Global AI Talent Report 2019

Voir la version française

Executive Summary

There is strong evidence that the supply of top-tier AI talent does not meet the demand. Yet there is little visibility on precisely how scarce this talent is or where it is concentrated around the world. This report summarizes our second survey of the scope and breadth of the worldwide AI talent pool. Our research relies on three main data sources. First, to get a picture of the researchers who are moving the field forward, we reviewed the publications from 21 leading scientific conferences in the field of AI and analyzed the profiles of the authors. This expands upon our 2018 report, when we looked at just three conferences. Second, we analyzed the results of several targeted LinkedIn searches, which showed us how many individuals are self-reporting that they have doctorates as well as the requisite skills in different regions around the world. Finally, we looked to outside reports and other secondary sources to help us put our findings in context and better understand the talent pool in a rapidly changing global AI landscape.

Our findings show that 22,400 people published at one or more of the top conferences in the field of machine learning in 2018, up 36% from 2015 and 19% from last year alone. The number of peer-reviewed publications rose in tandem, up 25% from 2015 and 16% from the year before. Women were underrepresented, making up just 18% of the researchers publishing in these conferences. We found that the AI talent pool is highly mobile, with about one-third of of researchers working for an employer based in a country that was different from the country where they received their PhD. Our analysis showed that about 18% of the authors who published their work at the 21 conferences included in this survey — around 4,000 people — contributed research that had a major impact on the overall field as measured by citation counts in the last two years (2017-2018). The countries with the highest number of high-impact researchers (i.e., those within the 18%) were the United States, China, the United Kingdom, Australia and Canada.

A complementary survey of LinkedIn profiles indicated a total of 36,524 people who qualified as self-reported AI specialists, according to our search criteria. This represents a 66% increase from the 2018 report.

The findings in this survey indicate that there has been notable growth and expansion, both in self-reported AI expertise and in the number of authors and scientific papers published at AI conferences, reflecting a field that is dynamic and unmistakably international.


AI experts are in higher demand than ever, as diverse organizations around the world position themselves to capture the benefits of AI awareness and implementation. Self-learning algorithms are expected to allow companies to better navigate complexity and access relevant but previously invisible signals, providing real-time insights that help employees do their jobs better.

Last year, with our first survey of the global talent pool, we confirmed a general assumption within the AI industrial community: PhDs with experience in artificial intelligence are very hard to come by. In this year’s follow-up, we expanded the number of academic conferences that made up our data sample from three to 21. We investigated gender, the flow of talent across national borders and which countries, based on our results from the 21 conferences, are producing the highest-impact research as per citation count. In parallel, we collected professional profiles on LinkedIn to assess trends in self-reported AI expertise. Finally, we contextualized our findings by correlating data from outside reports and sources.

At my company Element AI, we are continually recruiting for technical expertise related to the field of AI and have done a lot of work to understand where technical talent and accumen is based geographically. As someone who believes in the importance of an open and vibrant AI commons, I hope the community will find this report to be a useful resource for addressing the talent shortage. With this exercise, the team and I aim to help bring a fuzzy picture into better focus.  

Participation in academic conferences is growing fast

The primary source of data for this survey was academic conferences in the field of machine learning. To assess expertise, we looked at the authors of papers that were published over the past year at leading international academic conferences in the field. We prioritized the following 21 conferences:1

  • Annual Conference of the North American Chapter of the Association for Computational Linguistics (NAACL)
  • Association for the Advancement of Artificial Intelligence Conference (AAAI)
  • Association for Computational Linguistics Conference (ACL)
  • Conference on Computer Vision and Pattern Recognition (CVPR)
  • Conference on Empirical Methods in Natural Language Processing (EMNLP)
  • Conference on Learning Theory (COLT)
  • Conference on Neural Information Processing Systems (NeurIPS)
  • Conference on Uncertainty in Artificial Intelligence (UAI)
  • Genetic and Evolutionary Computation Conference (GECCO)
  • International Conference on Acoustics, Speech, and Signal Processing (ICASSP)
  • International Conference on Artificial Intelligence and Statistics (AISTATS)
  • International Conference on Autonomous Agents and Multiagent Systems (AAMAS)
  • International Conference on Computer Vision (ICCV)
  • International Conference on Intelligent Robots and Systems (IROS)
  • International Conference on Machine Learning (ICML)
  • International Conference on Medical Image Computing & Computer Assisted Intervention (MICCAI)
  • International Conference on Robotics and Automation (ICRA)
  • International Joint Conferences on Artificial Intelligence (IJCAI)
  • Interspeech
  • Robotics: Science and Systems (RSS)
  • Winter Conference on Applications of Computer Vision (WACV)

This focus on 21 conferences was a methodological shift from last year. In our first report, we surveyed the authors of just three highly subscribed AI conferences. By expanding from three to 21 in this year’s report, we hope to capture a more accurate picture of the researchers who are advancing the field.

