www.ifc.org/thoughtleadership NOTE 71 • SEPT 2019 Artificial Intelligence: Investment Trends and Selected Industry Uses By Xiaomin Mou The global race to fund, develop, and acquire artificial intelligence technologies and start-ups is intensifying, with commercial uses for AI proliferating in advanced and emerging economies alike. AI could increase GDP growth in both advanced countries and emerging markets. In energy, AI can optimize power transmission. In healthcare, diagnosis and drug discovery will benefit enormously from AI. In education it can improve learning environments and learning outcomes and can better prepare youth for transition to the workplace. In manufacturing, AI can help design better products in terms of functionality, quality, and cost, and improve predictive maintenance. AI can help extend credit and financial services to those who lack them. The potential impact of AI on transportation and logistics goes far beyond automation and road safety to span the entire logistics chain. Yet with the exceptions of China and India, emerging markets have received only a modest share of global investment in this advanced technology, despite the fact that they may benefit more from AI implementation than advanced economies. Artificial intelligence, or AI, has the potential to imitate the dynamically learn traffic patterns and create efficient human brain, which makes it unique among technologies routes; smartphones use AI to recognize faces and in that it can learn and solve problems that would normally verbal commands; AI enables efficient spam filters in require human intelligence. In general, AI includes natural email programs, smart assistants such as Alexa, and language and processing, visual perception and pattern recommendation engines. These are a small sample of recognition, and decision making. These processes in familiar technologies that leverage AI’s capabilities. AI combination give AI enormous potential in multiple applications can be found in virtually every industry today, disciplines and across many economic sectors.1 And they may from marketing to healthcare to finance. help address persistent development challenges such as a lack Of course, the development and implementation of AI is of infrastructure or underdeveloped healthcare or financial not without its share of controversy, and the debate about sectors, which can leave many individuals underserved. the risks and rewards of this unique and revolutionary Despite its revolutionary potential, AI—at least in its most technology tend toward extremes, with many observers basic form—has existed for decades. First-generation predicting that AI will destroy jobs and even eventually AI-equipped computers played chess, solved puzzles, and threaten humans. Some scenario analyses2 suggest a performed other relatively straightforward tasks. potentially positive impact of AI on GDP growth, but Yet the sophistication level of AI has evolved dramatically virtually all are focused on developed economies. In general, in recent decades, and the technology is now prevalent the aggregate impact is predicted to hinge on several factors in many areas of everyday life. Google Maps uses AI to including skills, availability of open source data, and About the Author Xiaomin Mou, Senior Investment Officer, Private Equity Funds—Disruptive Technologies and Funds, IFC. Her email is xmou@ifc.org. 1 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. technological progress, with some countries expected to cards or traditional mortgages. In China, companies like gain more than others. The impact on jobs is much more AntFinancial and Tencent have credit scoring solutions that uncertain, as it depends on the particular economic sector leverage e-commerce data, as well as payment platforms and the skill composition of the labor force. that provide insight to credit-based decision making. These Controversial or not, the race to develop AI proceeds technologies, much like those in agribusiness and other apace. Because the distribution of venture capital (VC) sectors, have enormous potential to be applied to other investment into AI-specific technologies closely tracks the emerging markets such as Sub-Saharan Africa. flow of overall VC flows, the latter can be used as a proxy Machine learning, in which machines are inspired by for interest in AI by country. And it is clear from the data biological processes and learn from observation and that the United States and China lead in AI investment, with experience, is the most invested category of AI. The AI China dominating global AI funding. Chinese AI companies industry is moving toward consolidation, with large raised a total of $31.7 billion in the first half of 2018, corporates and industrial players making frequent almost 75 percent of the global total of $43.5 billion. China acquisitions of start-ups, a phenomenon that tends to drive looks poised to lead the AI space in several sectors including up valuations and limit opportunities for VC investors. healthcare and autonomous driving. China’s progress with AI is largely the result of strong and direct government ($ billion) India SE Asia (ex. Singapore) support