www.ifc.

org/thoughtleadership NOTE 69 • JUL 2019

The Role of Artificial Intelligence in Supporting

Development in Emerging Markets
By Davide Strusani and Georges Vivien Houngbonon

Artificial Intelligence has enormous potential to augment human intelligence and to radically alter how
we access products and services, gather information, make products, and interact. In emerging markets,
AI offers an opportunity to lower costs and barriers to entry for businesses and deliver innovative business
models that can leapfrog traditional solutions and reach the underserved. With technology-based solutions
increasingly important to economic development in many nations, the goals of ending poverty and boosting
shared prosperity may become dependent on harnessing the power of AI. While emerging markets are
already using basic AI technologies to solve critical development challenges, much more can be done, and
private sector solutions will be critical to scaling new business models, developing new ways of delivering
services, and increasing local markets’ competitiveness. All of these solutions require innovative approaches
to expand opportunities and mitigate risks associated with this new technology.

Artificial Intelligence (AI) designates “the science and Basic AI imbeds cognitive abilities such as memory,
engineering of making machines intelligent, especially attention and language, as well as some executive
intelligent computer programs,”1 with intelligence defined functions like anticipation and decision-making with
by the AI100 Panel at Stanford University as “that quality limited reference to the past. It is typically used to
that enables an entity to function appropriately and with enhance the performance of business analytics solutions
foresight in its environment.”2 Other experts define AI as a and to improve the functioning of digital platforms.
computerized system that can think and act like humans. Examples include credit scoring, online matching,
More comprehensive definitions consider AI to mean chatbots, and smart speakers. In emerging markets,
all computer systems that can continuously scan their existing services include AI-enabled credit scoring in
environment, learn from it, and take action in response to Madagascar (M-Kajy), Kenya (M-Shwari), Egypt (ValU,
what they sense, as well as to human-defined objectives.3 Fawry Plus), and India (Aye Finance).
AI combines large volumes of data with computing power Advanced AI goes further in the simulation of human
to simulate human cognitive abilities such as reasoning, cognitive abilities such as perception, vision, and spatial
language, perception, vision, and spatial processing. Three processing. It closely mimics the human mind and enables
types of AI applications can be identified based on which the analysis of unstructured data such as texts, images, and
cognitive abilities are simulated and automated: audio data. Immediate application domains include facial

About the Authors

Davide Strusani ([email protected]) is a Principal Economist—Telecom, Media, Technology, Venture Capital and Funds, at IFC.
Davide has worked extensively for private sector organizations, industry associations, and governments on the role of technology,
digital services, and communications to deliver economic and social growth, and he has led numerous in-country campaigns
to focus governments and sector regulators on the benefits of ICT. Georges Vivien Houngbonon ([email protected]) is an
Economist—Telecom, Media, Technology at IFC. He has studied the impacts of digital technology in Africa for both academia and
the private sector.

1
This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.
and speech recognition, medical diagnoses, transportation, algorithms generally outperform rules-based approaches.
and urban planning, as well as logistics, security, and They can be classified into two broad categories according
safety. Outside of China, advanced AI applications are not to the number of stages involved in the learning process:
yet widely adopted in emerging markets. In China, Yitu
Basic learning algorithms involve only one stage of
Technology, a startup founded in 2012 and valued at $1
learning and are suitable for analyzing structured data
billion in 2017, provides facial recognition technology.
like price, quantity, or time; for predicting an outcome
Infervision uses AI to provide medical diagnosis solutions
given a set of inputs; or for clustering items according to
to more than 280 hospitals, 20 of which are outside China.
their characteristics. Examples include the prediction of a
In the transportation and automotive industry, NIO
customer’s churn, probability of default (credit scoring),
develops autonomous vehicle solutions.
and fraud detection in financial transactions. Deep learning
Autonomous AI is expected to become self-aware with the algorithms, on the contrary, involve several learning
ability to interact with human beings and learn on its own, stages organized similar to the structure of the brain.
thereby augmenting humans at home, outside the home, They are suitable for analyzing unstructured data such
or in work environments. Autonomous AI is still far from as images, audio recordings, or texts, and can be useful
widespread commercial use due to limitations in handling for face recognition, speech-to-text transcription, or text
irregular objects, biased decision-making involving value reconstitution. Unlike basic learning algorithms, deep
judgment, and an inability to learn from a few examples. learning algorithms radically open new avenues for data-
Prototypes from Fetch Robotics, Boston Dynamics, and driven decision-making, as few alternative methodologies
Hanson Robotics are early examples of autonomous AI. exist to process unstructured data.
AI performance has been enhanced by a new generation
Why is AI Gaining Prominence Now?
of algorithms labelled machine learning, or ML. Unlike
standard rules-based approaches, ML algorithms are Adoption of AI has significantly accelerated over the
automatically built from data, and the richer the dataset, last five years due the diffusion of digital technologies
the better they perform. Rules-based approaches tell and major breakthroughs in algorithmic capabilities,
the algorithm what to do in each state of the world. As access to richer data, and increasing computing power.
a result, they are limited to predictable outcomes and According to a Gartner global survey, 14 percent of large
perform poorly out of sample. ML, by contrast, involves companies used AI in 2019, up from 3 percent in 2018,
the use of algorithms to parse data, learn from it, and and that is expected to increase to 23 percent in 2020.
make a determination or prediction as a result. By learning The most popular AI applications include chatbots,
patterns from the data—including unpredictable ones—ML process optimization, and fraud analysis on transactional