Based on the publication information for the 21 conferences available on the DBLP Computer Science Bibliography, we counted the number of authors on conference papers. This gave us a raw total number of 22,400 unique individuals who published at one or more of these top conferences in 2018.2 To see how that number compared against previous years, we collected publication data from the same 21 conferences for 2015, 2016 and 2017.3 A clear growth trend emerged, with the total number of published authors up 36% from 2015 and 19% from last year alone. Research was also ramping up, with the total number of manuscripts published at these 21 conferences up 25% from 2015 and 16% from the year before.

It’s important to note that a data set based on peer-reviewed publications submitted to academic conferences will reflect certain biases. This survey did not consider research that has only been published in peer-reviewed journals. In addition, this survey risks leaving out some vibrant research and development communities, such as private labs and think tanks, as well as independent researchers and consultants, whose work may not be well represented at international academic conferences. Moreover, all of these 21 conferences take place in English, which may result in the underrepresentation of some research communities.

With the goal of better understanding the profiles of these researchers, we extracted a random sample of 4,500 authors. We then crowdsourced the following metadata for each name on our list of authors: 1) Gender;4 2) Country where they earned their PhD; 3) Current place of work; and 4) Country of their employer. Scrutinizing the metadata for this sample allowed us to make observations about the representation of women in the field, the geographical distribution of conference authors and global hotspots for the highest-impact research. We present these findings below.

Women continue to be underrepresented in AI conference publications

In a collaboration last year with Wired, we analyzed the talent pool to try to see what proportion of machine-learning researchers were women. Our review suggested that the field was still a far way off from reaching anything close to a gender balance: across the three leading academic conferences in AI that we surveyed last year, we found that just 12% of the authors were women.

This year’s survey, which looked at 21 academic conferences, found that women are also underrepresented in this broader group, with an overall gender ratio of 18% women. According to our survey, this gender imbalance in AI exists in both industry and academia: our data indicated that 19% of the conference authors who were in academia were women versus 16% in industry.

Women’s participation in the development and deployment of AI technology is an important question, given the potential society-wide impact of machine learning. Interviewed for the 2018 Wired article, Professor Joelle Pineau, who heads the Facebook AI Research lab in Montréal, made a case for taking steps to increase the number of women in the field: “We have more of a scientific responsibility to act than other fields because we’re developing technology that affects a large proportion of the population,” she said. Sam Altman, the CEO of OpenAI, has made similar comments, noting in a recent conversation with Recode’s Kara Swisher that machine learning is both “the most skewed field I know of right now” in terms of the gender of PhD graduates and also the field “that will have the most effect on the future of the world that we live in.”

According to the AI Index 2018 Report, published by Stanford University, women are also underrepresented in undergraduate AI and machine learning courses: Stanford’s 2017 Intro to AI course was 74% male, and UC Berkeley’s course was 73% male, according to the report. An even lower percentage of women enrolled in the universities’ Intro to Machine Learning courses, with men accounting for 76 percent of the students in the Stanford course and 79% of the students in the UC Berkeley course.  The same report found that in the United States, a majority of applicants (71%) for AI jobs are men.

Our data showed that gender ratios varied across countries analyzed in this survey, and some countries had a percentage of female authors that was higher than the 18% average, with Spain (26%), Taiwan (23%) and Singapore (23%) topping our list. China and Australia (both 22%), the United States (20%), Switzerland (19%), the United Kingdom and Italy (both 18%) and India (17%) also made the top ten list of countries with the highest percentage of female authors. Some countries had a higher-than-average percentage of female researchers — Iran’s author pool was 71% female, for example — but an overall cohort that was too small to be included in the analysis. In absolute numbers, the United States led the countries with the most female authors, followed by China, the United Kingdom, Germany, Canada, France, Australia, India, Italy and Singapore.

Countries that train top AI experts are also leading employment

Our conference researcher data allowed us to make some observations about where published authors are receiving their training. First, the United States continues to graduate PhDs who are productive publishers in terms of number of accepted publications: among our sample of conference authors, more than 44% earned a PhD in the United States.5 Authors trained in China accounted for almost 11% of the authors, followed by the United Kingdom (6%), Germany (5%) and Canada, France and Japan (4% each).

A similar geographical distribution characterized the employment data. Our survey showed that American employers continued to attract researchers for work, with 46% of the sample working for a U.S.-based employer. China, where more than 11% of our sample worked, was second on the list of top countries for employment, followed by the United Kingdom (7%). Canada, Germany and Japan each accounted for 4% of the sample. Overall, the 18 biggest countries accounted for 94% of the authors.6  The top five countries — the United States, China, the United Kingdom, Germany and Canada — accounted for 72% of the authors.

A large majority of the conference sample (77%) worked in academia, while 23% worked in industry. While some of the biggest private-sector players in the field continue to draw talent to their headquarters, we are seeing many that are also trying to recruit experts to work at offices in their home country.

