comes with colossal opportunities but
Open source policy initiatives by region, 2000–2009 also threats that are difficult to predict. Whoever becomes
the leader in this sphere will become the ruler of the world’
(Russia Today, 2017).
Approved Proposed Failed Total
China’s Next Generation Artificial Intelligence Development Plan
Europe 126 27 10 163
(2017) fixes the targets of attaining the same level in AI as the
Asia 59 20 2 81 USA by 2020 and of becoming the world leader in AI by 2030
Latin America 31 15 11 57 (Asgard and Roland Berger, 2018).
and the Chinese start-ups in AI already numbered 383 by 2017. This
Caribbean places China well behind the world leader, the USA (1 393), but
ahead of Israel (362), the UK (245), Canada (131), Japan (113),
North America 16 11 10 37
France (109), Germany (106) and India (82). The Republic of
Africa 8 1 – 9 Korea ranks 12th and the Russian Federation 20th (Asgard and
Middle East 5 2 – 7 Roland Berger, 2018).
Source: CSIS, 2010
5.5. What will greater automation

mean for low-skilled jobs?
Open-source platforms enable
anyone with a computer The same economies dominate both AI patents and industrial
and an Internet connection robots. Although most industrial robots are not yet equipped
with AI, robots and other cyberphysical systems are now
to modify… the code being designed to monitor production and make independent
anonymously and legitimately decisions on the factory floor.
Five economies accounted for 74% of global sales in industrial

robots in 2016: China (excluding the Taiwan Province of China,
These safety concerns are very real. AI software and hardware
which is the sixth-biggest market), the Republic of Korea,
are vulnerable to the usual security breaches in cyberspace.
Japan, USA and Germany. The most automated countries in
However, open-source AI development projects also enable
the world are the Republic of Korea, Singapore, Germany and
anyone with a computer and an Internet connection to
Japan. However, China has become the world’s biggest market
modify open-source machine learning libraries or the code
for industrial robots: between 2013 and 2016, Chinese robot
anonymously and legitimately. This makes it hard to attribute
suppliers expanded their global market share from 25% to
ownership to a given AI system, and thus accountability,
31% (IFR, 2017).
should something go wrong. In light of the growing number
of open-source AI development projects, Jayshree Pandya, In China’s Guangdong Province, the Midea manufacturing
Founder of Risk Group, deems it urgent to define a security company has reduced its workforce by half since it introduced
risk governance framework for AI. ‘How will we regulate AI robots onto the factory floor in 2011. These robots were
when humans will likely lose control of its development?’ she bought from Kuka, Germany’s biggest and most advanced
wonders. She observes that ‘there may also come a day when manufacturer of robots. In 2016, Midea purchased Kuka
AI systems are capable of changing their own code or self- (Tsang, 2016). As of 2013, Midea had about 100,000 employees
replicating’ (Pandya, 2019). in China and abroad.
12 See: https://openai.com/about/
Between 2011 and 2016, the supply of industrial robots worldwide

increased by about 31% per year. By 2016, there were about
1.8 million operational robots and their number is expected
to top 3 million by 2020. Most robots are employed in the
automotive industry (35%) but, thanks to strong growth in
Asian demand, the electronics and electrical industry took
31% of the market in 2016, including in China, Japan and the
Republic of Korea (IFR, 2017).
Automation driven by AI will

see many industries replaced
by new ones.
Automation driven by AI will see many industries replaced by
new ones. ‘Up to 47% of US jobs in 2010 were rated as highly Ju1978 / Shutterstock.com
likely to become computerized in the next 10–20 years’ (WEF, Figure 4). These jobs would be threatened by self-driving
2016). According to a study by the University of Minnesota that trucks, as would the secondary economy built around them,
drew on US census data, truck-driving was the most common such as the roadside cafés and motels that rely on the custom
job in 29 out of 50 US states in 2014 (Quoctrung, 2015; see of truck-drivers.
Figure 4.
Most common jobs in the USA by state, 2014
Software Developer
Truck Driver Farmer Truck Driver

Truck Driver
Truck Driver Truck Driver
Truck Driver
Secretary
Truck Driver Primary School Teacher
Truck Driver Farmer
Nursing Aide Primary School Teacher
Truck Driver Truck Driver Primary School Teacher
Truck Driver Truck Driver Secretary

Truck Driver
Retail Clerk Truck Truck Driver Secretary
Secretary
Software Truck Driver Driver Lawyer
Developer Truck
Software Developer Driver Software
Truck Driver Developer
Truck Driver Truck Driver Truck Driver
Truck Driver
Truck Driver
Customer Service Truck Driver
Truck Driver
Secretary
Truck
Driver Truck Driver Truck Driver
Truck Driver
Truck Driver
Primary Primary
School Teacher School Teacher
Cook
Source: IPUMS-CPS/University of Minnesota, credit Quoctrung Bui/National Public Radio, USA, Reproduced with permission
Jobs are poised to disappear from places such as the factory According to a study of employment trends tied to automation
floor, fast food and retail outlets, call centres and trucking in England between 2011 and 2017 by the UK Office for National
companies but certain skilled professions will also be affected Statistics (ONS, 2019), ‘as the proportion of employees in
by greater automation, such as accounting or publishing. The high‑skill occupations increases, the probability of automation
media and entertainment industry has become the top digital decreases. For those industries where half of the workforce
industry, followed by retail and high technology (UNCTAD, 2017). or more are highly skilled, their probability of automation is
Product digitization has had a sweeping impact on jobs in at least 35%’ (see Figure 5). Women in England account for
cultural industries: trade in recorded music declined by 27% and 70.2% of employees in jobs with a high risk of automation but
that in movies by 88% between 2004 and 2013, even as audio- only 42.6% of employees in jobs with a low risk of automation.
visual services as a whole have gained ground (UNESCO, 2016). Over the six years to 2017, retail outlets in England installed
additional automatic check-outs, resulting in the loss of one speech capabilities may do away with the need for call centres,
in four (25.3%) supermarket cashier jobs. Most of the cashiers for instance, which are often situated in developing countries.
were women. At least 15% of laundry workers, farm hands The drop in the availability of lower-skilled jobs could also
and tyre-fitters in England also lost their jobs to automation exacerbate gender inequality, as more men compete with
over this period, the study found. It estimated that about women for the same jobs (WWWF, 2017).
1.5 million of the country’s workers were at a high risk of losing
their jobs to automation (ONS, 2019).
Is the digital economy creating
Figure 5. fewer jobs than traditional
Probability of automation in England in 2017, by the proportion of industries?
employees in high-skilled jobs for different industries (%)
There is also a risk that value produced in low- and middle-

Proportion of employees in high-skill occupations
income countries could revert to high-income countries,
80
creating a form of cybercolonialization, such as in the case
Computer programming and consultancy
of Uber drivers who would no longer share company profits if
the multinational corporation replaced them with self-driving
70
cars (WWWF, 2017).
Scientific research and development
More broadly, is the digital economy creating fewer jobs than
60 traditional industries? Frey and Rahbari (2016) argue that,
‘while today’s technology sectors produce fewer jobs than the
Architecture and engineering
ones that preceded them, their indirect impacts on job creation
Education
Activities of head offices are far greater…. The fate of workers thus depends less on
50
Manufacture of basic pharmaceutical products job opportunities created by biotech companies or computer
firms [than] on the demand for local services created by those
companies’.
40 Extraction of
crude petroleum Legal and accounting
and natural gas Notwithstanding this, in 2010, only 0.5% of the US labour
Publishing
force was employed in industries that didn’t exist in 2000,
Information services
such as online social media networks or video and audio
30
Programming streaming. Compared to the major corporations behind the
and broadcasting Vehicle manufacturing early development of computer technology, such as IBM or
Wholesale trade
Dell, companies leading the digital revolution have relatively
20 Rubber and plastic manufacturing
Motion picture, video
Real few employees. In 2013, Facebook counted only 7,185
estate
and television programme Textile manufacturing employees, compared to 431,212 and 108,800 respectively
production Retail trade
for IBM and Dell (Berger and Frey, 2015). Berger and Frey
10 (2015) conclude that ‘because digital businesses require only
Sports and recreation Postal and courrier
Food and limited capital investment, employment opportunities created
beverage
Farming services by technological change may continue to stagnate as the US
0 economy is becoming increasingly digitized’.
25 30 35 40 45 50 55 60 65
Probability of automation The digital economy is certainly undermining job security.
According to Muntaner (2018), ‘most digital economy workers
Source: ONS (2019) Annual Survey Population, UK Survey of Adult Skills, are exposed to the health-damaging precarious employment
Figure 8 (open access) conditions characteristic of the contemporary working class
in high-income countries’. Online digital platforms in these
countries are seizing on the opportunities offered by the gig
What impact will demographic growth have on the labour
economy to limit the cost of labour. They treat their delivery
market in countries where agriculture, manufacturing and
staff like freelancers, rather than employees. In France, a
services are increasingly automated? Could widespread
cyclist paid by a digital platform like Deliveroo or Take Eat
automation be a boon for countries like Japan that are
Easy to deliver meals or a chauffeur hired by Uber to collect
confronted with depopulation?
the platform’s customers must provide their own means of
The drive for automation could have the biggest impact on transport. They have no employment contract with the platform
employment in low- and middle-income countries, since most and, thus, no job security, no right to an unemployment benefit,
of the lower-skilled jobs requiring automation in high-income no statutory holidays, no sick leave. Once their costs have been
countries have already gone through this transformation. deducted, they earn less than the hourly minimum wage in
Progress in developing AI systems with voice recognition and France, according to Amellal (2018), for whom these digital
platforms are eroding more than a century of social gains in labour markets. The Japanese government’s selection
France. criteria for most large university grants now take into account
the proportion of foreigners among teaching staff and
The courts are beginning to regulate these platforms. In
researchers13 (UNESCO, 2015, p.35).
November 2018, the French supreme court, the Cour de
Cassation, awarded a cyclist working for Take Eat Easy the status Employers increasingly consider mobility among skilled
of company employee. In January 2019, the Court of Appeal of employees to be an asset. ‘Studies conducted across Europe
Paris rendered a judgment affirming that a chauffeur working have shown that a high level of mobility by qualified personnel
for Uber was entitled to an employment contract (Collas, 2019). across countries and between the public and private sectors
contributes to the overall professionalism of the labour force,
as well as to the innovative performance of the economy’
5.6. Growing competition for skills (UNESCO, 2015, p.81).
The European Union now requires public-funded bodies to