Box 1: Differences between AI and other disruptive technologies

Standard inference and data mining techniques Cloud Computing, the practice of using a network
differ greatly from AI and machine learning. For of remote servers hosted on the Internet to store,
instance, to predict whether a customer is likely to manage, and process data, rather than a local server
churn, a statistical approach would assemble data on or a personal computer, is often used to implement
the characteristics of all customers, including whether AI, but it remains distinct. Cloud computing can be
they have churn or not, specify a statistical relationship used for non-AI related purposes such as data sharing
between customers’ statuses and their characteristics, and securing. Likewise, Blockchain, a decentralized
estimate parameters, and use this relationship to infer data sharing architecture, can also be used to store
the probability that a customer will churn given his or and process data, but does not include an automated
her characteristics. An ML approach would first train and data-driven learning process like AI. In addition,
an algorithm linking customers’ churning statuses while several digital platforms imbed AI technology
with their characteristics on a subset of data (training into their functioning, others do not. Digital platforms
dataset), use another subset of data to validate the that match users from different sides of the platform,
construction of the algorithm (validation dataset), recommend contents, or use targeted advertising
further test the predicting power of the final algorithm typically use an AI technology. However, peer-to-peer
on another subset of data (testing dataset), all before platforms or job boards, for instance, do not use AI,
predicting the probability of churn. Deep learning although they could eventually upgrade.
algorithms involve several iterations of this process after
partitioning the unstructured dataset.

2
This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.
 Opportunities Risks

• New products and business models—including leapfrogging • Obsolescence of traditional export-led path to economic growth
solutions, solutions for bottom of pyramid individuals, and • Increased digital and technological divide
easier access to credit
• Transformation of job requirements and disruption of traditional
• Automation of core business processes—leading to lower job functions
product costs
• Privacy, security, and public trust
• Human capital development
• Innovation in government services

FIGURE 1 Selected development opportunities and risks from AI in emerging markets