To assess the location of the expert’s place of work for the purposes of this report, we looked at the location of the headquarters of the business, not the country where the individual was physically working and living. So if someone did her PhD at a French institution, and now works for Google at their Paris office, our data would show that she was trained in France but works for an American company, since Google is U.S.-based. In a scenario like this one, the foreign company’s presence would bring certain benefits to the host country, including local investment, training and the fact that expertise is still physically located inside the country. Yet the foreign company would still retain ownership over the locally created IP. While we are aware that this method of counting depreciated the numbers for many ecosystems, mostly in favor of the United States, we feel it gives a better representation of how and where talent is flowing around the world.

The country that hosts a PhD student is not always the country that benefits

Our conference data also lent insight into where researchers are moving for work after completing their PhDs. Overall, we found that almost a third (27%) of the researchers in our sample were working for an employer based in a country that was different from the country where they received their training. Among the countries with at least 150 authors, the percentage was even higher at 32%. The global map of these movements is complex and the story behind each move is inevitably unique and personal. Nonetheless, this data allowed us to make some observations about the flow of AI talent across national borders.7

First, our data suggested that certain countries are particularly attractive for researchers in the field of machine learning. U.S.-based employers had the highest chance of attracting researchers who were trained abroad, according to our survey. China was the second-most-likely country to draw researchers who had received their PhDs in another country, bringing in almost one-fourth the number of researchers the United States was able to attract in absolute numbers. We presume that several different factors could be contributing to this observation, including the availability of jobs in each country.

According to our data, ten countries had a higher percentage of inbound researchers than outbound researchers: Taiwan, Sweden, Republic of Korea, Spain, the United States, Switzerland, China, Japan, the United Kingdom and Australia.8 The overall leaders when it came to receiving experts from around the world were Switzerland and Sweden, with 50% and 49%, respectively, of their talent pool receiving their training abroad. The United Kingdom was third, with 44% of its researchers obtaining their PhDs abroad, according to our data. In future projects we hope to explore why some countries received a higher percentage of inbound researchers; we expect that it could be due to a range of different reasons, including the availability of jobs and researchers returning to their home country.


This data also allowed us to compare the talent inflow and outflow in each country as a percentage of the country’s overall talent pool. Talent inflow represents the number of individuals who worked in country x but received their PhD in country y, divided by the total AI talent in the country. We posit that this measure shows how much pull an ecosystem is able to exert on talent.

Talent outflow, on the other hand, represents the number of individuals who received their PhD in country x but now work for an employer based in country y, divided by the total AI talent who got their PhD in country x. By looking at the ratio of individuals who leave the country to work for a foreign company versus those who stay in the country after their PhDs, this metric gives us insight into a country’s ability to retain its talent.

To see how countries compared in this push-and-pull dynamic, we calculated the average inflow and average outflow of all the countries and then looked at each individual country’s distance from the average inflow and distance from the average outflow. We plotted these with the talent inflow on the x-axis and the talent outflow on the y-axis, where the units represent the number of standard deviations from the mean. These values allowed us to categorize countries into the four distinct groups outlined below.

Australia, Spain, Sweden and Taiwan all saw more inflow and less outflow, as a proportion of the country’s talent pool, than average. This means that these countries are relatively more successful at both retaining the talent they’ve trained at home and attracting talent from other ecosystems. We call these ecosystems inviting countries. By contrast, we think of France and Israel as producer countries, because they saw less inflow and more outflow, as a proportion of the country’s talent pool, than average. These countries had just slightly more outflow than average, meaning they were categorized as producer countries by a small margin.

According to our data, the United States had less talent inflow and less talent outflow, as a proportion of the country’s overall talent pool, than average. This does not reflect the size of its talent pool: in absolute numbers, the United States remains the leading global talent magnet. Rather, it signals the relative stability of its talent pool. (Since our study defines a researcher’s location based on the headquarters of the company where the researcher works, these results could be seen to inflate U.S. numbers. This is because many of the companies establishing labs around the world are headquartered in the United States.) In addition to the United States, the same pattern was seen in China, Germany, India, Italy, Japan and the Republic of Korea. We term these ecosystems anchored countries.  

Finally, several countries saw both more inflow and more outflow, as a proportion of the country’s talent pool, than average. These countries are succeeding at attracting workers who were trained abroad while also seeing more post-graduate movement than average. These ecosystems, which we term platform countries, include Canada, the Netherlands, Singapore, Switzerland and the United Kingdom.

Our conference data also shed light on notable talent exchanges between some countries. There was a particularly strong exchange between China and the United States, with neither making notable gains over the other: we found that around 500 experts in our data set of 22,400 researchers received their PhD in China and then went on to work for a U.S. employer, with 500 more receiving their PhD in the United States and then going on to work for a China-based employer. A similar phenomenon was observed between the United States and the United Kingdom, where we found about 325 experts moved from the United States to the UK, with about the same number heading in the other direction.