for the technology, leadership from Chinese tech Brazil industry giants, and a robust venture capital community. Poland With the exceptions of China and India, emerging markets Turkey LatAm (ex Brazil) have received a modest share of global investment in CGCC & Levant advanced technologies. Total VC flows to emerging markets Sub Sah Africa between 2008–2017 excluding China and India was just CCE (ex Poland) $24 billion, compared with global flows to the United North Africa States over the same period of $694 billion (Figure 1). US ($ billion) 50.0 China India UK 2.4 Israel 2.8 Germany 185 Singapore 1.8 1.3 France 7.6 4.9 2.4 0.7 0.2 Rest of the World 50 31 FIGURE 2 EM VC Investment 2008–17 (ex-China) 19 15 Source: Pitchbook 2019 694 24 17 12 The Evolution of AI FIGURE 1 Global VC Investment 2008-17 Artificial intelligence can be categorized into three basic Source: Pitchbook 2019 stages of development. AI development in China has important implications for Basic AI or Artificial Narrow Intelligence (ANI) is other emerging markets, too. A microlending algorithm limited in scope and restricted to just one functional area. developed using the credit scoring of Chinese consumers AlphaGo, a computer program that plays the board game can be much more readily applied in another emerging Go, is an example. market than one developed using credit reports of Advanced AI or Artificial General Intelligence (AGI) is American consumers. That’s due to the fact that, unlike advanced and usually covers more than one field, such as borrowers in advanced economies, borrowers in China power of reasoning, abstract thinking, or problem solving and other emerging markets often do not have credit on par with human adults. 2 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. Autonomous AI or Artificial Super Intelligence (ASI) is the intelligence that allows for some degree of derived action final stage of intelligence expansion in which AI surpasses beyond explicit coding; and statistical analysis that human intelligence across all fields. This stage of AI is not mimics the results of human reasoning without having to expected to be fully developed for several decades. “understand” that reasoning. Natural language processing uses this latter approach with a departure from grammar The Rapid Growth of Data building to use statistical rules. Today, advances in other technologies are creating an Also, as AI becomes more widely adopted, its basic toolsets environment conducive to the rapid acceleration of AI and functionality will become available as commercial technology. Massive amounts of data that are being created services from large tech platforms. Examples include by increasingly ubiquitous connected devices, machines, and Amazon Machine Learning Services, Google DeepMind global systems—including the Internet of Things, or IoT—are and TensorFlow, IBM Watson, and Microsoft Cortana becoming increasingly helpful in training learning systems to Intelligence Suite, among others. Platform operators will make them more realistic and humanlike in their behaviors. offer an AI layer to add stickiness to existing offerings, and For example, electric vehicle carmaker Tesla aggregated with this horizontal toolset available, start-ups will be able some 780,000,000 miles by the close of 2016—a rate of to scale AI more quickly and cheaply. one million miles every ten minutes—through its connected cars. The data generated by those miles can be instrumental Funding Trends in AI for AI applications. As commercial uses for artificial intelligence proliferate, The more data available, the better the AI algorithms the race to acquire AI technologies and start-ups is become. In addition, significantly faster computers allow intensifying. Big corporations in every industry, from retail for much more rapid processing of the data. Lower-cost to agriculture, are attempting to integrate machine learning computing power, particularly through cloud services and into their products. new models of neural networks, have dramatically increased Perhaps as a result, machine learning leads AI technology the speed and power of AI. Graphic processing units (GPUs), investments. Machine learning, as opposed to learning repurposed to work on data, allow for faster training of according to rules and logic, occurs through observation machine learning systems compared with more traditional and experience. Instead of a programmer writing the central processing units (CPUs). While CPUs load and process commands to solve a problem, the program generates data sequentially, GPUs can “parallel” process data, which its own algorithm based on example data and a desired allows AI to manipulate vectors and matrices in parallel. By output. Essentially, the machine programs itself. repurposing these graphics chips, networks can iterate faster, As of January 2019, Venture Scanner, an emerging leading to more accurate training in shorter time periods. technology research firm, analyzed over 2000 AI start-ups GPUs can also replace expensive high-performance hardware. and classified them into 13 functional categories that The effect of these chips has been described as allowing