Source: IFC

data. Emerging applications include consumer and market Furthermore, the wide range of sectors that AI can
segmentation, computer-assisted diagnostics, call center transform has attracted significant venture capital
virtual assistants, sentiment analysis and opinion mining, investment. AI-related VC deals rose from 150 in 2012 to
face detection and recognition, and human resources 698 in 2017, with 90 percent annual growth in the volume
applications such as resume screening. These applications of investment, from just $0.6 billion in 2012 to $14 billion
are most common in insurance, software and IT services, in 2017.5
telecom, and retail. AI companies tend to have higher
valuations and raised more investment rounds than Reducing Poverty and Boosting Shared
equivalent non-AI startups since 2015.4 Prosperity
Companies and users are now creating an unprecedented Traditional pathways to a country’s economic development
amount of data. In 2017, the amount of digital data are increasingly subject to technology-based disruptions.
created was more than eight times that of 2009. Progress AI is highly disruptive in that it can result in a step
in telecommunications networks, the ongoing deployment change in the cost of or access to products or services,
of the Internet of Things, and the upcoming large-scale or can dramatically change how we gather information,
deployment of 5G networks will all enable even greater make products, or interact. As development challenges
data generation. In 2022, Internet data traffic is projected become more and more intertwined with technology-
to be three times that of 2017, and the share of licensed based disruptions, the twin goals of ending poverty and
IoT devices is predicted to rise from 13 percent in 2018 to boosting shared prosperity become critically dependent on
28 percent in 2025. Because AI needs data to learn, these harnessing the power of technologies such as AI, while at
trends are set to supercharge the development of more the same time seeking to limit the associated risks.6
powerful AI technologies.
Emerging markets, including some of the world’s poorest
Vast improvements in computing power and capacity countries, are already using basic AI to solve critical
to store data have supported this growth in data. For development challenges, particularly in the provision of
example, Google recently introduced its Tensor Processing financial services to unserved and underserved populations.
Unit, which has a processing power of 15 to 30 times that Early progress in basic machine learning algorithms,
of a Graphical Processing Unit, a key computer element combined with the limited burden of legacy technologies
that was central to the implementation of deep learning and a growing mass of technology users, have enabled
algorithms. emerging markets to implement basic AI solutions such as
credit scoring and targeted advertising. Ant Financial in
The rising importance of AI is also driven by demand-
East Asia, M-Shwari in East Africa, M-Kajy in Madagascar,
side effects such as the deployment of digital platforms
and MoMo Kash in Cote d’Ivoire are early examples of AI
and the emergence of other disruptive applications such
delivering financial services to the poorest. M-Shwari uses
as blockchain and cloud computing. The digital platform
machine learning to predict the probability of default of
business model hinges on the successful targeting of users
potential borrowers, which allowed it to deliver small loans
for marketing purposes and the provision of personalized
to 21 million Kenyans by the end of 2017.
Internet content to catalyze usage. These functions are
essential to the ability of digital platforms to reach the AI applications have the potential to address challenges
minimum number of users necessary for profitability. Thus, faced by individuals at the bottom of the income
online platforms like Google and Amazon rely on AI to distribution, particularly the bottom 40 percent.
attract both users and advertisers. While these individuals lack the means to purchase AI-

3
This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.
technologies or AI-enabled equipment, they can benefit used by many companies in developed markets. AI-enabled
from AI-as-a-service solutions through their mobile devices. productivity growth directly raises output and employment,
Recent examples include a machine learning app, Nuru, and also does so indirectly through increased consumption.
that has been used on farms in Kenya, Mozambique,
Cost reductions stemming from automation of certain
and Tanzania to identify leaf damage in photos taken by
functions can combine with increased access to credit—a
farmers and to send information to authorities to help
critical advantage AI technologies are already delivering—
monitor the presence of an invasive pest that threatens farm
to reduce overall business costs. This can increase both
revenue and food security across East Africa.
the volume of bankable business opportunities and the