Since the conference data doesn’t tell us anything about a researcher’s nationality, we don’t know how many researchers are coming from abroad to the United States (or any other country) for their doctorates and then either staying or leaving. However, we do know that, overall, U.S. universities receive a high number of graduate students from abroad. In 2015, for example, international students earned about a third of science and engineering graduate-level degrees granted in the U.S., and 76% of these graduates said they expected to stay in the country. At some universities, the percentages of foreign graduate students is notably higher: at New York University’s Tandon School of Engineering, for example, 80% of graduate students are reported to come from abroad.

This trend is also reflected among the cohorts graduating from American universities with computer science PhDs. Over the 2012-15 period, 7,851 foreign students earned doctorates at an American institution in the fields of mathematics or computer science, according to the U.S. National Science Foundation. Of that group, more than 79% said they had plans to stay in the United States, and 53% had definite plans to stay. There is therefore evidence that many non-U.S. nationals are moving to the United States to do their PhDs (or sooner) and then staying in the country to work. These early-stage talent migrations, which presumably occur across countries and are not captured in our data, likely indicate that there is even more talent flow across national borders than is shown by this survey.  

United States, China, United Kingdom, Australia and Canada lead in high-impact research

This year’s survey found that the total number of authors at the top international academic conferences was up 19% from last year. With the aim of assessing the impact these authors are currently having in the field, we analyzed the citations for each of their 2017 and 2018 publications.9 We found that 18% — about 4,000 people — are doing research that is having a notable impact on the overall field as measured by citations received in the last two years. These are AI experts who are publishing the most-cited papers at the top academic conferences and whose knowledge is deep enough, we believe, to continue to make substantial contributions to the field. These experts may also be a potential source of applied talent for teams working to bring theory to application.

Our conference data showed that these researchers were concentrated more in some countries than in others. The top five countries for total number of these researchers were the United States (1,095), followed by China (255), the United Kingdom (140), Australia (80) and Canada (45).

The picture changes somewhat when we look at researchers doing the most impactful work as a percentage of the total number of AI researchers in a given country. These are countries where a higher-than-average percentage of the local talent pool is making high-impact contributions to the field, suggesting that these countries may be doing something right when it comes to nurturing top-tier talent. Here, Australia comes out on top, with 18% of its overall author pool publishing high-impact work, followed by the United States, the United Kingdom and China (13% each), Switzerland (11%), Singapore (9%), Sweden and Spain (8% each) and Israel, Canada and Italy (7% each).   

In all the countries, the most impactful research was more likely to come from academia than industry, but for some more so than others.10 China was the country where high-impact research was most likely to come from academia (90%), followed by Italy (86%), the United States (84%), Germany (83%) and Taiwan (81%). France was the country with the highest percentage of impactful research coming from industry (30%), followed by India and Israel (29% each), Spain (28%) and the United Kingdom (27%).

Social network data suggests more people are self-reporting AI expertise

Our conference researcher data showed notable year-over-year increases in the number of authors and papers published at 21 of the top academic conferences in the AI field. To try to get a sense of whether the job market was growing in tandem, we surveyed LinkedIn, the most widely used professional networking site in the world.

In our last talent report, we also analyzed the results of several targeted LinkedIn searches to get a picture of the broad talent pool. Just as we did for last year’s report, this year we defined “talent” as people who have proven technical competency in machine learning, several years of work experience and can collaborate and thrive in an interdisciplinary environment. These individuals should be able to identify a problem that can be solved with modern machine learning techniques, build and implement that solution from scratch and then optimize the solution to work efficiently.

To capture these experts’ profiles on LinkedIn, we started by setting our search parameters to include the following job titles: “data scientist,” “research scientist,” “machine learning engineer,” “machine learning researcher” and “data analyst.”11 Our search was also designed to only capture individuals with PhDs. While a doctorate is not an objective requirement to be an AI expert, we worked with the assumption that it is still a useful proxy for the highly technical skills required to qualify as a specialist. Finally, individuals must have described their work as including “machine learning.”

Based on these search queries, our survey indicated a total of 36,524 people who qualified as self-reported AI specialists. Last year, by contrast, our survey of LinkedIn indicated that worldwide there were 22,064 experts. This represents a 66% increase from the 2018 report.

Similar to last year, our 2019 survey found that AI specialists who are on LinkedIn are concentrated in the United States, the United Kingdom, Canada, France and Germany. Yet those are not the countries in the LinkedIn data that saw the biggest increases from last year: Italy, Tunisia, Israel, Estonia and Argentina saw the most relative growth (see the infographic above).