collectively raised $48 billion in funding since 2011. Start- processing speeds to “jump ahead” seven years, relative to ups developing machine learning applications make up half what Moore’s Law would have allowed.3 of this funding. These companies utilize computer algorithms New Approaches to AI to automatically optimize some part of their operations. Examples include CustomerMatrix, Ayasdi, Drive.ai, and Beyond data generation and computing power, new Cylance. Many other AI categories include pioneers and approaches to artificial intelligence are driving the display enormous potential for growth and development. technology forward. The first such approach involves modelling the human brain. This includes physically Market intelligence firm CBInsights has identified 100 building an electronic model of the brain, as well as using of the most promising private companies applying AI logical approaches like neural networks that mimic the way algorithms across more than 25 industries, from healthcare neurons in the brain interact. to cyber security. These start-ups have collectively raised Alternative approaches involve sophisticated logical $11.7 billion across 367 deals. rules. These include logical programming to code human Perhaps due to this rapid growth in the AI space, there is reasoning into software; evolutionary computational now an acute shortage of AI talent in many workforces. 3 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. And this is accelerating the race to acquire early-stage AI China and AI companies with promising technologies and personnel. AI pushed total VC funds flowing to China to a record Notable acquisitions include Amazon’s purchase of AI $40 billion in 2017, up 15 percent from the previous year. cybersecurity start-up Sqrrl and Oracle’s acquisition of The Chinese government is active in promoting the AI cybersecurity firm Zenedge. While tech giants continue to industry and initiatives; its stated goal is to develop an AI hunt for AI technology and talent, traditional insurance, sector worth $150 billion by 2030. The Chinese private retail, and healthcare incumbents are also on the chase. sector is also active in the space. Internet firm Baidu has The largest deals in AI history include the 2018 Roche actively pursued an “AI first” agenda since launching the Holdings acquisition of New York-based Flatiron Health Institute for Deep Learning in 2013 and establishing the for $1.9 billion, and Ford Motor’s acquisition of auto tech Silicon Valley AI Lab the following year. In January 2018 start-up Argo AI for over $1 billion in 2017. Google is the the Beijing Frontier International AI Research Institute top acquirer of AI start-ups, with 14 acquisitions under was established under the leadership of Kai-Fu Lee of its belt. Sinovation Ventures. The growth of VC funding since 2012 has followed a There are also local AI initiatives in China with multiple similar path. In 2017 AI attracted $12 billion of investment cities—Beijing, Shanghai, Hangzhou, Zhejiang, and Tianjin from VC firms, which is double the volume of 2016, among them—developing plans and policies for AI. For according to KPMG. Around 42 percent of AI companies example, Shanghai plans to establish a special fund to acquired since 2013 had VC backing. invest in AI development, while Hangzhou has launched its AI Acquisitions and Funding are Scaling Rapidly own AI industrial park along with a fund that will invest According to ABI Research, AI start-ups in the United approximately $1.5 billion in it. States raised $4.4 billion from 155 investments, while PricewaterhouseCoopers predicts China’s GDP will reach Chinese start-ups raised $4.9 billion from 19 investments,4 $38 trillion by 2030, with $7 trillion of that coming from AI as they tend to focus more on mature AI applications. The through new business creation in fields such as autonomous most vibrant AI hubs worldwide are California’s Silicon driving and precision medicine, as well as existing business Valley, New York City, Beijing, Boston, London, and upgrades in terms of improved efficiencies and reduced Shenzhen. These hubs benefit not only from the creation costs. From 2010 through Q3 2017, a total of 704 AI deals of highly skilled and highly paid jobs, but also knowledge were made in China, representing $6.67 billion.5 and innovation spillovers. Employees at AI firms tend to become AI entrepreneurs, AI workers switch between AI B2B Services companies, and innovative AI products can be developed 2% 1% Lifestyle & for and deployed in local markets, exposing even more Consumption people to the technology. 3% 3% Transportation & Silicon Valley is the top global hub for start-ups (12,700 to 3% Automobile 5% 15,600 active start-ups) and tech workers (two million). It 32% Health is the global leader for VC investment and the headquarters Fintech 10% for many top technology firms. Education New York is the leading hub for the financial and media Security industries; it has an AI talent pipeline from universities; 11% Tools & Software and it has a strong funding ecosystem—the world’s second Manufacture 17% largest after Silicon Valley in terms of the absolute number 13% Logistics of early-stage investments. E-Commerce Beijing leads the volume of academic research output in AI, which comes from Tsinghua, Beihang, and Peking Universities; it has extensive involvement of tech leaders, FIGURE 3 China AI Investment by Subsector (# of deals) especially Baidu; and the Chinese government considers AI Source: ITJuzi to be of strategic importance. 