SCIENCE & ENGINEERING INDICATORS

Artificial intelligence is defined by the Oxford English dictionary as “the theory and development of computer systems able to perform tasks normally requiring human intelligence.” A core objective of AI research and technologies is to automate or replicate human learning and cognition (Executive Office of the President 2016). AI technologies are rapidly integrating machine learning with increasingly available data, and these changes are predicted to have profound implications for the economy and society, with influences on both the production and characteristics of a wide range of products and services, and the nature of work (Cockburn 2018; WIPO 2019). These rapid changes are widely expected to have large and long-term economic and technological effects on society, including sectors and occupations not historically impacted by technology.

AI is expected to automate or augment a broad range of work tasks compared to the narrower range of routine, highly structured, and repetitive tasks that were automated by ICT (Brynjolfsson and Mitchell 2017). For example, AI has the potential to impact the nature of work and employment in two industries—medicine and finance—that have not had significant disruptions in employment from past technologies (Frank et al. 2019). AI is expected to have broad impacts like other general purpose technologies (GPTs), such as the steam engine, electrification, and IT. GPTs are a special category of technologies that are widely used and capable of ongoing technical improvement and of enabling innovation in application sectors (Bresnahan 2010). AI's impacts on the economy and society are likely to take at least several decades to become apparent based on the experience of past GPTs (Brynjolfsson, Rock, and Syverson 2018).

Experts predict that AI will increase the effectiveness and the frequency of cyberattacks on computer networks, thefts of personal data, and spread of computer viruses (Yampolskiy 2017). The federal government and corporations are supporting a wide range of research on enhancing cybersecurity, including developing AI to detect and defend against attacks (Dietterich and Horvitz 2015). Furthermore, there is a concern that AI may increase inequality across individuals, groups, and countries through the “digital divide” (Williams 2018:103–4). This digital divide refers to gaps in access to information and communication technology (OECD n.d.). In addition to inequalities in access to digital technologies, two additional potential impacts are inequalities in digital skills and technology use as well as outcomes, both benefits and harms (Lutz 2019:144).

AI research areas and technologies include machine learning, autonomous robotics and vehicles, computable statistics, computer vision, language processing, virtual agents, and neural networks (Furman and Seamans 2018). The data presented in this section cover a variety of indicators related to AI, including publication, job opening, business spending and capabilities, and venture capital funding. The Technical Appendix to this report describes the data sources.

The production and citation of peer-reviewed publications is an indicator of the creation, dissemination, and impact of S&E knowledge about AI (for a comprehensive discussion of S&E publication data, see Indicators 2020 report “Publications Output: U.S. Trends and International Comparisons”). Based on an analysis by Elsevier (2018), U.S. scientific publications in AI are more frequently cited than those from Europe and China, suggesting that U.S. research has relatively more impact (Figure 6-16). Elsevier (2018) also suggests that the impact of China’s AI research has been rising in the last several years (Figure 6-16).

By number of publications, China’s AI output has grown far more rapidly than the United States and Europe over the last decade, resulting in China becoming the leading country producing AI scientific articles (25% share) followed by the United States (17%) (Figure 6-17). China’s AI scientific publications have had a significant degree of international collaboration. For example, one study noted that more than half of China’s AI papers were authored jointly with other countries (Allen 2019:11). In addition, the expansion in the volume of scientific articles outside of Europe, China, and the United States suggests that research capability is strengthening in the rest of the world.

The number of publications grew most rapidly in three AI topic areas: machine learning and probabilistic reasoning, neural networks, and computer vision (Figure 6-18). Machine learning is a branch of computational statistics that develops algorithms to execute or solve specialized tasks and problems. Neural networks are an AI technology that closely mimic the human brain’s underlying architecture to provide capability in reasoning and problem solving. Examples that illustrate the rapid advancement of this form of AI technology include (1) the improved accuracy of AI translating news between languages, (2) the declining training time for AI to recognize an image from an image database, and (3) the improving score for AI answering grade school-level multiple-choice science questions.

While comprehensive business sector data are limited, a recent McKinsey and Company survey of businesses internationally finds AI has been adopted to some degree by a wide range of industries as measured by AI embedded in at least one business unit of the company (Figure 6-19). Industries with the highest adoption rates (greater than 55%) were automotive and assembly, telecommunications, high technology, and financial services. Many companies within these industries produce or use AI technologies in manufacturing their products or delivering their services. Firms reporting in the McKinsey survey indicate global AI adoption, including North America, Europe, Asia-Pacific, Middle East, India, and other developing countries (Table 6-3). The adoption rates of developing countries and regions rival or exceed those of developed regions, including North America. The most widely adopted AI technologies are conversational interfaces, robotic process automation, and computer vision. Technologies that have the lowest adoption rates include natural language text understanding, natural language speech understanding, and autonomous vehicles. The sidebar Commercialization of Artificial Intelligence has more information on the adoption of AI in the marketplace.

Résumé databases provide another indicator of the skills that are most searched for in relation to AI. Searches on Monster.com show that for the United States the number of jobs requiring AI skills is small but rapidly growing (Figure 6-20). Job openings requiring AI skills on Monster.com jumped from 5,000 in 2015 to 36,000 in 2017, led by machine learning techniques, deep learning, and natural language processing.

In 2016, the U.S. government released its first initiative on AI. The stated goals include public funding of AI R&D, development of a skilled AI workforce, regulation to encourage innovation while protecting the public, and support for workers that are negatively affected by AI (Executive Office of the President 2016:3–4). In 2019, the United States released the “American AI Initiative” proposing sustained investment in AI R&D and increased collaboration among government, industry, academia, nonprofits, and other countries. The plan also includes as stated goals policies to increase the pipeline of AI talent; enhance access to U.S. government data, models, and computing resources; and protect U.S. AI research and technology against strategic competitors and foreign adversaries (Metz 2019).

In 2017, China announced the Next Generation Artificial Intelligence Plan with the stated goal of becoming the “world’s premier global AI innovation center” with its AI industry projected to be valued at about $150 billion in 2030 (Kania 2017). Other stated goals in the plan include the development of an AI talent pipeline, large-scale government procurement of AI technology, promotion of collaboration between the private sector, universities, and government and investments in mitigation of AI’s potential risks and societal disruption, including a large-scale replacement of jobs by AI (Allen and Kania 2017; Lee 2018:84). China’s plan reflects a collaborative initiative where China’s central, regional, and local governments all play roles (Lee 2018:98–99).

AI startups depend heavily on venture capital funding. This section presents data from PitchBook, a company that collects venture capital data by country, industry, and technology area. These data are shown more extensively across technology areas in the Indicators 2020 report “Invention, Knowledge Transfer, and Innovation.” For context, the global total of venture capital investment was $244 billion in 2018, with $113 billion invested in the United States and $76 billion in China.

AI-related venture capital funding has grown rapidly from less than $1 billion in 2010 to $36 billion in 2018 with the majority of the funds being invested in the United States followed by China (Figure 6-21). Investment in China soared after 2016, resulting in its 28% global share in 2018. The U.S. global share dropped from an average of 83% in 2010–16 to 54% in 2017–18. The number of AI startups in the United States receiving venture capital investment more than doubled from almost 600 in 2014 to more than 1,500 in 2018 (Table 6-4). The number of startups in China, although smaller than that in the United States, also grew rapidly.