According to a 2018 survey of nearly 1,400 CEOs in advertise vacancies internationally and has introduced a
91 countries, many companies face ‘a shortage of skilled ‘scientific visa’ to slash the red tape for non-EU nationals
talent to clean, integrate and extract value from big data and submitting job applications. These policies have been inspired
move beyond baby steps toward AI’… It’s not only a matter of by the need to augment the number of researchers in the
hiring or developing AI specialists and data scientists. It is EU, a significant share of whom will have to come from third
equally important to cultivate a workforce ready to use AI- countries (UNESCO, 2015, p.242).
based systems’. In Asia-Pacific and Africa, as many as 35%
Countries of all income levels are striving to attract foreign
and 45% of company CEOs respectively express ‘extreme
talent – but not necessarily for the same reasons (see
concern’ about the availability of necessary skills (PwC, 2019).
Figure 6). Malaysia’s general population is not shrinking14 but
The current skills gap is preventing companies from embracing it does suffer from a shortage of skilled personnel, a situation
AI. Although 85% of company CEOs surveyed feel that AI will which also poses a dilemma to the numerous multinational
significantly change the way they do business in the next five electronics firms it hosts. Brain drain is a problem, with
years, nearly one-quarter of them have no plans to pursue AI Singapore alone absorbing 57% of the diaspora. Malaysia
‘at the moment’. A further 35%‘have plans do so in the next plans to become the sixth-biggest destination for international
three years (PwC, 2019). students by 2020; by 2012, it already tied with Japan for
eighth place, with 2.9% of all international PhD students in
science and engineering enrolling in Malaysian universities
The current skills gap is (see Figure 6).
preventing companies from
embracing AI Figure 6.
Distribution of international doctoral students enrolled in science
According to the survey, globally, CEOs see retraining and and engineering programmes by host country, 2012 (%)
upskilling as the best answer but it will take ‘time and money’. UK USA
More than one-quarter of Middle East CEOs see ‘hiring from
outside their industry’ as a potential solution, as do one in five France
9.2
CEOs in Western Europe (PwC, 2019).
7.4
The survey suggests that ‘governments and businesses need Australia
to work together to help their people adjust to the disruptive 4.6

impact of new technologies. A culture of adaptability and Canada 49.1
3.9
lifelong learning will be crucial to spreading the benefits of AI
and related technologies widely through society’ (PwC, 2019). Germany 3.5 2012
3.1
Switzerland
2.9
The skills shortage is fostering Japan 2.9

2.0
growing competition [in] both Malaysia
11.4
the domestic and international

Sweden
labour markets Other countries

Source: UNESCO Science Report: towards 2030 (2015), Figure 2.12
The skills shortage is fostering growing competition, as 13 By 2013, 15.5% of graduate students in Japan were of foreign origin
companies and institutions vie to attract and retain talent. (UNESCO, 2015).
This trend is affecting both the domestic and international 14 Malaysia’s population grew by 1.6% in 2014 (UNESCO, 2015).
Around the world, regional economic communities are Figure 7.

taking inspiration from the EU model by twinning scientific Long-term growth of tertiary-level international students,
integration with economic integration. Scientific mobility 1975–2013
is a key element of this strategy. The Association of South
East Asian Nations (ASEAN), for instance, plans to remove
restrictions to the cross-border movement of people and
services, as part of the ASEAN Economic Community’s drive
to create a common market and production base. In parallel, 4.1m 2013
ASEAN’s Plan of Action on Science, Technology and Innovation

2016–2025 ‘aims to strengthen scientific capacity in member
states by fostering exchanges among researchers both within
the region and beyond’ (UNESCO, 2015, p.76).
2.8m 2005
Scientists have never been

so mobile 1.7m 1995
1.1m 1985
Scientists have never been so mobile, particularly when
it comes to university students and the private sector. The
0.8m 1975
number of those opting to study abroad grew by almost 50%
between 2005 and 2013 (see Figure 7). In parallel, ‘increasingly,
firms are relocating their research laboratories abroad.
Source: UNESCO Science Report: towards 2030 (2015), Figure 1.4
Universities, by and large, remain much more immobile, with
only a small minority setting up campuses abroad’ (UNESCO,
2015, p.57).
5.7. The need for skills to keep pace
with technology
One in four scientific articles 5.7.1. A trend towards greater investment
was cosigned by a foreign in R&D
collaborator in 2014 For countries to reap the benefits of the Fourth Industrial
Revolution, they will need to invest in research infrastructure
and equipment and to train a critical mass of scientists and
engineers.
The desire to attract foreign talent is driving not only physical
The good news for developing countries is that the availability
mobility but also virtual mobility, thanks to the development
of cloud computing infrastructure and the increasing
of digital technologies. In research, there has been a global
processing power of common computers are lowering the
move towards open science through online international
cost of AI research infrastructure.
collaboration and open access to scientific publications and
underlying data. One in four scientific articles produced around However, as we have seen in section 5.3, much of the current
the world was cosigned by a foreign collaborator in 2014, investment in AI is being driven by the private sector, which
compared to one in five a decade earlier (UNESCO, 2015). has less access to financial support in developing countries.
There is a correlation between government support for R&D
With 60% of people now living in countries with a stagnant and business investment in the same. ‘Once countries are
or shrinking workforce, companies ‘that can’t find the talent prepared to invest more in research personnel and in publicly
they need in one country use remote working to employ people funded research, the propensity of businesses to invest in R&D
elsewhere’ (WEF, 2016). also grows’ (UNESCO, 2015, see Figure 8).
Figure 8.
Mutually reinforcing effect of strong government investment in R&D, 2010–2011
The size of the bubbles is proportionate to business sector funding of R&D as a share of GDP (%)
8 000
Finland
7 000
Denmark
Singapore
6 000
Luxembourg Korea, Rep.
Norway
Japan Sweden
5 000
Researchers (FTE) per million inhabitants
Portugal Canada
Austria
Slovenia UK Germany
4 000
USA
Belgium France
New Zealand
Netherlands
Estonia
Ireland
Russian Fed.
3 000
Slovakia Czech Republic
Lithuania Spain
Hungary
2 000
Latvia
Poland Italy
Malta Bulgaria
Malaysia
Argentina Ukraine
Costa Rica
Serbia
1 000
Turkey
Romania China
Kazakhstan Brazil
Uruguay
Colombia Mexico South Africa