Data generated through mobile phones can be highly level of competition within markets and industries. AI
correlated with financial status, educational attainment, solutions can also help overcome the lack of infrastructure
and health status and therefore can enable mobile AI apps and large information asymmetries in emerging markets
to deliver microlending, personalized tutoring, health by supporting product innovation in the form of new
diagnoses, and medication advice. In addition, AI’s speech business models and leapfrogging solutions tailored to serve
recognition and speech-to-text functionalities remove previously unserved and underserved populations.
literacy barriers typically encountered by the poorest
AI has the potential to deliver significant productivity
individuals when accessing text-based applications. And
gains for businesses. This includes the combination
image recognition can be used to assess microinsurance
of the accelerated pace of technology diffusion, the
claims of farmers in distant rural communities.
convergence of multiple technologies, and the emergence
AI technologies can enable new approaches to the of digital platforms. Through automation, AI is set to
monitoring and evaluation of development interventions bring significant cost reductions across all core business
to target those most in need.7 Emerging countries functions, including human resources management,
often lack the data necessary to fine-tune development marketing, accounting, and inventory. For instance,
interventions. AI’s ability to tackle unstructured data such employee recruitment often involves the costly review
as text, images, and audio can be useful for extracting the of dozens of candidate profiles, a process that can
information needed to improve development outcomes. be automated using AI solutions. Automation of the
For instance, an experiment in rural India relies on textual recruitment process typically curbs time-to-hire from 10
transcription of village assemblies to identify the topics weeks to 2 weeks, and the time to shortlist candidates from
discussed and how the flow of conversation varies with the 2–3 weeks to almost instantaneous. Also, repetitive human
gender and status of the speaker, thereby shedding light on review of accounting documents or inventory can instead
the functioning of these deliberative bodies, an important be performed by AI, generating significant cost savings.
aspect of political accountability.8 Other experiments For instance, automation of accounting services in Brazil
include the use of machine learning on VAT tax data in is set to alleviate the cost of bureaucracy—tax filing is an
India to better target firms for audits and to predict travel example—incurred by medium-sized enterprises. These
demand patterns after hurricanes9 as well as where food improvements are likely to drive the growth of informal
insecurity will occur to help target aid interventions.10 businesses, which make up to two-thirds of GDP in some
lower-income countries.11
Despite the potential risks of AI, failing to take advantage
of the opportunities it offers could be even more costly. Productivity improvements also stem from more efficient
The economic and societal transformations brought about investment in human capital thanks to automation. AI
by disruptive technologies can be accelerated with AI and can reshape high-quality education and learning through
can dramatically speed up progress toward the Sustainable precisely targeted and individually customized human
Development Goals and the twin goals. Yet if countries capital investments. The integration of online courses with
cannot compete in the future global economy, they will AI offers the opportunity to improve access to affordable
be left behind. To harness the potential of new business education and raise learning and employment in emerging
models, new ways of delivering services, and shifting markets. Edtech companies like Coursera, Andela, and
sources of competitiveness, countries and companies in the Udemy are generating data on student performance across
private sector will need to implement innovative approaches emerging markets and are poised to leverage this data to
to expand AI’s opportunities and mitigate its risks. deliver upskilling recommendations. In India, UpGrad
has enrolled 2,000 students in entrepreneurship, digital
Identifying the Development Opportunities of AI marketing, data analytics, and product management
courses, whereas Edutel uses two-way satellite technology
AI can expand and increase development opportunities
to deliver live lessons by specialist teachers in science, math,
in emerging markets in several ways. Improved business
and English to about 2,000 primary and secondary schools
productivity stemming from automation of core business
in the southern state of Karnataka. Other companies are
processes and human capital development can significantly
combining data from online education and job platforms to
lower production costs. These improvements are already
deliver automatic upskilling recommendations with the goal