4 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. Do the Rewards of AI Outweigh its Risks? The timing is particularly fortuitous, as the energy grid is New technologies come with risks, and there is much changing from constant baseload systems to intermittent uncertainty around advances in AI and machine learning, renewable generation, a shift that greatly increases system particularly with regard to the technology’s impact on complexity. For example, AI could be used to optimize society and the economy. AI’s potential to imitate human distributed energy resources such as rooftop solar behavior has given rise to concerns that the technology photovoltaic and batteries to match load and capacity. poses a significant threat to jobs, privacy, and the nature of Electricity meter data can be disaggregated with heuristics human society itself. machine learning, generating insights for additional energy savings. And renewable energy sales and deployment can Concerns about AI-driven job losses assume that humans be accelerated with AI. won’t be needed to manage and monitor AI machines and regulate inputs and outputs. Yet a study by the Economist AI can deliver increased energy efficiencies at the grid Intelligence Unit that looked at the manufacturing, healthcare, level by reducing standby reserves of thermal base load energy, and transportation sectors found that AI would boost generation by allowing the grid to follow load and GDP by 1 percent under all scenarios it examined, with even renewables more closely. This directly reduces the use of more significant gains in developing Asian nations.6 The EIU coal, oil, and gas, and thereby reduces greenhouse gases. study also projects that employment in the manufacturing Also, through its greater level of flexibility, AI can increase sector will remain relatively steady after AI technology renewables generation by lifting the ceiling on the amount penetration. The study does predict that certain job categories of renewables that can be accommodated. would be eliminated by AI, though there will be offsetting job At the building level, AI can increase efficiency by using creation among higher-skilled job categories. Still, job losses machine learning to predict building heating and cooling have historically been associated with the introduction of loads based on weather, time of day, weekday, etc. And AI revolutionary technologies, especially in manufacturing. can empower consumers through better disaggregation of electricity meter data, allowing for resource conservation Bias in AI through behavior modification. As AI technologies have emerged and spread, a phenomenon known as AI bias has been noticed. It occurs AI in Healthcare when an algorithm produces results that are prejudiced due There are many uses for AI technology in the healthcare to erroneous assumptions in the machine learning process. sector. These technologies are maturing rapidly and are And it can lead to and perpetuate biases in hiring, lending, already being used in a number of applications—from aiding and security, among other areas. Bias can creep in at many diagnosis to improving operational healthcare workflow stages of the learning process7 including (1) setting what efficiencies. The goal of many of these applications is to the model should achieve (potential predatory behavior to do what humans do but faster, more accurately, and more maximize profit); (2) collecting data that reflects prejudices reliably. That makes them potentially beneficial in resource- (selecting one gender over another, for example) or is not constrained environments with limited access to doctors and representative of reality; and (3) preparing the data and other health professionals, as well as in cost-containment selecting which attributes the algorithm should consider or constraints. Top uses include: ignore. Mitigating these biases can be challenging, but there 1. AI-enhanced medical imaging and diagnostics is is a strong movement within AI to do so, and researchers designed to improve the speed and reliability of analysis are working on algorithms that help detect and mitigate and can be particularly beneficial in contexts where hidden biases in training data and models, processes there is a lack of trained doctors, radiologists, etc. that hold the users of these models accountable for fairer 2. AI-triage plugs into tele-health platforms and provides a outcomes and defining fairness in different contexts. pre-consult triage, even flagging potential diagnosis, to AI in Energy save physicians time. AI’s potential in the energy industry mostly leverages the 3. Patient data and risk analytics. AI promises data technology’s ability to analyze highly complex systems analytics and machine learning on patient data such in real time and optimize them in ways not possible with as electronic health records, facilitating predictive conventional information technology. diagnostics, and ultimately improving outcomes. 