Kuwait
0
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60
GERD funded from non-business sources as a share of GDP (%)
Source: UNESCO, 2015, Figure 1.3
As of 2015, the G20 still accounted for a disproportionate share UNESCO is currently developing a series of indicators to
of both human and financial investment in R&D: two-thirds monitor countries’ implementation of the Recommendation,
(64%) of the global population but 91% of research expenditure including those that track progress toward Sustainable
and 86% of the world’s researchers, according to the UNESCO Development Goal (SDG) 9.5, namely to Enhance scientific
Institute for Statistics. research, upgrade the technological capabilities of industrial
sectors in all countries, in particular developing countries,
Investment in R&D is, nevertheless, growing in developing
including, by 2030, encouraging innovation and substantially
countries. Between 2007 and 2013, investment in R&D
increasing the number of research and development workers per
among low-income countries even doubled to PPP$3.9 billion,
1 million people and public and private research and development
equivalent to PPP$38 000 per researcher15 (UNESCO, 2015,
spending.
Table 1.2).
The UNESCO Institute for Statistics, which is the United
Nations’ custodian for indicators SDG 9.5.1 and 9.5.2,
Once countries are prepared already surveys UNESCO member states regularly to collect
to invest more in research data on research expenditure and research personnel. The
personnel and in publicly Recommendation goes a step further by making it a legal
obligation for all Member States to supply national data
funded research, the regularly. It also provides a package intended to build up every
propensity of businesses to country’s community of scientists, which is the specific sense
invest in R&D also grows of SDG 9.5.
Despite the global economic and financial crisis, global

5.7.2. An underexploited pool of talent
expenditure on R&D grew from PPP$1,132 billion to
$PPP1,478 billion between 2007 and 2013. Even as Only three out of ten researchers in the world are women
governments in high-income countries reacted to the crisis (2013), meaning that most countries are underexploiting their
with austerity budgets, the business enterprise sector in many talent pool, including in the developed world. The notable
countries maintained or even augmented its own investment exceptions are Latin America and the Caribbean, as well as
in R&D as a buffer against the crisis (UNESCO, 2015, p.30). states in Europe and Central Asia which formed part of the
Communist bloc at the time of the Union of Soviet Socialist
Many developing countries, meanwhile, used the global
Republics (UNESCO, 2015, Figure 3.2).
commodities boom to raise their own investment in R&D,
including Argentina, Brazil, Ethiopia, Malaysia, Mexico and In the great majority of countries, women researchers work
Turkey. China invested massively in R&D, doubling its global mainly in the public sector. This phenomenon is uncorrelated
share of expenditure to 20% in the six years to 2013 (UNESCO, with a country’s level of development. For instance, women
2015, p.30). make up one-third of researchers in the business enterprise
sector in Botswana, Kenya, Namibia, Portugal, the Russian
Between 2007 and 2013, the number of researchers worldwide
Federation, South Africa, Thailand and Vietnam, compared
grew by 21% to 7.8 million. Researcher density grew in
to 26% in Belgium, Chile, Colombia, Singapore, Slovenia and
countries of all income levels. Among low-income countries,
Sweden, 20% in Slovakia, France and Poland, 14% in Germany
the number of researchers rose from 99 to 121 per million
and the Republic of Korea, 15% in Ethiopia, 11% in India and
inhabitants (in full-time equivalents) between 2007 and 2013.
8% in Japan. In the Arab world, 37% of researchers are women
Growth in some upper middle-income countries has been
but, for half of the Arab countries reporting data, barely any
spectacular, such as in Argentina (from 984 to 1,256 per
women at all work in the business sector (UNESCO, 2015).
million); China (853 in 2009 to 1,071); Malaysia (368 in 2006 to
These trends have implications for domestic innovation,
1,780 in 2012); and Turkey (715 to 1,189 per million). The global
insofar as the private sector is the main driver of the Fourth
average rose to 1,083 per million (UNESCO, 2015, Table 1.3).
Industrial Revolution.
In November 2017, UNESCO adopted the Recommendation
Women graduates dominate life sciences in many countries
on Science and Scientific Researchers, following broad
but are still a minority in mathematics, physics, engineering
consultations. This normative instrument provides countries
and computer sciences, fields that play a vital role in the digital
with guidelines on how to nurture endogenous research and
economy (UNESCO, 2015).
development.
15 Dollar amounts are calculated in purchasing power parities to facilitate

international comparisons. According to this approach, a given sum
of money, when converted into US dollars at the purchasing power
parity rate (PPP$), will buy the same basket of goods and services in all
countries.
Box 4.
Overcoming the physical limits to miniaturization of brain implants
In 2017, Dr Nazek El-Atab won the L’Oréal-UNESCO For
Women in Science International Rising Talent award for
her research into the fabrication of atomic-scale devices.
Thanks to the ‘quantum-confinement effect’, these devices
can tune electronic and optical properties of materials with
more precision than conventional devices, saving time and
money. Using a technique known as Atomic Layer Deposition
(ALD), the device can construct the active layer of nanoscale
transistors and memory cards in a single step.
Dr El-Atab’s current research focuses on the fabrication of

3D nanotube-based nano-electronics that will potentially
overcome the decade-long challenge of developing
electronics for implantation in the brain. Brain implants
could enable the deaf to hear, the blind to see and the
©l L’Oréal Middle East paralyzed to control robotic arms and legs.
Dr Nazek El-Ateb in her lab Dr El-Atab is a Postdoctoral Research Fellow at the MMH
labs of the Computer Electrical Mathematical Science and
Engineering Division of the King Abdullah University of
Most electronic components have been shrinking in recent Science and Technology in Saudi Arabia.
decades, including those found in cellular telephones, laptop
She is one of a growing number of L’Oréal-UNESCO For
computers, cameras and medical devices. The problem is
Women in Science laureates and fellows whose research
that traditional miniaturization techniques have reached
is nurturing the Fourth Industrial Revolution.
their physical limits. Below a certain size, the capacity for
memorization of these tiny devices degrades. For details, see: https://en.unesco.org/science-sustainable-future/women-
in-science
Since the establishment of the three most prestigious

international prizes for mathematics (Fields, Wolf and Abel),
5.7.3. The case for investing in cross-
only two women mathematicians have been recognized, out disciplinary education and new
of a total of 142 laureates. knowledge creation
AI and machine learning, together with a host of other
Women are still a minority innovative technologies that include robotics, nanotechnology,
in mathematics, physics, 3D printing and additive manufacturing, genetics and
biotechnology, are forming the bedrock of new markets for
engineering and computer products and processes that will shape the way people live
sciences, fields that play a vital and do business in the decades to come.
role in the digital economy This trend offers opportunities for developing countries to
tap into these markets. Many of the aforementioned fields
The L’Oréal–UNESCO Programme for Women in Science has
possess a set of both softcore and hardcore applications,
been raising the profile of exceptional women researchers for
the former being much cheaper than the latter to produce.
the past two decades through a system of annual prizes and
For example, softcore applications in nanotechnology utilize
research fellowships, in order to change attitudes towards
locally available chemicals and chemical reactions to produce
women researchers and provide young girls with positive role
new nano-materials in a bottom-up approach that is much
models (see Box 4). The programme’s slogan is ‘The world
needs science. Science needs women’. more cost-effective than the top-down alternative, which
focuses more on manufacturing solid-state components that
In 2019, the L’Oréal Foundation and UNESCO extended require costly clean rooms. Nanotechnology has been dubbed
the international awards dedicated to materials science to the ‘technology of the poor’ on account of these softcore
mathematics and computer science. Two mathematicians applications.
figure among the five laureates of the 2019 edition of the
Women in Science programme, Claire Voisin (France) and
Ingrid Daubechies (USA).
Many products developed using nanotechnology, biotechnology specific needs but it does mean that the company utilizes the
and nano-biotechnology are gradually replacing more existing technology without taking part in developing any of its
expensive alternatives as demand for them grows. This is components. This creates a lost opportunity for the company
the case for nano-based pesticides and insecticides, for to hone the skills of its employees while creating value.
instance. New low-cost fabrics designed using biotechnology,
The same goes for AI. Many AI solutions and algorithms are
meanwhile, are expected to replace current non-sustainable
purchased off the shelf by customers who played no part in their
petrochemical fabrics, which also come with a big price-tag.
development. The extent to which companies in developing
countries can draw on technical knowledge will largely
It is from [the basic research determine their ability to generate dynamic, environmentally
sustainable productive capacities. These companies will only
laboratories of universities] be able to leapfrog over costly investment in infrastructure by
that the next wave of high ensuring they possess a sufficient absorptive capacity to build
technologies will come, in on the backlog of unexploited technologies and benefit from
intimate synergy with digital the associated lower risks.
technologies At the government level, it will be of utmost importance for