4
This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.
 AI Benefits Economic Drivers Economic Impact

Human capital development

Productivity growth
Automation of core business Increased output
processes
Lower barriers to entry for
businesses
Increased consumption due
Easier access to credit to income growth

Market expansion

New products and business

models—including
leapfrogging solutions

FIGURE 2 Channels to economic development supported by AI technologies

Source: IFC

of improving employability. Examples include 51job.com in AI is also driving innovations in business models through
China and Revelo in Brazil. automation that delivers more affordable services, thereby
expanding the market to underserved consumers. For
AI is driving innovation in financial services through
instance, TaxiJet, a ride-hailing company with a business
better data processing, increasing access to credit. By
model similar to Uber, uses AI to match users with taxi
relying on nontraditional data such as mobile phone call
drivers in Cote d’Ivoire at lower cost than traditional taxis.
records, mobile-money transaction data, text messages,
and address books, AI can reduce information asymmetry AI can also alleviate constraints from poor infrastructure
in contexts where borrowers lack credit history, enabling in emerging markets by providing alternatives and cost-
access to financial services for first-time borrowers and the effective solutions to deliver social services to those who
unbanked. AI has the potential to make financial services need them most, including remote communities. Taking
more affordable thanks to the automation of credit scoring, advantage of the widespread coverage of mobile networks,
a process that requires human resources in traditional AI is being used in telemedicine for early diagnosis of
financial institutions. Machine learning algorithms can disease, thereby saving costs associated with maintaining
parse large amount of mobile phone data to deliver an extensive network of community health workers.
instantaneous credit scores to users in developing countries. Likewise, education resources planning often fails to
Once a user is offered a loan, the scoring algorithm account for the geographical distribution of learning
continues to improve by absorbing credit history data. An outcomes due to poor data, which leads to an unequal
example of this approach is Branch, a fintech company that distribution of resources. Automated processing of student
offers microloans to first-time borrowers and customers performance can help target areas with the greatest
without banking accounts in Africa (Kenya and Nigeria), challenges. AI-enabled matching of students with teachers
India, and Mexico. can also improve access to higher quality education.
AI’s capacity to handle unstructured data has the Productivity growth, lower barriers to entry, and
potential to enable product innovation in sectors such as market creation and expansion have the potential to
pharmaceuticals, transportation, and logistics. Because of raise consumption and ultimately output. Creating
limited statistical capacity, emerging markets often lack and expanding markets can help create jobs and raise
the structured data needed to power business analytics consumption to the benefit of the wider economy. For
solutions. AI, however, can handle unstructured data such instance, online marketplaces that rely on AI solutions
as audio recordings or videos largely available in emerging across Africa are expected to create around 3 million jobs
markets to discover new ways of servicing customers, by 2025 by expanding the supply of goods and services,
new drugs, and new predictive healthcare solutions. Drug making assets more productive, and unlocking demand
discovery, for instance, involves searching through an in remote locations.12 Productivity growth of informal
almost infinite set of molecular combinations, a task for businesses and market expansion are more critical in
which AI is far more efficient than medicinal chemists. emerging markets, suggesting a higher economic potential