5 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. Disclosed equity funding 2013–18 The use of AI in education can not only improve learning environments and learning outcomes, it can also save teachers and faculty time and allow them to focus on learners with special needs and can make curricula more relevant to the needs of employers and industry. It also has the potential to democratize education by providing quality teaching in non-traditional learning environments. And AI can give parents a greater role in their children’s education through new tools and platforms, and can decentralize FIGURE 4 AI Investments in Healthcare, Global education to reduce school, campus, and class sizes. Source: CBInsights All of these applications are useful not only for academia, but also in making on-the-job training programs more 4. Drug discovery. Deep learning techniques using efficient. AI applications also have the potential to better convolutional neural networks8 are very effective in prepare youth transition to the workplace through predicting which molecular structures could result in specialized work readiness programs, while helping working effective drugs. Applications are being developed by adults remain competitive in the workplace through both in-house research and development departments customized reskilling/upskilling offerings. Experts predict as well as by independent start-ups that are focused on great potential for AI in assessments, intelligent tutoring, the vertical systems and are expected to accelerate drug development of global classrooms, language learning, and discovery. AI also supports personalized medicine, or matchmaking between the demand for and supply of skills. the targeting of medicines based on individual genetics and other genomic analysis. AI in Manufacturing 5. Pharmaceutical supply chain. Using AI to process real- Manufacturing offers multiple opportunities for AI time data and make predictive recommendations is technologies, with innovations encompassing both expected to drive data-driven supply chains, improving hardware and software. The top uses are: efficiency and cost management. 1. Product and process engineering. This includes the AI can increase access to quality healthcare through AI- use of AI in CAD (Computer Aided Design) systems enabled triage, leveraging the time of scarce doctors and to design better products in terms of function, quality, facilitating diagnoses. It can deliver more affordable care or cost. This area is by far the most promising for the through increased productivity, allowing available healthcare manufacturing sector because of the scalability of CAD professionals to focus more closely on patient care and software solutions. Thus, Generative Design, which uses human-interaction. It can lower costs through better data a mix of large databases of designs and an input of the management and more efficient drug discovery mechanisms. critical parameters and functions of a given product, can automatically create a product optimized in its function, AI in Education cost, and manufacturability. Artificial intelligence technologies can dramatically 2. Intelligent CAD systems can also be interfaced with process enhance the way students learn both within and outside simulation tools to seek the best ways to manufacture a of the classroom, as well as help expand access, relevance, and efficiency of education, although the use of this given product (for instance, deciding between 3D-printing technology in the sector is still at a nascent stage of or traditional molding for plastic parts). As we have development. Machine learning can customize learning already seen with traditional CAD systems, where the content by providing teachers and faculty with actionable cost has fallen to 1 percent of what it was 20 years ago, insights from student performance to better understand such tools could quickly become affordable and therefore and serve student needs. AI can improve online tutoring, widespread—even in emerging markets. help teachers automate routine tasks such as grading, and 3. Production management. AI-enhanced predictive fill gaps in their curricula, and can give students immediate maintenance is aimed at improving asset productivity by feedback to help them better understand concepts at their using data to anticipate machine breakdowns, particularly own pace and with a greater degree of individualization. in cases when traditional statistical analysis tools have 6 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. already been fully deployed and costs and benefits justify 4. Visual identification and verification can be used to identify adding AI to them. customers and documents, streamlining processes such as In addition, collaborative and context-aware robots can account creation and loan and insurance origination. For recognize their environments, enabling them to alter example, Irisguard supports customer identification. their actions based on what is needed of them. And 5. Humanlike