developing countries to set up innovative practical policies and
Disciplines such as molecular biology, biotechnology, programmes that can successfully deliver food security and
nanotechnology and materials science are rooted in the robust rural development, in order to promote the desired
basic research laboratories of universities. It is, thus, from pro-poor economic growth, whereby the income of the poor
these laboratories that the next wave of high technologies will grows at a faster rate than that of the rest of the population.
come, in intimate synergy with digital technologies (UNESCO,
2015, p.427). This is why it is so vital for a country to invest in
basic research, if it wishes to be an active player in the Fourth
Scientists and engineers will
Industrial Revolution. also need to be retrained
Given the ways in which AI is already changing the conduct
The use of ready-made of research, scientists and engineers will also need to be
solutions has tended to retrained. A risk for developing countries is that some of
be favoured over spending their best researchers in AI may be attracted to multinational
corporations by lucrative salary packages and other benefits.
valuable time on developing a On 13 June 2018, Google announced the creation of the first
bespoke solution or product. centre for AI in Accra, the capital of Ghana. A month later,
Google began recruiting researchers specializing in machine
However, value creation nowadays often depends on using learning.
existing knowledge better rather than generating new
How can local employers compete effectively to retain the
knowledge (basic research), whether process technologies are
skills they need? It will be important to raise awareness among
developed domestically or not. Ever since the 1990s, the use of
policy- and decision-makers in developing countries of the
ready-made solutions has tended to be favoured over spending
policy options at their disposal to tackle internal and external
valuable time on developing a bespoke solution or product.
brain drain, among other challenges.
During the heady years of high oil prices in the first years of the
21st century, Russian companies preferred to import advanced As we have seen, the world of work is changing radically,
technologies, rather than use the boon to develop their own, as automation becomes more widespread. For a policy to
even though this led to a growing dependence on developed succeed, it will need to extend to the general population. In
countries in certain areas, such as pharmaceuticals and high- order to ensure the broad application of AI and proper quality
tech medical equipment (UNESCO, 2015, p.343). control of output and recommendations, AI technologies must
be disseminated and the general population must be trained to
In the software industry, ready-made solutions have been
use them, including through the use of open access systems.
extremely effective in reducing the time and cost of software
development but this has created a form of dependence. This is being done through UNESCO’s HOPE initiative
Buying a software off the shelf may work well when the end- (see p. 31), for instance. Over the past two years, HOPE has
user lacks the knowledge and skills to tailor a solution to their brought together 18 universities, centres and other bodies
around the FREE and open source tools for WATer resource ◼◼ entrepreneurship programmes for secondary schools and
management (FREEWAT16) project, a Horizon 2020 project universities, including robust pre-incubation and incubation
financed by the European Commission. FREEWAT takes an programmes, as well as programmes that build leadership
innovative participatory approach, bringing together technical skills among university graduates.
staff and relevant stakeholders, including policy- and decision-
The integrated science teaching approach fluidifies the
makers, to design scenarii for the application of conjunctive
boundaries between biology, chemistry, mathematics and
water policies. The HOPE consortium has also organized
physics. Current education systems produce graduates who
capacity-building workshops and seminars and trained
lack a combination of skills. The integrated science teaching
700 participants.
approach is striving to overcome this relative disciplinary
FREEWAT is part of the ICT4WATER cluster, which combines isolation by establishing internal, external and historical
ICT and water management projects, all of which are co- connections between science (such as biology), mathematics
funded by the European Commission. Their common goal is to and technology (such as nanotechnology or robotics).
increase efficiency in water management and enable greater
It is important to start with school curricula and to ensure
cooperation among water regulators, operators and users by
continuity through the secondary and tertiary levels of
deploying technical solutions. The FREEWAT approach is being
education to overcome the relative disciplinary isolation
used for 14 case studies within the European Union, three case
of scientists in the traditional (fragmented) approach to
studies in neighbouring countries (Switzerland, Turkey and
science teaching. School curricular need to emphasize
Ukraine) and is being applied to a large transboundary aquifer
three dimensions: blurring the boundaries between
in Africa (UNESCO’s Groundwater Resources Governance of
traditional scientific disciplines; fostering original ideas and
Transboundary Aquifers project). Training courses have been
creative thinking and; reducing the amount of rote learning
run in 53 countries spread across all five continents.
(memorization of detail).
A good example of a creative educational programme is

5.7.4. An integrated approach to science enquiry-based (hands-on) science education for the first
teaching 12 years of schooling. Enquiry-based (hands-on) learning
teaches children the scientific method, which hinges on
The trends outlined above imply that developing countries will observation, measurement and experimentation. Pupils learn
need to take a cross-disciplinary approach to education, in to formulate hypotheses that they then test and adjust, in light
order to equip the workforce, in general, and new university of the results of their experiment. Criticism is a basic tenet of
graduates, in particular, with the requisite skills to reap the the scientific method. As Sir Peter Medawar put it in his Advice
benefits of the Fourth Industrial Revolution. The integrated to a Young Scientist (1979), ‘all experimentation is criticism.
approach to science teaching has become a priority for If an experiment does not hold out the possibility of causing
UNESCO’s International Basic Sciences Programme. one to revise one’s views, it is hard to see why it should be
done at all’.
Developing countries will need

to take a cross-disciplinary Enquiry-based (hands-on)
approach to education learning teaches children the
scientific method
The development of an innovation ecosystem should also
include: Enquiry-based science education promotes both conceptual
understanding and the development of capabilities widely
◼◼ innovative programmes and projects related to the
recognized as being vital for everyone in the twenty-first
Sustainable Development Goals to tackle infrastructure
century, such as critical thinking, team work, consideration
problems in the spheres of water, health, energy, agriculture
of alternatives and effective communication. Children are
and, most importantly, ICT networks to ensure Internet
motivated by the satisfaction of having made sense of
connectivity everywhere;
something they did not previously understand. Enquiry is
◼◼ a creative set of educational programmes that include well- only one of a range of ways of learning but it is a particularly
designed science and engineering fairs (competitions) and important one, being based on research and modern views of
well-established science clubs in all schools, as well as how pupils learn, as well as the importance of progressively
science parks within regional knowledge hubs that could developing a child’s understanding of phenomena in the world
be associated with these schools; around them, as their experience grows.
16 See: www.freewat.eu
Currently, it is difficult to add new courses such as A growing number of least developed countries are putting
nanosciences, robotics, etc. early on in a child’s education, cross-disciplinary programmes in place to stimulate a vocation
as this overloads the curriculum, even though students will for science among the young and give graduates the skills
encounter this cross-disciplinary approach at a later stage they need to devise innovative solutions for key development
in their education. For instance, advanced computer skills, challenges in priority fields such as agriculture, industry,
quantitative thinking and mathematics all are prerequisites civil engineering, security and health. Digital technologies
for any undergraduate studying microbiology today. are being introduced into schools to facilitate this strategic
transformation.
Scientists educated in an integrated manner, regardless of their
ultimate professional specialty, will share a common scientific
language, facilitating both cross-disciplinary understanding
A growing number of least
and collaboration by enabling interdisciplinary communication
and nurturing the type of researcher who is well-equipped to developed countries are
cope with the rapid and far-reaching advances of the Fourth putting cross-disciplinary
Industrial Revolution in terms of understanding, experimental programmes in place
techniques and computational analysis.
Active learning workshop at RobotsMali
©UNESCO Bamako Office, active learning workshop at RobotsMali
Mali is one such example. The country’s Strategic Framework Through its engineering programme, meanwhile, UNESCO
for Economic and Sustainable Development (2016-2018) deems plans to organize short courses on robotics in 2019 in Africa
‘it particularly timely to use ICTs to develop the teaching of and beyond in partnership with companies working on AI like
science and technology at secondary level and to encourage Quanser. The aim is to demonstrate how secondary schools
girls to opt for scientific careers’17. In 2016, the government can incorporate AI into teaching using the hands-on learning
created an annual Science Fair (Fête de la science) and Miss approach. The programme is using the same approach to
Science contest. incorporate ethical issues into AI teaching.
On 19 March 2018, the government inaugurated the National