5
This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.
for AI in these countries than in developed markets. China For instance, advanced AI applications such as natural
offers an example of the size of the opportunities AI holds language processing could replace outsourced customer care
for emerging markets: It is estimated that AI could boost services, an industry which employ thousands of workers
China’s GDP by 26 percent by 2030, compared with 14 in countries like Vietnam, South Africa, and Morocco.
percent in the United States.13 Likewise, autonomous AI, enabling robots to sew, could
replace hundreds of thousands of workers in Bangladesh and
These gains could be further reinforced by AI-driven
Ethiopia. Job displacement, accentuated by the decreasing
efficiencies in the delivery of public services. Governments
importance of cross-country labor cost arbitrage and
in emerging markets could benefit from AI due to
combined with slower output growth, could further widen
potentially significant cost-savings, improved delivery
the gap between advanced and emerging countries, as well as
of social services, and better risk management. While
increase inequality within countries, and thereby limit upside
few studies have investigated the gains from automation
opportunities for nascent middle classes. This risk is more
for governments in emerging economies, estimates from
acute in the medium term for countries that have already
advanced economies suggest that it could be substantial.
developed these types of jobs, while for the poorest countries
In the United States, estimates suggest that the federal
these jobs may not be there to lose.
government could save up to $41 billion through AI-enabled
automation.14 Potential government services that could be AI may challenge existing local businesses that fail to
automated include data entry with automated handwriting catch up with the latest technologies. A key element of AI
recognition, planning and optimization algorithms, and performance is access to large volumes of data, and this
customer service using speech recognition and natural tends to increase the initial advantage of a successful first
language processing. For instance, electronic document mover. Such a trend has the potential to create ‘winner
discovery locates 95 percent of relevant documents in the takes all’ outcomes. Successful AI-enabled businesses are
discovery phase of legal cases, compared to an average of 50 more competitive and therefore attract more customers
percent for humans, in a fraction of the time.15 and accumulate more data, which further improves their
AI algorithm and reinforces their initial competitive
Other opportunities include risk management—disease
advantage. This is often the case with mobile operators
prevention, natural disaster management, humanitarian
that offer electronic financial services and employ AI
crisis management—and citizen engagement through
to optimize their distribution networks. If the enabling
automated and real-time analysis of online activities,
environment to be competitive does not adapt, firms
including social network and telecommunications metadata.
will not be able to pursue new opportunities, widening
productivity differences, creating larger first-mover
Managing the Risks that AI Poses
advantages, and fostering growth accelerations only in
Disruptive technologies including AI pose new risks to certain sectors and locations.
economic and societal inclusion. Technologies are reshaping
Acute societal challenges include privacy, security, public
the nature of work and could exacerbate inequality within
trust, algorithmic biases, and ethical use of AI. Survey
countries.16 Shifts in the demand for labor and the skills
results suggest that these issues are more severe in emerging
that complement technology can reward those with access
markets. For instance, only 20 to 21 percent of businesses
to new technologies and skills—at the expense of those
leaders in the Asia-Pacific and Europe, the Middle East and
who lack them. With advances in AI and ML, highly-
Africa regions provide their boards with adequate reporting
skilled routine tasks may be disrupted. There will thus be a
metrics for cyber and privacy risk management, compared
premium on skills that complement technology—not only
to 35 percent in North America.17 Because they are often
technical skills but also socio-behavioral and creative skills
trained on imperfect data, AI applications in real world
for greater adaptability and lifelong learning.
settings like job screening, insurance approval, and policing
One concern is whether AI will disrupt the potential tend to reproduce social biases typically related to gender
for emerging economies to catch up through traditional and race. Policy makers are taking steps to mitigate these
export-led manufacturing. Emerging countries have risks, with member states of the OECD and some emerging
been able to take advantage of an abundant skilled but countries like Brazil and Peru having recently endorsed a
low-wage labor to attract foreign manufacturing firms set of principles to promote responsible stewardship of AI.
and outsourced services and gain global competitiveness
On top of these challenges, there are risks that AI may
in export-oriented sectors. Countries like China today,
widen gaps between countries, reinforcing the current
or South Korea and Japan yesterday, have succeeded by
digital divide.18 AI leaders (mostly in developed countries)
relying on this model. However, AI, more than other
could increase their lead in AI adoption over developing
disruptive technologies, embeds the cognitive abilities
countries. Many developed countries may have no
mobilized by this labor force, which might make it more
choice but to develop local AI industry to capture higher
difficult for emerging countries to exploit this important
productivity growth as their GDP growth momentum
traditional development ladder.