chatbots, similar to the popular Siri functions can be altered in real time. application, can intelligently interact with customers, answer questions, and reduce loads for customer service 4. Yield enhancements are a consequence of root cause departments. NextIt is an example of a chat bot provider. analysis of defective products and improved manufacturing processes in real time to boost output. AI can help in the 6. Using AI technology and data analytics to support cases when traditional statistical analysis has already consumer access to mortgage financing, especially for been fully deployed and if costs/benefits justify it. Some those who are informally employed and applicants with AI applications are being developed both in-house and weak documentation. Aavas, a specialized housing by start-ups focused on the industry whenever costs finance company in India, relies on data analytics and and benefits analysis can justify them. As with other AI AI tools to assess the creditworthiness and willingness of applications, access to large data sets and the involvement households with undocumented and documented incomes of data scientists who are also technical experts in the to repay loans received. specific application targeted are critical to successfully AI can significantly lower the cost of asset management, deploying AI in manufacturing. making it available to average investors and not just high It’s been proven over decades that Total Predictive net worth individuals. AI-enabled fraud detection can allow Maintenance (TPM) programs can significantly reduce banks to accurately predict if an account is at risk. And AI factory or assembly line downtime and maintenance costs. can help eliminate human error from compliance processes, Automated, sensor-based inspection of critical parameters a challenging area for many financial institutions. From coupled with Statistical Process Control (SPC) is also proven extending investment opportunities to the underbanked to reduce online rejects significantly. In specific cases, when and the average investor, to detecting fraud and mitigating scale is large enough and maintenance or quality issues investment risks, AI has the potential to improve the cannot be solved with traditional TPM or Total Quality financial health of people and institutions globally. Management (TQM) tools, AI tools could be considered. AI in Transport Similarly, collaborative and context-aware robots could improve productivity in specific cases. Autonomous vehicles tend to dominate the discussion of AI in transportation, but the effects of AI on transportation AI in Financial Services and logistics extend far beyond AVs and even roads. An AI is likely to have a game-changing impact in the financial entire spectrum of transportation modes is expected to go services industry in six major areas. driverless or crewless, including railways, ships, and various delivery vehicles, all of which are potentially viable in the 1. Gaining insights that can accurately predict customer short-to-medium term. behavior. An example is using AI to look at a potential borrower’s past behavior and accurately predict his AI technologies have enormous potential to address challenges or her creditworthiness. IBM Watson is just one of in transportation, particularly with regard to safety, reliability hundreds of applications here. and predictability, efficiency, and environmental issues such as pollution. AI can provide innovations in traffic management 2. Early detection and prevention of cybersecurity threats. for solutions to more effectively route cars and avoid accidents, Generative Adversarial Networks can generate real and crashes, and fatalities, as well assist law enforcement. Routes fake data sets and learn over time, increasing accuracy can be optimized to reduce traffic and increase reliability, of identification and verification. while optimal transit networks for communities can be 3. Supporting financial institutions in complying with KYC/ designed with smarter traffic signals and other transport AML regulations as AI can learn, remember, and comply infrastructure. Delivery routes for trucks and motorcycles with all applicable laws. This can significantly reduce in intra-city deliveries can be altered for quicker delivery operating costs in an increasingly complex regulatory world. times, while commute times can be reduced for individuals. 