Centre for Collaborative Education in Robotics (Centre
national collaboratif de l’éducation en robotique, RobotsMali)
5.7.5. Building awareness of AI and critical
in the new Cité universitaire of Kabala south of Bamako. In thinking
addition to teaching robotics and coding, RobotsMali provides
In spite of the indubitable promise of AI, the use and deployment
training in AI, electronics and the Internet of Things, as
of AI as black boxes without a proper understanding of how
well as in entrepreneurship, marketing, design and project
AI works and the consequences of these technologies is
management.
problematic.
In addition to teaching

robotics and coding, Specific efforts need to
RobotsMali provides training be devoted to training
in AI, electronics, the Internet researchers from the
of Things,…, entrepreneurship, developing world in basic
marketing, design and project research related to AI
management
This is why research and education in AI are a key factor in
curbing effects that may not only limit access to AI potential
As part of its Youth Capacity-building Programme in but also damage our societies. These effects include biases in
STI, UNESCO’s Bamako office provided RobotsMali with AI, privacy issues and the proliferation of misleading or false
educational material and equipment in 2018, in coordination information. Moustapha Cissé from Facebook AI Research
with UNESCO’s Africa Department and with the financial cites the examples of facial recognition systems which work
support of China. This material included 40 laptop computers better with European faces than with African ones and AI
and kits of robots and drones for different age groups: children systems that can identify skin cancers better in white patients
up to 13 years, teenagers and university students. than in black ones (UNESCO, 2018b).
The project was implemented in partnership with the Ministry There will be no way to address these issues unless
of Education and the Ministry of Innovation and Scientific education and basic research in AI expand at the same rate
Research, along with the Malian Association of Women as technological advances in AI. The developing world is
Engineers. UNESCO’s Youth Capacity-building Programme particularly vulnerable in this regard. With AI having been
in STI also plans to train teachers in primary and secondary developed in Western societies, the detrimental effect of
schools across Mali in how to teach robotics and to support biases are expected to be particularly acute in developing
schools wishing to create their own robotics clubs. countries, which means that specific efforts need to be devoted
to training researchers from the developing world in basic
research related to AI (Zou and Schiebinger, 2018).
UNESCO plans to organize The ICTP is in the process of establishing a diploma course in
short courses on robotics in Quantitative Life Sciences that includes a strong component
2019 in Africa and beyond on AI and data science.
In parallel, the ICTP has been raising awareness and

fostering critical thinking about AI in the scientific community
17 « II parait particulièrement opportun de développer l’enseignement des through a series of training workshops and research
sciences et techniques par l’introduction de TIC au secondaire en favorisant activities. For instance, the ICTP hosted a school on Large-
l’accès des filles aux séries et filières scientifiques. » Cadre stratégique pour
la relance économique et le développement durable (2016-2018) Scale Problems in Machine Learning in August 2002.
The ICTP– International Centre for Theoretical Sciences programmes. The ICTP has also been reaching out to the
Winter School in November 2018 focused on the theme of wider public through conferences and a series of video clips
Learning and AI. for Raiscuola18.
ICTP is working with centres of excellence in Africa, such as During Mobile Learning Week in March 2019,UNESCO also
the network of the East African Institute for Fundamental reminded a young audience that popular environmental
Research (a category II centre) in Kigali, Rwanda, to integrate applications (apps) bring many benefits but also demand
AI into tertiary education curricular and basic research sacrifices on the part of the user (see Box 5).
Mobile apps can help people to identify a particular animal species.
© Getty Images/FTiarei
18 See, for example (in Italian): http://www.raiscuola.rai.it/speciale-unita/

antonio-celani-intelligenza-artificiale-dalleorigini-ad-oggi/286/41405/
default.aspx
Box 5.
Building awareness of the pros and cons of common apps
In 2019, the theme of Mobile Learning Dr Will Logan, who heads the difference between apps that were
Week at UNESCO headquarters was International Center for Integrated simply connected objects and those
artificial intelligence and sustainable Water Resources Management in the that incorporated AI components to
development. USA, a ‘category 2’ centre of UNESCO, enable an automatic identification
presented iRain, a free app that can process.
At a workshop organized jointly by
be used by individuals, schools, local She cautioned budding citizen
UNESCO’s Natural Sciences Sector
governments and national hydrological scientists to remain vigilant when
and Social and Human Sciences Sector
services alike to estimate rainfall (see using apps equipped with AI. Machines
on 5 March during Mobile Learning
also Box 1). excelled at data processing and at
Week, the public was reminded that
popular environmental applications Dr Logan demonstrated how the apps compiling statistical records, she said,
(apps) bring many benefits but also could be used to anticipate water- but deep learning worked like a black
demand sacrifices. These sacrifices related hazards such as flooding, box. The machine did not explain how
might be obvious, such as the amount droughts and storms but also to it arrived at a certain conclusion, so
of time a person spent using the app track rainfall over the past 35 years, the user did not learn how to recognize
or their decision to relinquish their for instance, to see whether climate future specimens.
right to privacy by giving the app their change had affected precipitation over Apps which were simply connected
geographical location. a given area, such as the Sahara. He objects, on the other hand, and
mentioned that people were tailoring thus incapable of providing an
Other sacrifices were more insidious,
the app to their own needs and finding automatic identification process,
the public heard. Clients tended to
uses for them that the developer had could communicate verbally to offer
develop a relationship of dependency
never imagined, a trend applauded by advice and an explanation of how they
on the satellites and cloud used by
Dr Nurock. arrived at their findings, enabling the
apps by accepting the information
user to learn the right terminology for
retrieved from them as the ultimate Dr Régine Vigne-Lebbe explained that
describing a plant or animal. Armed
truth, explained Dr Vanessa Nurock, the apps developed by her university
with this knowledge, the user would
Associate Professor in Political Theory in collaboration with the Museum
be better equipped in future to identify
and Ethics at Paris 8 University. In this national d’histoire naturelle (National
a species and read relevant literature.
way, apps had become the modern Museum of Natural History) had been
equivalent of the Gods or oracles designed to broaden the public’s Dr Vigne-Lebbe predicted that future
worshipped in the past, she said. environmental awareness. These apps apps would combine human active
help people to identify a particular observation with automatic recognition
The session focused on whether
plant or animal species. for optimal results.
environmental apps were effective
educational tools that could also Professor of Biodiversity Informatics
For details, see: https://en.unesco.org/news/
foster citizen science – or just another at the Sorbonne University in Paris, experts-debate-whether-environmental-apps-
consumer product. Dr Vigne-Lebbe underscored the are-fostering-citizen-science
The implications of AI
for policy-making and
foresight

The implications of AI for policy-making and foresight
6. The implications of AI for policy-making