6
This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.
slows, a phenomenon often related to aging populations. In terms of entrepreneurial ecosystems, few emerging markets
Moreover, wage rates in these economies are high, have AI startups. As of 2018, 20 countries hosted 95 percent
providing more incentive than in low-wage developing of worldwide AI enterprises, only three of them are emerging
countries to substitute labor with machines. markets. China is second with 1011 AI enterprises, India is
5th with 152, and Russia is 20th with 17.21 A lack of access
Developing countries tend to have other ways to improve
to expertise and data often discourage private investors
their productivity, including catching up with best practices
from pursuing AI projects in emerging markets. Scarce
and restructuring their industries, and may therefore have
AI expertise in low-income countries increases the cost of
less incentive to develop local AI industry. This does not
implementing any AI project. Recent initiatives tackling this
mean that developed economies will reap the biggest gains
issue in Africa include Andela (Nigeria, Kenya, Rwanda, and
from AI and that developing economies are destined to
Uganda), a Google AI Lab in Accra (Ghana) and the creation
lose the AI race. Countries can choose to strengthen their
of a Master in Machine Intelligence at the African Institute of
digital economy foundations and develop the supporting
Mathematical Science in Kigali (Rwanda).
capabilities needed to reap the potential of AI. While
China has become the second largest AI powerhouse, paths
Looking Forward
remain open to other economies, and support from the
private sector will be critical to accelerating adoption and Development Finance Institutions including IFC are
dissemination. pursuing a broad range of strategies to help private
companies and governments across emerging markets
Supporting Private Sector-led AI Solutions in EMs implement AI solutions. 22 VC investment as well as
investment in funds is enabling the growth of AI startups.
The private sector is well positioned to harness the
Further investment in online educational platforms that
opportunities AI offers in emerging markets because
offer machine learning and programming courses are
of the significant need for innovation and the potential
building local expertise across emerging markets. However,
gains in productivity, market expansion, and business
private sector investment in AI projects in emerging
opportunities in the public sector. However, aside from
markets remains limited to a few countries, partly due to
China, private sector involvement in the diffusion of
uncertainty about consumer interest in AI products.
AI in emerging markets has been limited so far. India,
the second largest emerging market nation, remains a The private sector alone cannot make AI succeed
significant laggard, with just 152 AI startups as of 2018, in emerging markets. Governments must level the
compared with 1011 in China.19 playing field by providing open access to big data; by
catalyzing network effects through standards setting and
Most private sector initiatives are focused on microlending
interoperability enforcement; and by supporting trial-
and use machine learning algorithms in conjunction with
and-error phases, including potentially through public
mobile phone data to predict the probability of default by
subsidies to AI incubators.
potential borrowers, with fintech companies and mobile
operators leading the race. AI applications in promising
ACKNOWLEDGEMENTS
sectors such as transportation, education, health, and
agribusiness are rarely available in most emerging markets. The authors would like to thank the following colleagues
for their review and suggestions: Simon Andrews, Chief
Critical constraints to the adoption of AI solutions include
Investment Officer, Telecom, Media, Technology, Venture
the lack of a developed digital economy and a supporting
Capital and Funds, IFC; Ed Hsu, Senior Advisor, Global
entrepreneurial ecosystem capable of driving innovation
Practice Infrastructure, World Bank; Jennifer Keller, Lead
and attracting financing; a scarcity of local AI expertise;
Economist, Global Practice Macro, Trade and Investment,
and a lack of government support in key areas such as
World Bank, currently on assignment to Thought Leadership,
open access to data, system interoperability, trust, and
Economics and Private Sector Development, IFC; Gordon
acceptance of trial and error.
Myers, Chief Counsel, Legal Department, IFC; Kiril Nejkov,
While basic AI applications such as credit scoring and Counsel, IFC Compliance Risk, IFC; Kevin Matthees, Research
online platforms (mobile-based or fixed) can rely on Assistant; Thought Leadership, Economics and Private
traditional connectivity like 2G, advanced AI applications Sector Development, IFC; and Thomas Rehermann, Senior
such as facial and speech recognition require broadband Economist, Thought Leadership, Economics and Private
connection to transmit bandwidth-consuming files such Sector Development, IFC.
as images and audio. Likewise, the smooth functioning of
Please see the following additional reports and EM
autonomous vehicles requires a web of connected objects
Compass Notes about technology and its role in
and a network architecture closer to 5G. 20 Data centers
emerging markets: Reinventing Business Through Disruptive
are critical infrastructure for data storage and high-speed
Technologies—Sector Trends and Investment Opportunities
computation and parallel computing, yet they remain
for Firms in Emerging Markets (March 2019); Blockchain:
deficient in many emerging markets, particularly in Africa.