Our LinkedIn sample indicated that these individuals were trained in various academic disciplines. Computer science, which 28% listed as their academic discipline, was the leading field. This percentage was higher in some countries, including France (47%) and China (44%). Likewise, some countries were overrepresented in other disciplines. Take physics: 9% of all experts said they were trained in physics, but in Germany it was 28%. Another example was math and statistics. Eighteen percent listed it as their academic discipline, a share that jumped to 27% of the Israel- and U.S.-based profiles and 35% of the Russia-based profiles.12


There are some important caveats to this data. First, all information on LinkedIn is self-reported: individuals opt into the site and then describe their education, experience and current work in their own words. Second, although it has broad reach, some countries are poorly represented on the site. For example, LinkedIn reports that some 144 million Americans currently have profiles on LinkedIn — representing more than 44% of the U.S. population. LinkedIn is also widely used in Canada, where about 38% of the population is reportedly registered on the site. In Russia, by contrast, the penetration rate is reportedly just 5%. In China, a major player in machine learning, the penetration rate is even lower at 3%.

Despite these particularities, we have found that in countries where it is more widely used, LinkedIn activity can give insight into changes in the size of and interest in the AI field. In the case of this survey, we found major increases in self-reported machine-learning expertise. We believe this likely reflects an expanding talent pool propelled by a market that is increasingly prizing AI skills and expertise. While there may be an element of professional ‘rebranding’ reflected in this data, we tried to control for this possibility with searches that targeted specific educational requirements.

AI Talent Hotspots Around the Globe

To provide more context on how AI research is being fostered around the world, below we survey some of the countries that, according to our survey, are driving high-impact work. Different countries have different approaches to attracting and sustaining top-tier talent, and we describe a few of these strategies and dive deeper into the country-level data.



Our data, which pulls exclusively from LinkedIn and academic conferences, likely has a number of blind spots, and some countries are potentially underrepresented in the report. China is the most conspicuous of these possible omissions. The LinkedIn penetration rate in China is just 3%. Given the highly international nature of the AI industry, it is possible that AI experts in China may be better represented on LinkedIn than are other industries. Nonetheless, the picture of China painted by LinkedIn data alone is without a doubt incomplete.

China’s representation in our conference data is more robust. About 11% of the overall conference authors received their training in China, and the same percentage of authors were working for a Chinese employer. China accounted for about 12% of the female authors in the sample and 14% of the authors publishing the most impactful research. Looking at China’s overall total number of authors who published at the top academic conferences in 2018, 13% of them were included in this high-impact group. This is the same percentage as in the U.S. Of these high-impact researchers, nine out of ten were working in academia, the highest proportion in our sample.  

China has an active publishing ecosystem, and not all papers published in China are released in the common international conference language of English. It is therefore possible that surveying international academic conferences does not fully capture the amount of impactful research coming out of China, where the government has made the development of AI a national priority and is investing accordingly. In the July 2017 announcement of its national plan for AI, the Chinese government set out its aim to become “the world’s primary AI innovation center,” with an industry worth $150 billion, by 2030.

China has a number of advantages when it comes to AI, including a massive amount of data, its current vibrant entrepreneurial environment and government support, according to Kai-Fu Lee, the former president of Google China and author of AI Superpowers: China, Silicon Valley and the New World Order. Indeed, the China AI Development Report 2018, published in June 2018 by Tsinghua University, claims that China leads the world in AI-related scientific publications, patents and total venture capital investment but lags behind when it comes to one crucial aspect: top AI talent. “Developing countries such as China are underrepresented by top AI talent,” according to the report. “The United States maintains its safe lead…in the world’s top-tier AI talent pool based on the H-index,”13 followed by the United Kingdom, Germany, France, Italy and China, according to the report’s ranking system. When it comes to total overall talent, however, the report asserts that China is second only to the United States. Nonetheless, there are indications that the gap between the two countries could be narrowing rapidly: a recent study released by the Allen Institute for Artificial Intelligence found that China is set to outpace the United States in high-impact14 publications by 2020. This may be one of the reasons why China’s AI strategy ranks the cultivation of top-tier AI talent as a major priority.

Part of that strategy seems to be to bring Chinese researchers working abroad back to their home country. In 2008, China established the “Thousand Talents Program” aimed at attracting foreign and Chinese researchers working in other countries. Offering distinguished research positions, bonuses and grants, the Thousand Talents Program has created incentives for thousands of researchers to bring their work to China, a majority of them from the United States. In parallel, more than 300 “entrepreneurial parks” have been built to house students and workers returning from abroad, according to the state-owned newspaper China Daily.

Data from the U.S. National Science Foundation shows that, compared to previous years, a smaller percentage of Chinese PhD students who obtain their doctorates in the United States in the field of mathematics or computer science are reporting that they plan to stay in the country following graduation. Over the 2012-15 period, 42% of non-American PhD graduates in these fields were Chinese nationals, with 87% saying they had plans to stay and 57% saying they had definite plans to stay. Although the percentage of Chinese PhD students in these fields who said they had plans to stay in the United States following their studies was higher than the average across all countries,15 it has nonetheless fallen as compared to the 2004-07 period, when 91% of Chinese PhD graduates in math and computer science said they planned to stay in the country after their degree, and 65% reported that they had definite plans to stay. This data suggests that “sea turtles” — as students who return home after a long period studying abroad, are commonly called in China — are more and more likely to be coming back with American-minted math and computer science doctorates.