and foresight
6.1. A growing number of national The Natural Sciences Sector has a number of tools at its
disposal. It already provides Member States with policy advice
AI strategies to strengthen the governance of national innovation systems.
For example, UNESCO’s Global Observatory of Science,
The first national AI strategies were adopted by Canada Technology and Innovation Policy Instruments (GO-SPIN)
and Japan in March and April 2017. Since then, the list has maps national innovation systems to inform policy-making.
lengthened. Policy researcher Tim Dutton is regularly updating
The issues outlined in the preceding section with regard to
a list19 of countries which have adopted AI strategies, or are
the growing concentration of AI-related research in the hands
contemplating doing so. In addition to Canada and Japan,
of a minority also underscore the need to incorporate issues
it includes Australia, China, Denmark, Finland, France,
related to AI in the policy advice provided by UNESCO, in much
Germany, India, Kenya, Republic of Korea, Singapore, United
the same way governments were advised to mainstream the
Arab Emirates, United Kingdom and USA.
concept of sustainable development in planning processes
after publication of the Brundtland Report, Our Common
How can UNESCO ensure that Future, 30 years ago. There are plans to launch a series of
occasional papers and policy briefs on emerging issues in STI
developing countries are more policy, within the UNESCO Science Report library. One of the
than simply providers and first papers will be looking at policy implications of AI.
users of data, more than just UNESCO will also be examining the possibility of mapping
clients of AI? global trends in AI implementation and research, to detect
hotspots and coldspots and provide evidence for capacity-
Dutton has observed that ‘no two strategies are alike, with building strategies and policies.
each focusing on different aspects of AI policy: scientific
research, talent development, skills and education, public and
private sector adoption, ethics and inclusion, standards and
regulations, and data and digital infrastructure’. 6.2. Existing policy tools can frame
The UNESCO Science Report 2020 will be monitoring the AI strategies
development of national and regional AI strategies since 2017
as part of its mandate for tracking global trends in science AI is not the first disruptive technology with which policy-
governance. makers have been confronted. There have been precedents,
such as the emergence of biotechnology and gene editing. In
Should STI policies promote AI research in developing
developing a national strategy for AI, governments can draw
countries? Currently, much of the developing world tends to
on earlier approaches to emergent technologies.
be a testing ground for technologies developed elsewhere.
How can UNESCO ensure that developing countries are more Existing policy tools can frame national strategies for AI.
than simply providers and users of data, more than just clients This mix includes legislation and other forms of regulation,
of AI? technology foresight exercises, research priority-setting
exercises, participatory approaches to policy formulation and
19 See https://medium.com/politics-ai/an-overview-of-national-ai-
recourse to ethics committees.
strategies-2a70ec6edfd
It will be important to carry out analytical studies of the Data Protection Regulation (GDPR), meanwhile, imposes
socio-economic impact of AI and to devise policies which stricter data privacy rules on all entities doing business in
ensure accountability and transparency, data privacy, respect the European Union (Bradsher and Bennhold, 2019).
for civil rights and so on. The development of AI will need to
In Africa, the Policy on Science and Technology adopted in 2011
be regulated at both the national and international levels to
by the Economic Community of West African States’ (ECOWAS)
ensure that, in our globalized world, AI does not become an
agent of human and environmental exploitation or neglect. ‘is an integral part of Vision 2020’, the subregion’s development
blueprint to 2020 (UNESCO, 2015, p.476). The East African
Community’s Common Market Protocol (2010) makes provision
for market-led research, the promotion of industrial research
and the transfer, acquisition, adaptation and development of
6.3. AI systems can contribute modern technology. States are encouraged to collaborate with
to policy-making the East African Science and Technology Commission and to
establish a fund for R&D for the purpose of implementing
Policy-makers should complement the focus on AI governance the provisions in the protocol (UNESCO, 2015). In 2017, the
by using AI to extend existing statistical and econometric Southern African Development Community ratified its Protocol
models for better decision-making. on Science and Technology.
Together with big data, AI can provide keen insights into trends Even though the focus of the Association of South East Asian
or answer ‘black box’ type questions where there is not a Nations (ASEAN) has always been on the creation of a single
clear relationship between input and output. For example, market along the lines of the European model, ‘leaders have
topic modelling can be used to characterize the evolving, long acknowledged that successful economic integration
dynamic policy environment and thereby identify priority areas. will hinge on how well member states manage to assimilate
Similarly, clustering algorithms can help to identify non-trivial science and technology’. ASEAN’s Plan of Action on Science,
taxonomies, in order to develop tailored policy instruments. Technology and Innovation 2016–2025 ‘aims to strengthen
scientific capacity in member states by fostering exchanges
among researchers both within the region and beyond’.
Could AI produce models that Established in 2015, the ASEAN Economic Community is
provide an evidence-based expected to spur scientific cooperation among member
justification for why countries countries, while enhancing the role of the ASEAN University
Network, which counts 30 members (UNESCO, 2015, p.7620).
should invest in science,
technology and innovation?
Could AI even serve as an advocacy tool? For instance, could 6.5. Global strategies will also be
AI produce models that provide an evidence-based justification
for why countries should invest in science, technology and
necessary to ensure collective
innovation? choices
Given the ethical issues raised by AI and the fact that
multinational companies are driving much of the research in AI,
6.4. Regional economic global strategies and guidelines will also be necessary. In the
USA, the University of Stanford has set up a Global Digital Policy
communities can help to frame Incubator21 to provide ‘a vehicle for global multistakeholder
AI strategies collaboration between technologists, governments, private
sector companies, diplomats, international organizations,
Regional economic communities may yet prove to be a fertile academics and civil society in developing norms and policies
breeding ground for national and regional AI strategies. that enhance security, promote economic development and
Increasingly, communities see regional scientific integration reinforce respect for human rights’.
as an integral part of regional economic integration. Many are
taking inspiration from the European model of integration. The 20 See also this blog: www.unesco.org/new/en/media-services/
single-view/news/regional_economic_communities_a_conduit_for_
European Commission adopted the Digital Single Market in southsouth_co/
2015 and is currently working with member states to develop 21 https://cddrl.fsi.stanford.edu/global-digital-policy-incubator/front-
a coordinated strategy for AI. The European Union’s General global-digital-policy-incubator
Within the United Nations, UNESCO’s Commission on the Our goal is to ensure consistency and complementarity
Ethics of Science and Technology (COMEST) is mandated between OECD work and other international initiatives,
to formulate ethical principles that could provide decision- including under the G20 and G7, the European Commission,
makers with criteria that extend beyond purely economic organizations such as the Institute of Electrical and Electronic
considerations. COMEST published a report on Robotics Ethics Engineers and the United Nations, including, importantly,
in 2018 and submitted a preliminary study on the ethics of AI22 UNESCO itself’. A succession of government ministers and
to UNESCO’s Executive Board in April 2019. COMEST began participants from academia, intergovernmental organizations,
reflecting on the ethics of the Internet of Things in 2017, the the private sector, technical community, media and civil society
year UNESCO’s International Bioethics Committee adopted all called for the development of ethical principles to govern
its own report on Big Data and Health. AI on the basis of transparency and accountability.
The time is more than ripe to If we let the invisible hand of
define the ethical principles the market operate freely, we
[which will] ensure that AI will get useful apps but our
serves collective choices, privacy will be eroded and
based on humanistic values inequalities will grow
Audrey Azoulay Fabrizio Hochschild
Fabrizio Hochschild Drummond, Assistant Secretary-General

For the Director-General of UNESCO, Audrey Azoulay, ‘the for Strategic Coordination within the Executive office of the
issues raised by artificial intelligence are not technological. Secretary-General of the United Nations, warned that ‘the
They concern our own humanity, raising scientific, political, current preference for non-binding international agreements,
philosophical and ethical questions […]. The time is more part of the shift from cooperation to competition, is making
than ripe to define the ethical principles that must serve as non-binding treaties much more attractive but we cannot rely
a foundation and framework for this disruption, to ensure on everyone to be good. […] If we let the invisible hand of the
that artificial intelligence serves collective choices, based on market operate freely, we will get useful applications but our
humanist values’. She made these comments in Paris on 4 privacy will be eroded and inequalities will grow, contributing
March 2019, while opening UNESCO’s global conference23 to the polarization of our societies’, he argued. Recently,
on the theme of Principles for AI: towards a Humanistic some leading public figures have come out in favour of global
Approach? governance of data (see Box 6).
At the conference, Angel Gurría, Secretary-General of the Within the global community, the United Nations’ Convention
OECD, spoke of the need to work with UNESCO in a concerted on Certain Conventional Weapons has, meanwhile, been
effort to ‘make AI less artificial and more intelligent’. He stated discussing an international treaty which would ban lethal
that ‘we must progress together on AI-related technical, ethical autonomous weapons. Several countries are investing heavily
and legal issues, to foster alignment of standards and codes in this field. As of late 2018, member states had not reached
of conduct, and inter-operability of laws and regulations... consensus on the need for such a treaty.
22 Preliminary Study of the Technical and Legal Aspects relating to the

desirability of a Standard-setting Instrument on the Ethics of Artificial
Intelligence (document 206 EX/42): https://unesdoc.unesco.org/
ark:/48223/pf0000367422
23 See : https://en.unesco.org/news/participants-global-unesco-
conference-artificial-intelligence-urge-rights-based-governance-ai
Box 6. Leaders in Davos endorse principle of global governance of data