7
This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.
Opportunities for Private Enterprises in Emerging Markets Could Increase Access to Financial Services in Emerging Markets
(January 2019); How Technology Creates Markets—Trends and (Note 59); Modelo Peru: A Mobile Money Platform Offering
Examples for Investors in Emerging Markets (March 2018); Basic Interoperability Towards Financial Inclusion (Note 54); Precision
Business Models for Banks Providing Digital Financial Services in Farming Enables Climate-Smart Agribusiness (Note 46); Digital
Africa (Note 68); The Case for Responsible Investing in Digital Financial Services: Challenges and Opportunities for Emerging
Financial Services (Note 67); Natural Gas and the Clean Energy Market Banks (Note 42); How Fintech is Reaching the Poor in
Transition (Note 65); How a Know-Your-Customer Utility Africa and Asia: A Start-Up Perspective (Note 34).

1
Definition attributed to Professor John McCarthy in 1956. Quoted in Moor, James. 2006. “The Dartmouth College Artificial Intelligence Conference: The
Next Fifty Years.” AI Magazine, Volume 27 Number 4, pp 87–91. https://pdfs.semanticscholar.org/d486/9863b5da0fa4ff5707fa972c6e1dc92474f6.pdf.
2
See One Hundred Year Study on Artificial Intelligence (AI100). “Defining AI”. Stanford University. https://ai100.stanford.edu/2016-report/section-i-
what-artificial-intelligence/defining-ai
3
Broader definitions of artificial intelligence are drawn from the following sources: Nilsson, Nils. 2010. “The Quest for Artificial Intelligence: A History
of Ideas and Achievements.” Cambridge University Press; Russell, Stuart and Peter Norvig. 2009. “Artificial Intelligence: A Modern Approach.” Pearson
3rd Edition; OECD. 2019. OECD. 2019. “Recommendation of the Council on Artificial Intelligence.” OECD Legal 0449 as adopted on May 21, 2019.
https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449; PwC. 2018. “The Macroeconomic Impact of Artificial Intelligence.” February
2018. https://www.pwc.co.uk/economic-services/assets/macroeconomic-impact-of-ai-technical-report-feb-18.pdf.
4
MMC Ventures. 2019. “The State of AI 2019: Divergence.” MMC Ventures Report in Partnership with Barclays. https://www.stateofai2019.com/.
5
CB Insights, 2019. “Artificial Intelligence Deals Tracker.” https://app.cbinsights.com/research/artificial-intelligence-deals-tracker-heatmap/
6
For examples about the human development opportunities that AI offers, see the following World Bank Blog: Osornprasop, Sutayut. 2018. “Artificial
Intelligence, Big Data: Opportunities for Enhancing Human Development in Thailand and Beyond.” World Bank Blogs, December 18, 2018. https://
blogs.worldbank.org/eastasiapacific/artificial-intelligence-big-data-opportunities-enhancing-human-development-thailand-and-beyond
7
Blumenstock, Joshua Evan, 2016. “Fighting Poverty with Data: Machine Learning Algorithms Measure and Target Poverty.” Science Magazine, Volume
353, Issue 6301, pp. 753–754. http://www.jblumenstock.com/files/papers/jblumenstock_2016_science.pdf.
8
Parthasarathy, Ramya, Vijayendra Rao and Nethra Palaniswamy. 2017. “Deliberative Inequality: A Text-as-Data Study of Tamil Nadu’s Village
Assemblies.” World Bank Group Policy Research, Working Paper WPS 8119. http://documents.worldbank.org/curated/en/582551498568606865/
Deliberative-inequality-a-text-as-data-study-of-Tamil-Nadus-village-assemblies.
9
Farley, Scott. 2017. “Tracing Hurricane Maria’s Wake.” Blog, October 25, 2017. https://blog.mapbox.com/tracing-hurricane-marias-wake-
7217a7d08380.
10
Knippenberg, Erwin, Nathan Jensen and Mark Constas. 2018. “Resilience, Shocks, and the Dynamics of Food Insecurity: Evidence from Malawi.”
Working Paper, March 2018. https://static1.squarespace.com/static/59c807a890bade1b5925f72e/t/5a9ef2b253450a57bf5f6a7c/1520366270740/
Resilience%2C+Shocks+and+the+Dynamics+of+Food+Insecurity+%28JMP%29.pdf
11
Leandro, Medina and Friedrich Schneider. 2018. “Shadow Economies Around the World: What Did We Learn Over the Last 20 years?” IMF Working
Paper WP/18/17. https://www.imf.org/en/Publications/WP/Issues/2018/01/25/Shadow-Economies-Around-the-World-What-Did-We-Learn-Over-the-
Last-20-Years-45583.
12
Boston Consulting Group (BCG). 2019. “How Online Marketplaces Can Power Employment in Africa.” https://www.bcg.com/publications/2019/how-
online-marketplaces-can-power-employment-africa.aspx.
13
PwC. 2017. “Sizing the Prize: What’s the Real Value of AI for Your Business and How Can You Capitalize?” https://www.pwc.com/gx/en/issues/
analytics/assets/pwc-ai-analysis-sizing-the-prize-report.pdf.
14
Eggers, William, David Schatsky and Peter Viechnicki. 2017. “AI-Augmented Government: Using Cognitive Technologies to Redesign Public Sector
Work.” Deloitte Center for Government Insights. https://www2.deloitte.com/content/dam/insights/us/articles/3832_AI-augmented-government/DUP_
AI-augmented-government.pdf.
15
Kershaw, Anne. 2005. “Automated Document Review Proves its Reliability.” Digital Discovery & e-Evidence, Talking Tech Note Volume 5, Number
11, November 2005, pp 1–4. www.akershaw.com/Documents/2004AEKDocReviewArticle.pdf.
16
World Bank Group. 2019. “World Development Report 2019: The Changing Nature of Work.” http://documents.worldbank.org/curated/
en/816281518818814423/pdf/2019-WDR-Report.pdf.
17
PwC. 2018. “Digital Trust Insights.” PwC 2018 Digital Trust Insights Survey. https://www.pwc.com/us/en/services/consulting/cybersecurity/digital-
trust/2018-insights.html
18
World Bank Group. 2016. “World Development Report 2016: Digital Dividends.” https://openknowledge.worldbank.org/bitstream/
handle/10986/23347/9781464806711.pdf
19
China Institute for Science and Technology Policy at Tsinghua University (CISTP). 2018. “China AI Development Report 2018.” CISTP, July 2018.
http://www.sppm.tsinghua.edu.cn/eWebEditor/UploadFile/China_AI_development_report_2018.pdf
20
Forecast of 12% 5G penetration rate in developing countries by 2025, compared to 48% in developed countries according to projections by GSMA
Intelligence. GSMA Intelligence, 2018. “Market Penetration: 5G.” https://www.gsmaintelligence.com/metrics/44/1529/data/?report=5d26d871dd310
21
China Institute for Science and Technology Policy at Tsinghua University (CISTP). 2018. “China AI Development Report 2018.” CISTP, July 2018.
22
These strategies are complemented by investment in digital infrastructure such as Helios Towers Africa and a 10,000 km sub-marine cable system in
East Africa. See Chapter 6 of “Reinventing Business Through Disruptive - Sector Trends and Investment Opportunities for Firms in Emerging Markets.”
IFC Report, March 2019. https://www.ifc.org/wps/wcm/connect/8c67719a-2816-4694-9187-7de2ef5075bc/Reinventing-business-through-Disruptive-
Tech-v1.pdf?MOD=AJPERES

8
This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.