Recruitment plans like the Thousand Talents Program, coupled with the new business opportunities offered by unprecedented state investment in AI, are motivating factors for some researchers considering moving their labs to China. Another consideration may be what’s known as the “bamboo ceiling”: the idea that while STEM graduates may be readily employable by foreign companies, they may not rise as easily through the ranks as some of their colleagues. A 2015 study looking at employment data from Google, Hewlett-Packard, Intel, LinkedIn and Yahoo supports this supposition: although Asians made up 27% of the professional workforce in these companies, they represented only 14% of the executives, according to the study.  


Out of our sample of more than 36,500 LinkedIn profiles, about 2% (607) was based in Singapore. Of this group, 72% reported at least six years of experience, and 93% said they had at least three years of experience.

Authors trained in Singapore and/or working for a Singapore-based employer accounted for 1% of our sample. One percent of researchers based out of an institution in Singapore qualified as top talent. When we looked at the number of authors compared to population, however, Singapore was ahead of every other country in the world, with five conference authors for every 100,000 individuals. Singapore was also a leader on the question of gender, with women making up 23% of conference authors.  

An aggressive stance on connectedness could put Singapore in a position to quickly accelerate in AI: in the 2017 Global Smart City Performance Index, which assessed cities based on their degree of connectivity when it comes to mobility, healthcare, public safety and productivity, Singapore was ranked first in the world in each category. Startup Genome’s 2018 Global Startup Ecosystem Report also included Singapore as one of the “Top 10 Ecosystems for Local Connectedness.” A 2017 report from McKinsey on AI in Southeast Asia found that of the nations that make up the Association of Southeast Asian Nations (ASEAN), “Singapore is leading the region in AI experimentation across multiple industries.”16 One example highlighted by the report is Singapore’s Smart Mobility 2030 plan, which aims to use AI for the real-time optimization of car, bus, train and bicycle traffic.

Indeed, the government of Singapore is proactively promoting AI. In 2017, Singapore’s National Research Foundation, which sets the country’s agenda for research and development, announced AISG, a five-year, $150-million initiative dedicated to developing AI solutions. A year later, noting that the “conditions are now ripe for us to take Government’s digital transformation to the next level,” Singapore’s Digital Government Blueprint called for all ministries to implement AI on at least one project “for service delivery or policy making” by 2023.

In parallel, Singapore is devising guidelines for the responsible implementation of AI. The government shared a first draft of its Model Artificial Intelligence Governance Framework in January of this year, asking organizations to pilot the framework and provide feedback for future versions.

In February, the Minister-in-charge of the Smart Nation Programme Office, Vivian Balakrishnan (who is also the Minister for Foreign Affairs), announced that the country is set to “double down” on AI with the aim of becoming a major hub for the deployment of the technology. Singapore is also investing in training and upskilling: AISG’s “AI for Everyone” initiative, for example, aims to teach basic machine learning to some 10,000 people ranging from high-school students to working adults. “We do not expect everyone to become an AI expert,” Balakrishnan said. “But AI … is a general purpose technology, and we want our workforce to be able to use [these] tools to participate meaningfully in the future AI-driven economy.”

Republic of Korea

Out of our sample of more than 36,500 LinkedIn profiles, about half of one percent (192) was based in the Republic of Korea (where the networking site is not widely used). Authors trained in the Republic of Korea and/or working for an employer based in the country accounted for 2% of our sample.

The Republic of Korea is making notable strides in developing AI capabilities, and these investments fit into the country’s rich history of supporting technology research and development. This year, for example, for the sixth year in a row, the Republic of Korea was ranked as the most innovative economy in the world, according to Bloomberg’s Innovation Index.

This commitment to innovation is reflected in recent investments in AI: in 2018, the Republic of Korea committed to allocating $2 billion over four years to AI research and development with a goal “to reach the global top four by 2022.” Talent will be a major focus of the program, which aims to foster 1,370 AI researchers, 350 of whom should qualify as top-tier talent. At the same, thousands of scholarships will be granted to students studying AI in the Republic of Korea, and an accelerated training program will be offered to help employers start filling open jobs more quickly.


Our data suggests that Australia is an emerging AI hotspot. Looking at the overall numbers, about 4% (1,299 profiles) of our LinkedIn data was based in Australia, and 3% of the total number of authors at the top 21 international conferences received their training in Australia. The same percentage of authors worked for an Australia-based employer.

When it came to the most impactful research, however, Australia had the fourth-highest number of researchers in the world. Of its overall pool of authors, 18% — the highest of any country — was included in this high-impact category. According to our data, Australia is also one of the countries bringing in more talent than it is losing to foreign companies and universities.  