There is already some public discourse in favour of global governance of data. In separate interventions at the World Economic
Forum in Davos in January 2019, for instance, the leaders of Japan, South Africa and Germany all called for greater international
oversight of data usage.
Prime Minister Shinzo Abe announced that Japan would be using its chairmanship of the Group of 20 nations (G20) in 2019 to
promote the idea of expanding World Trade Organization rules beyond goods and services to encompass trade in data as well; he
expressed the hope that the G20 in Osaka would be remembered as the start of global data governance.
Citing cybersecurity as a priority, President Cyril Ramaphosa of South Africa said that greater oversight of the technology sector
would be on the agenda of African Union leaders meeting in February 2019 in Addis Ababa, Ethiopia (Bradsher and Bennhold, 2019).
In Davos, Chancellor Angela Merkel of Germany called for a ‘common digital market’ within the European Union. She cited the need
for ethical standards in AI and genetic engineering, as well as for the handling and ownership of data. The European Union’s General
Data Protection Regulation (GDPR) took effect in May 2018; it imposes stricter data privacy rules on all entities doing business in the
European Union (Bradsher and Bennhold, 2019).
Speaking in Davos in January 2019, the CEO of Microsoft himself voiced support for a common standard on data privacy rules. Satya
Nadella argued that US citizens deserved the same level of protection for data misuse as Europeans and expressed the hope that ‘in
the United States, we do something similar [to the GDPR] and that the world converges on a common standard’. He tweeted from
Davos that ‘privacy is a human right, we need a GDPR for the world (see tweet)’.
Conclusion
7. Conclusion
From the foregoing, it is evident that the Fourth Industrial them through simple experiments. However, the accuracy of
Revolution offers UNESCO’s science programmes numerous machine learning techniques has been called into question,
opportunities to improve their service delivery to member since an AI algorithm is not always capable of producing the
states. The predictive and prescriptive capabilities of AI same answer twice to the same question. This inconsistency
systems have enormous potential for fields as diverse as poses a problem for reproductibility in science. It also sheds
medical research and diagnostics, ecological and biodiversity doubt on the reliability of some decisions currently being made
research, freshwater management, geoscience research, by AI algorithms, such as in determining the length of a prison
disaster risk reduction and strategic foresight. sentence. AI algorithms may also contain gender or racial
biases. It will be very important for scientists to preserve their
AI systems are already helping scientists working with critical thinking in approaching AI.
UNESCO to analyse an unprecedented volume of data in fields
ranging from theoretical physics to ecological research, water For countries to reap the benefits of AI, they will need to
have the requisite human and institutional capacity. That will
management and climate change assessments.
require investment in research infrastructure and equipment
AI also presents a number of challenges. In a mid- to long- and a critical mass of well-trained scientists and engineers.
term perspective, it could fundamentally alter the way in UNESCO can provide policy advice and guidance and help
which research is conducted, structured and understood. This countries to benchmark their progress against that of other
evolution will require scientists and engineers to learn new countries, through tools such as the UNESCO Science Report,
skills and could eventually result in AI outcompeting research World Water Development Report, the database of the UNESCO
organized by, and for, humans. AI is already mimicking the Institute for Statistics and the Global Observatory of Science,
scientific method by generating hypotheses and verifying Technology and Innovation Policy Instruments.
Conclusion
There are opportunities for developing countries. The must not become simply data subjects and users of data
availability of cloud computing infrastructure and the produced elsewhere. UNESCO can help them to collect,
increasing processing power of common computers are develop and manage quality data for their own benefit, thereby
lowering the cost of AI research infrastructure. Open science helping to narrow the scientific divide.
and open source movements are striving to level the playing
field. The adoption of policies encouraging transparency Up until now, data collection has been a costly, difficult task.
and sharing of core datasets across both public and private As a result, the data gathered are often incomplete, especially
actors could stimulate innovation-oriented competition while in developing countries and remote areas. The Internet of
enhancing research productivity, including in developing Things could radically change that, as this low-cost, low-power
countries. technology does not require any pre-existing infrastructure
and can be deployed in most remote areas. UNESCO has been
The current domination of AI research by a small number of
advocating the use of the Internet of Things for development,
multinational corporations, and their propensity to take over
since sensors can be attached to the nodes to support a wide
smaller competitors, is in danger of widening the technological
range of applications, such as water or air quality monitoring,
divide, if this trend goes unchecked. In developing countries,
the private sector has less access to finance, hampering animal tracking and disease mapping.
the emergence of start-ups in AI and reducing employment UNESCO could also add value to current studies mapping
opportunities. Talented researchers are being snapped up by trends in AI, which tend to focus on the leaders in this field
multinational corporations able to offer higher wages and (or hotspots). By mapping the so-called ‘coldspots’ in AI and
other benefits. UNESCO can advise governments on their analysing the reasons why these mainly developing countries
best policy options for tackling these and other challenges. are lagging behind, UNESCO could not only help to form a
The third key component of AI after computer power and more rounded picture of global trends in AI but also pinpoint
algorithms is data, vast quantities of it. Developing countries some of the policy challenges faced by these ‘coldspots’.
Bibliography
Bibliography
Amellal, Karim (2018) La Révolution de la servitude. ISBN: Kitano, Hiroaki (2018) Artificial intelligence to win the Nobel
9782354571283, Demopolis, Paris, 208 pp. Prize and beyond: creating the engine for scientific
discovery.
Asgard and Roland Berger (2018) Artificial Intelligence – A
Strategy for European Startups. Recommendations for AI Magazine. Association for the Advancement of Artificial
Policymakers. Intelligence, 37(1): https://www.aaai.org/ojs/index.
php/aimagazine/article/view/2642
Biggs, Phillippa; Garrity, John; LaSalle, Connie and Anna
Polomska (2015) Harnessing the Internet of Things McGovern, Amy et al. (2017) Using artificial intelligence to
for Global Development. A contribution to the ITU/ improve real-time decision-making for high-impact
UNESCO Broadband Commission for Sustainable weather. Bulletin of the American Meteorological
Development. Society: doi:10.1175/BAMS-D-16-0123.1
Bradsher, Keight and Katrin Bennhold (2019) World leaders Mulatu, K.; Mora, B.; Kooistra, L; and M. Herold (2017).
at Davos call for global rules on tech. New York Biodiversity monitoring in changing tropical forests:
Times, 23 January. a review of approaches and new opportunities.
Remote Sensing, 9(10), 1059.
Collas, Aurélie ((2019). Les travailleurs ubérisés sont les
prolétaires du XXIe siècle. Le Monde, 7 avril. Muntaner, C. (2018) Digital platforms, gig economy,
precarious employment and the invisible hand of
Cockburn, I. M.; Henderson; R.; and Stern S. (2017) The
social class. International Journal of Health Services,
Impact of Artificial Intelligence on Innovation: www.
48(4):597-600. doi: 10.1177/0020731418801413.
nber.org/chapters/c14006.pdf.
Nature (2018) Deep learning for biology. Nature, 554, 555–
Deutsche Welle (2018) Chinese takeover of Leifeld collapses
557: www.nature.com/articles/d41586-018-02174-z
ahead of expected German veto: https://www.dw.com/
en/chinese-takeover-of-leifeld-collapses-ahead-of- Norouzzadeh, Mohammad Sadegh; Nguyen, Anh; Kosmala,
expected-german-veto/a-44906055. Margaret; Swanson, Alexandra; Palmer, Meredith S.;
Packer, Craig and Jeff Clune. (2018) Automatically
Hinchcliffe, Tim (2018) Medicine or poison? On the ethical
identifying, counting and describing wild animals in
implications of AI implants in humans. The Sociable,
camera-trap images with deep learning. Proceedings
2 July. See: https://sociable.co/technology/ethics-ai-
of the National Academy of Sciences, 5 June:
implants-humans/
https://doi.org/10.1073/pnas.1719367115
Horowitz, Michael, Kania, Elsa B.; Allen, Gregory C. and
OECD (2016) Science and Technology Outlook. Organisation
Paul Scharre (2018) Strategic Competition in an Era
for Economic Co-operation and Development, Paris.
of Artificial Intelligence. Centre for New American
Security, Washington DC. IFR (2017) World Robotics OECD (2017) Science, Technology and Innovation Scoreboard
2017. Industrial Robots. Executive Summary. 2017: the Digital Transformation. Organisation for
International Federation of Robotics. Economic Co-operation and Development, Paris.
Jalalkamali, A ; Moradi M and N. Moradi (2015) Application ONS (2019) The Probability of Automation in England: 2011 and
of several artificial intelligence models and 2017. Office for National Statistics, UK.
ARIMAX model for forecasting drought using the
Otake, T. (2016) IBM big data used for rapid diagnosis of rare
Standardized Precipitation Index. International
leukemia case in Japan. Japan Times, 11 August.
Journal of Environmental Science and Technology, 2(4),
1201–1210: https://link.springer.com/article/10.1007/
s13762-014-0717-6.
Bibliography
Pax (2018) Belgium votes to ban killer robots. Pax: https:// UNESCO (2017) Report of COMEST on Robotic
www.paxforpeace.nl/stay-informed/news/belgium- Ethics. See: http://unesdoc.unesco.org/
votes-to-bankiller-robots. images/0025/002539/253952E.pdf
PwC (2019) CEOs’ Curbed Confidence Spells Caution. 22nd UNESCO (2018a) Countering the monopolization of
Annual Global CEO Survey. Launched at World research. Interview of Yoshua Bengio by Jasmina
Economic Forum, Davos, Switzerland, 23 January. Šopova. UNESCO Courier, July. UNESCO, Paris.
Quoctrung, Bui (2015) The most common job in every state. UNESCO (2018b) Democratizing AI in Africa. Interview of
National Public Radio, USA, 5 February: https://www. Moustapha Cissé. UNESCO Courier, July. UNESCO,
npr.org/sections/money/2015/02/05/382664837/map- Paris.
the-most-common-job-in-every-state
Russia Today (2017) ‘Whoever leads in AI will rule the
Rasp, Stephan; Pritchard, Michael S. and Pierre Gentine world’: Putin to Russian children on Knowledge Day.
(2018) Deep learning to represent subgrid processes Russia Today, 1 September: https://www.rt.com/
in climate models. Proceedings of the National news/401731-ai-rule-world-putin
Academy of Sciences: www.pnas.org/cgi/doi/10.1073/
Torney, C. J., Lloyd-Jones, D. J., Chevallier, M., Moyer, D.
pnas.1810286115
C., Maliti, H. T., Mwita, M. and G. C. Hopcraft (2019).
Singer, G. (2018) How is artificial intelligence changing A comparison of deep learning and citizen science
science? Interview given to Physics.org, 25 techniques for counting wildlife in aerial survey
May: https://phys.org/news/2018-05-artificial- images. Methods in Ecology and Evolution.
intelligence-science.html#jCp
Wang, Wen-Chuan; Chau Kowk-Wing; Tian Cheng, Chun
Tsang, Amie (2016) Midea of China moves a step closer to and Qiu Lin (2009) A comparison of performance of
takeover of Kuka of Germany. New York Times, 4 July. several artificial intelligence methods for forecasting
monthly discharge time series. Journal of Hydrology,
Torbert, Georgina (2019) Statistician raises red flag about
374, issues 3–4.
reliability of machine learning techniques. Digital
Trends, 17 February. See: https://www.digitaltrends. WEF (2016) Four changes shaping the labour market.
com/cool-tech/machine-learning-reliability- World Economic Forum: https://www.weforum.org/
reproducibility/ agenda/2016/01/four-changes-shaping-the-labour-
market/
UNCTAD (2017) World Investment Report 2017: Investment
and the Digital Economy. United Nations Conference WWWF (2017) Artificial Intelligence: the Road Ahead in Low
on Trade and Development, Geneva. and Middle-Income Countries. World Wide Web
Foundation.
UNESCO (2015) UNESCO Science Report: towards 2030.
UNESCO Paris. Zou, J. and L. Schiebinger (2018) AI can be sexist and racist
– it’s time to make it fair. Nature 559, 324-326.
UNESCO (2016) The Globalisation of Cultural Trade: a Shift
in Cultural Consumption – International Flows of
Cultural Goods and Services, 2004–2013. UNESCO
Institute for Statistics.
Glossary of terms
Glossary learning sets up basic parameters about the data and trains
the computer to learn on its own by recognizing patterns
of terms from processing many layers of data (sas.com).
Filter bubble (or Echo chamber) is an unwanted feature