7 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group. All of these solutions impact pollution, as route optimization markets such as China to become adopters rather than just reduces fuel use and emissions for all types of transportation, developers of AI, with AI applications poised to proliferate including ships, trucks, and cars, among others in and have a significant impact on major economic sectors. The effects of AI on transportation and logistics go beyond These developments will capture significant gains across automation and road safety management to span the entire the value chain, with cost savings stemming from more logistics chain—from origin to final destination of cargo accurate demand forecasting and tailored and targeted and goods. AI can offer shippers faster delivery times and user experiences. Along with the promises of AI come with increased reliability at lower costs to get products sent by the challenge of AI bias, though a growing contingency of sea from factories to land distribution centers. AI can also researchers are devoted to mitigating this bias to ensure enable much more accurate predictions of arrival times for fairness in AI systems across the spectrum. container ships and can spot trends and risks in shipping ACKNOWLEDGMENTS lanes and ports. Machine learning can help analyze The author would like to thank the following colleagues historical shipping data by considering factors such as for their contributions, review and suggestions: Simon weather patterns and busy or slow shipping seasons, which Andrews, Senior Manager, Development Partner Relations, Partnerships, Communication & Outreach, IFC; Gordon can highlight inefficiencies, errors, and duplications. AI can Myers, Chief Counsel, Legal, IFC; Kiril Nejkov, Counsel, also help provide digital chartering marketplaces for the IFC Compliance Risk, IFC; Disruptive Technologies, bulk maritime industry (VesselBot already does this). Disruptive Technologies and Funds, IFC: Karthik G. Tiruvarur, Investment Officer, and Yuntong Shi, Investment AI technologies are also being used to mimic human Analyst; Telecom, Media, Technology, Venture Capital perception and cognitive abilities such as seeing, hearing, and Funds, Sector Economics and Development Impact, reading, and interpreting sensor data, and this has IFC: Davide Strusani, Principal Economist, and Georges Vivien Houngbonon, Economist; Stevon D. Darling, benefits for user interfaces aboard ships, including speech Associate Investment Officer, Office of the Chief Operating recognition programs that directly control equipment. Officer, IFC; Charlene Sullivan, Research Officer, Climate Strategy and Business Development, Climate Business, Looking Forward IFC; Transport, Infrastructure & Natural Resources, IFC: Ian Twinn, Manager, and Maria Lopez Conde, Research Recent breakthroughs in deep learning have produced AI Analyst; Global MAS – Health and Education, Manufacturing, systems that match or surpass human intelligence in certain Agribusiness & Services, IFC: Salah-Eddine Kandri, Global key functions and economic sectors. The United States and Lead Education, and Alejandro Caballero, Principal Education China are leading the race in AI. While the capital flowing Specialist; Monique Mrazek, Senior Investment Officer, Regional MAS – Latin America & the Caribbean, Health to AI start-ups is similar in the two countries, in China and Education; Emmanuel Pouliquen, Principal Industry the average dollar amount per investment is much higher, Specialist, Manufacturing - Global MAS; Margarete Biallas, reflecting the reality that in China, AI applications are the Digital Finance Practice Lead and Senior Operations Officer, Financial Inclusion, Financial Institutions Group, IFC; Julia main focus, rather than fundamental AI development. Heckmann, Research Analyst, Global Infrastructure – Energy, AI development will force societies to confront the Global Infrastructure & Natural Resources, IFC; Thought Leadership, Economics and Private Sector Development, possibility of job losses, yet studies suggest that AI will IFC: Baloko Makala, Consultant; Matt Benjamin, Editor; add to GDP, with emerging markets standing to benefit Felicity O’Neill, Research Assistant; Prajakta Diwan, Research even more than developed countries. We expect emerging Assistant; and Thomas Rehermann, Senior Economist. 1 See also Strusani, Davide and Georges Vivien Houngbonon. 2019. “The Role of Artificial Intelligence in Supporting Development in Emerging Markets.” EM Compass Note 69, IFC, July 2019. 2 Economist Intelligence Unit. 2017. “Risks and Rewards - Scenarios Around the Economic Impact of Machine Learning.” https://eiuperspectives. economist.com/sites/default/files/Risks_and_rewards_2018.2.7.pdf. 3 Moore’s Law is the general observation that the number of transistors in an integrated circuit doubles about every two years, allowing computing power to increase at an exponential pace. 4 ABI Research. “Artificial Intelligence Investment Monitor 2017.” https://www.abiresearch.com/marketresearch/product/1030415-artificial-intelligence- investmentmonitor/ 5 “Sizing the Price – PwC’s Global Artificial Intelligence Study – Exploiting the AI Revolution”. PwC Global. June 2017. https://www.pwc.com/gx/en/ issues/data-andanalytics/publications/artificial-intelligence-study.html 6 Economist Intelligence Unit. 2017. 7 Hao, Karen. 2019. “This is How AI Bias Really Happens—And Why it’s so Hard to Fix.” MIT Technology Review. https://www.technologyreview. com/s/612876/this-is-how-aibias-really-happensand-why-its-so-hard-to-fix/. 8 According to www.deepai.org a convolutional neural network is “a subset of deep learning and neural networks most commonly used to analyze visual imagery.” https://deepai.org/machine-learning-glossary-andterms/convolutional-neural-network. 8 This publication may be reused for noncommercial purposes if the source is cited as IFC, a member of the World Bank Group.