North America


Canada has about one-tenth the population of the United States, and its representation on LinkedIn was roughly proportional, with 1,487 profiles indicating that they were based in Canada. Looking at the conference data, researchers who did their PhDs in Canada as well as those researchers working for a Canada-based employer both accounted for about 4% of our sample. Canada was a leader in top-tier talent, with the fifth-highest number of authors who had published high-impact research.

There are clues, however, that the Canadian AI ecosystem is larger than this data set would suggest and is growing fast. The Canadian government made a major investment in AI in 2017, when it allocated $125 million for the Pan-Canadian Artificial Intelligence Strategy, an initiative aimed at nurturing AI talent. In 2018, our research found that the number of active AI-related startups in Canada had risen 28% over the year prior, representing almost 650 AI startups across the country. In the same period, we also noted the increasing number of international players opening AI labs in some of the most vibrant cluster cities in the country, including Toronto, Montréal, Vancouver, Edmonton, Waterloo, Ottawa and Québec City.

United States

Out of our sample of more than 36,500 LinkedIn profiles, nearly half (15,747) were based in the United States. A notable proportion of these experts — about one in five — currently work or have previously worked for Microsoft (1,077), IBM (667), Google (697), Amazon (511) and/or Apple (393). Eighty-seven percent of these profiles have at least six years of experience, and almost all of them (97%) say they have at least three years of experience. About a third (36%) of the U.S.-based experts in this talent pool are working in the San Francisco Bay Area.  

When it comes to conference data, the United States topped almost every measure in absolute numbers. People trained in the United States accounted for nearly half (44%) of the published authors in the sample. Forty-six percent were employed by a company or university based in the United States. Among women, that percentage was even higher: of the 4,085 female authors in our data set, 1,960 (48%) were working for a U.S.-based employer. Authors working at a U.S. institution accounted for 60% (1,095) of the researchers doing the most impactful research as measured by citations received in the last two years. About 13% of the total pool of U.S.-based researchers were at or above the threshold to qualify as “high-impact” authors. Of this top talent, 84% worked in academia and 16% in industry.

That the United States would lead across several metrics is unsurprising. For one thing, American universities continue to lead in fundamental AI research. According to Kai-Fu Lee, the United States is set to maintain a major global advantage in “research leadership” and graduate education for at least the next 30 to 40 years.

Moreover, American companies and the U.S. government are investing heavily in AI research and development. Facebook, Apple, Microsoft, Google and Amazon, all based in the United States, are doubling down on machine learning. Since 2011, the majority of AI startup equity investments have gone to those based in the United States, “both in the number of investment transaction[s] and in U.S. dollars invested, accounting for two-thirds of the total value of investment,” according to a December 2018 report from the OECD. The U.S. military is also highly invested in AI: the Defense Advanced Research Projects Agency (DARPA), for example, announced in November 2018 that they would earmark $2 billion over a five-year period for “new and existing programs to create the third wave of AI technologies.”

The European region

Out of our sample of more than 36,500 LinkedIn profiles, about 37% (13,553) was based in the European region.17 Of these, there were seven countries that accounted for at least 2% of the sample: about 9% (3,387 profiles) was based in the United Kingdom, 4% (1,426) in France, 4% (1,351) in Germany, 3% (975) in Spain, 2% (687) in the Netherlands, 2% (681) in Italy and 2% (625) in Switzerland.  

Of the total number of researchers who published at the top 21 conferences last year, 27% of them received their training in the European region, and 25% worked for an employer based in one of the countries included in this group. A quarter (25%) of the female researchers in our sample were working for employers based in the European region. Among the researchers doing the most impactful research, 16% of them were working for a company based in one of these countries.

The European Union (EU) and several individual countries in the European region are making major investments in developing AI skills and capabilities. In 2018, the European Commission (EC) said that it would commit €1.5 billion to fund AI research over a two-year period as part of Horizon 2020 program, representing a 70% yearly increase. At the country level, member states are also allocating money toward research in the AI field: last year in the United Kingdom, for example, the government and the private sector teamed up to invest some £1 billion in AI research. France, meanwhile, invested €1.5 billion through 2022. The EC is set to release a plan later this year detailing how, together with member states, Europe can reach a total AI spend of €20 billion per year for the next ten years.

A recent report compiled by Charlotte Stix, a Research Associate at the Leverhulme Centre for the Future of Intelligence at the University of Cambridge and an Advisor at  Element AI, found that brain drain poses a challenge for AI research in the EU. Stix also flagged lagging VC investments in startups in the EU, which may be an indicator for future talent movement. A 2018 report published by the OECD showed that 8% of the world’s AI equity investment went to the EU in 2017, up from 1% in 2013. The majority of this investment is concentrated in three countries: from 2011-2018, the United Kingdom accounted for 55%, while Germany received 14% and France 13%. Despite these trends, Stix’s review of the EU AI ecosystem makes a strong case that Europe is set to emerge as a global leader in “ethical AI,” notably through the ongoing development of relevant/suitable guidelines, frameworks and regulations (e.g., the General Data Protection Regulation18). One prominent example is the European Commission’s independent advisory group, the High Level Expert Group on AI (AI HLEG).19 This group, made of up 52 experts, is tasked with drafting recommendations for the development of new policy on AI and related ethical, legal and societal issues.  