of an algorithm which, when used as a black box, can do
things that were not explicitly intended by its programmer.
For example, the algorithm that builds a person’s Facebook
feed filters information to show the person the things they
like most. In this way, it is easy to convince a person of an
idea that is false, such as that vaccines cause autism. The
Artificial intelligence makes it possible for machines
person can then live in a virtual bubble in which almost
to learn from experience. Intelligent systems use a
everyone in their feed is convinced that vaccines cause
combination of big data analytics, cloud computing,
autism. A programmer can ‘open’ the black box to see how
machine-to-machine communication and the Internet
it works then try to fix the unwanted feature.
of Things to learn to perform cognitive tasks: sensing,
processing of oral language, reasoning, decision-making, Internet of things refers to the extension of Internet
displacement and manipulation of objects, etc. AI is connectivity to physical devices by embedding the latter
empowering new kinds of software and robots to act as with electronics, software, sensors, etc. to enable them to
self-governing agents capable of operating much more communicate and interact over the Internet while being
independently from the decisions of their human creators remotely controlled and monitored. By 2017, there were
and operators than machines have previously done. more than 8 billion ‘smart’ devices embedded in clothing,
There is also potential for them to adapt to new working vehicles, electrical goods (watches, home appliances), etc.
environments without any reprogramming (OECD, 2016).
The term artificial intelligence was coined at a conference at Machine learning is a method of data analysis that
Dartmouth College in New Hampshire, USA, in 1956. automates analytical model building. It is a branch of
artificial intelligence based on the idea that systems can
Black box refers to the use of a device without an
learn from data, identify patterns and make decisions with
understanding of its inner workings. Black box solutions
minimal human intervention (sas.com).
may discriminate between the signal and background,
for instance, without providing the operator with an Overfitting of data is a cause of poor performance in
understanding of the relationship between input and output. machine learning. Machine learning algorithms use training
datasets to build a model. If they learn for too long, they
Blockchain technology is a distributed, decentralized and
begin to assimilate irrelevant detail in the training dataset,
peer-to-peer database that can be used to store information
impairing their ability to apply the learned model to new
in a transparent and tamper-proof manner. It acts as a
data. (When a machine-learning algorithm learns for too
ledger that houses historical data in the form of blocks.
short a time, this is called underfitting.)
Blockchain can also be used to execute trade among
parties through a protocol called smart contracts that Robot refers to a machine characterized by four
does not require a centralized authority to authenticate the main features: mobility, in order to function in human
trade. The technology has been used extensively within a environments like hospitals and factories; interactivity,
cryptocurrency framework but, recently, has also found an made possible by sensors and actuators; communication,
outlet in energy management-related areas. made possible by computer interfaces or voice recognition
and speech synthesis systems; and autonomy, in the sense
Data mining is a process by which anomalies, patterns and
of an ability to ‘think’ for themselves and make their own
correlations are identified within large data sets to predict
decisions to act upon their environment, without direct
outcomes. Using a broad range of techniques, you can use
external control. Robots perform tasks through two types of
this information to increase revenue, cut costs, reduce risks,
algorithm: deterministic algorithms and AI (or stochastic)
etc. (sas.com).
algorithms. A deterministic robot’s behaviour is basically
Deep learning is a type of machine learning that trains a preprogrammed and essentially determined. However,
computer to perform tasks such as recognizing speech, AI‑based, cognitive robots will learn from past experiences
identifying images or making predictions. Instead of and calibrate their algorithms themselves, so their
organizing data to run through predefined equations, deep behaviour will not be perfectly predictable (UNESCO, 2017).
ARTIFICIAL
INTELLIGENCE
FOR SUSTAINABLE
DEVELOPMENT:
In 2018, UNESCO’s Natural Sciences Sector undertook

a foresight exercise, in order to identify emerging
opportunities and challenges arising from the rapid
development of artificial intelligence and evaluate the
implications of this silent revolution for its programmes.
As part of this exercise, the Natural Sciences Sector
surveyed staff at headquarters and in the field, as well
as UNESCO chairholders and centres. Respondents
were asked to identify opportunities and challenges
for UNESCO’s work in the following areas: scientific
research and experimental development; ecosystem and
environmental management; and science governance and
foresight.
The current paper summarizes the findings of
this exercise.
For more information:
[email protected]

2019 UNESCO AI SustDev

Uploaded by

Copyright:

Available Formats

2019 UNESCO AI SustDev

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

2019 UNESCO AI SustDev

Uploaded by

Copyright:

Available Formats

Principles for

INTRODUCTION4 3.4.3. AI can improve climate change assessments 18

4.1. Ensuring quality data and data management 22

5.1. Smart solutions are already here 

5.7. The need for skills to keep pace

THE IMPLICATIONS OF AI FOR POLICY-

6.1. A growing number of national AI strategies 48

6.2. Existing policy tools can frame AI strategies  48

6.3. AI systems can contribute to policy-making  49

6.4. Regional economic communities can help

6.5. Global strategies will also be necessary to

We are living through what has been termed the Fourth

Since the second half of the 20th century, biotechnology,

Digital technologies include big data and analytics, robots

Progress in artificial intelligence was initially slow, hampered

UNESCO’s own International Hydrological Programme began

Given the speed with which artificial intelligence is evolving

Developing countries will only be able to reap the benefits of

The current paper sets out to identify emerging opportunities

iStock/Getty Images Plus

2. The use of AI in scientific research and

power observed that these techniques have no way of saying ‘I don’t

Could AI systems be designed to ask the ‘right’ questions

2.2.2. Quantum computing may be the next

Quantum computing promises

The TQT provides a hub for the study of the future of AI on

In November 2018, a researcher from the ICTP’s Condensed

For the first three-year phase of the Flagship, ICTP’s Marcello

iStock/Getty Images Plus

3. The use of AI for environmental

Motion-sensor cameras can …

The main advantage of using AI 3.2.3. Enhancing the visitor’s experience

◼◼ Efficient, safe and sustainable Earth resources exploration

◼◼ Innovative renewable energy production and CO2 mitigation,

3.3.1. Machine-learning algorithms already in 3.3.2. The advantages of a three-pronged

The G-WADI PERSIANN-CCS GeoServer

iStock/Getty Images Plus

4. Fostering quality data for

4.1. Ensuring quality data and data Box 2.

Applications based on the Internet of Things exist in many

iStock/Getty Images Plus

5. Issues at the science–policy nexus

In July 2018, Yoshua Bengio, an expert in deep learning, told

Getty Images / Baranozdemir Delivery by drone

Policies encouraging China

transparency and sharing of 0 10 20 30 40

core datasets across both Other economies

public and private actors could Germany 3,2%

stimulate innovation-oriented France 2,2%

competition and enhance Netherlands 1,6%

research productivity Finland 1,2%

AI may already be widening the technological divide, even Belgium 0,1%

◼◼ African governments should invest in basic and applied

◼◼ African stakeholders should invest in the Internet of

According to the World Information Technology and Services

However, many buyers still attach importance to the reputation

Source: CSIS, 2010

5.5. What will greater automation

Five economies accounted for 74% of global sales in industrial

INTRODUCTION4 3.4.3. AI can improve climate change assessments 18

4.1. Ensuring quality data and data management 22

5.1. Smart solutions are already here

6.1. A growing number of national AI strategies 48

6.2. Existing policy tools can frame AI strategies 48

6.3. AI systems can contribute to policy-making 49

Quantum computing promises

Motion-sensor cameras can …

The main advantage of using AI 3.2.3. Enhancing the visitor’s experience

Policies encouraging China

Automation driven by AI will

The skills shortage is fostering Japan 2.9

Scientists have never been

Developing countries will need

In addition to teaching