Our 2019 Global AI Talent Report shows that the number of authors and impactful scientific papers published in the field of AI, as well as self-reported AI expertise, have notably increased over the past year. Women continue to be underrepresented, but some countries are closer to reaching gender parity than others. Looking at the geographical spread of talent, the United States leads in absolute numbers on nearly every metric. Yet the field is unmistakably international, with each local ecosystem characterized by its own unique strengths and strategies. Countries that focus on nurturing this expertise are helping to grow the pie of worldwide AI talent needed to bring the field forward.

Lire la version française

Written with Grace Kiser and Yoan Mantha

Research by Yoan Mantha

Data visualization by Santiago Salcido

Translated into French by Melissa Guay

Website and packaging by Doriane Soulas, Benoit Lavoie-Lamer, Morgan Guegan, Kayla Gillis, Vincent Caruana, Wei-Wei Lin and Genevieve Jacovella Remillard

Special thanks to those at Element AI and associated consultants who provided invaluable commentary:

Adélaïde Andriot | Caroline Bourbonnière | Chang Sub Chang | Kevin G. Clark | Christophe Coutelle | Gregg Delman | Sabina Dessertine | Guillaume Gagnon | Paul Gagnon Peter Henderson | Simon Hudson | Valérie Lafontaine | Clare Lee | Vyara Ndejuru | Lara O’DonnellMarc-Etienne Ouimette | Clement Soh | Charlotte Stix | Jeffery Yune


  1. Many thanks to the Element AI Office of the Chief Science Officer (OCSO), which helped us identify these conferences.
  2. All duplicates were removed from this data set.
  3. With two exceptions: we didn’t have access to NAACL data for 2017 and 2018 or to data from ICCV, a biannual conference, for 2016 and 2018.
  4. Unfortunately we were unable to capture non-binary gender categories in this report, but we hope to develop a more inclusive approach for future efforts.
  5. To see where experts received their training, we crowdsourced PhD location data for each author in the random sample. In addition to gathering educational profiles, this step also allowed us to weed out any undergraduate authors or edge cases that didn’t fit the scope of this project.
  6. This group includes any country with 150 or more authors in our sample. These countries are: Australia, Canada, China, France, Germany, India, Israel, Italy, Japan, the Netherlands, Singapore, Republic of Korea, Spain, Sweden, Switzerland, Taiwan, United Kingdom and the United States.
  7. These observations are limited given the narrow scope of the survey and myriad variables that were not taken into account, including, for example, GDP, citizenship, the total number of PhDs produced per country and immigration policy.
  8. Note: this only includes countries for which there was enough data to obtain statistical significance.
  9. Many thanks to the Element AI Office of the Chief Science Officer (OCSO), which helped us define this task. We looked at 2017 and 2018 citations, rather than H-index, because we wanted to try to capture as best we could the researchers’ current impact, rather than their full career impact. We counted the citations for all 2017 and 2018 publications and defined the threshold as the inflection point of the sorted number of citations for individual publications.
  10. Note that our methodology did not allow us to distinguish dual affiliations.
  11. Many thanks to the Element AI Talent team for helping us define these search terms.
  12. Note that on LinkedIn, users have the option to list multiple areas of study, and indeed many of the profiles included more than one academic discipline.
  13. The H-index, a measurement of an author’s scholarly impact, is defined as “the highest number of publications of a scientist that received h or more citations each while the other publications have not more than h citations each.”
  14. “High-impact” in this case was defined as the top 10% of papers by citation count.
  15. Of the overall group of foreign students obtaining a PhD in the U.S. in the field of mathematics or computer science, more than 79% said they had plans to stay in the U.S., and 53% had definite plans to stay, according to the National Science Foundation.
  16. ASEAN includes Brunei, Cambodia, Indonesia, Malaysia, Myanmar (Burma), Philippines, Singapore, Thailand and Vietnam.
  17. In this report, what we term the “European region” is not limited to the official member states of the European Union. We include the following non-EU countries, many of which are actively involved in European programs and initiatives: Albania, Belarus, Bosnia-Herzegovina, Iceland, Israel, Kosovo, Macedonia, Moldova, Norway, Russia, Serbia, Switzerland, Turkey and Ukraine. The United Kingdom is also included in this group.
  18. The General Data Protection Regulation (GDPR) is a major EU data protection and privacy law that went into effect in May 2018.
  19. Disclosure: I am a member of the AI HLEG.