Key trends in AI research - plus twenty red-hot topics in the field 🔥
Insights from ETO's Research Almanac and Map of Science
Over the past several months, we've been busy improving the research data that powers many of our tools, such as the Map of Science and Research Almanac. Our last post used the data to explore trends in AI safety research. Today, we'll look at the broader field, exploring how AI research is growing, differentiating, and spreading around the world.
Key findings
Global AI research more than doubled from 2017 to 2022, fueled in large part by rapid growth in natural language processing and computer vision research. Robotics research saw slower but still notable gains.
AI safety research is growing much faster, but from a much smaller base. We estimate AI safety research comprises only 2% of overall AI research.
China leads in AI research output - but when only highly cited articles are counted, the U.S. takes the top spot.
The Chinese Academy of Sciences leads the world in both overall AI research output and highly cited AI research output.
Over half of the top research clusters in the Map of Science are AI-related, on topics ranging from basic frameworks to specific applications and implications of AI.
Overall trends
According to the latest data from the Research Almanac, about 1.2 million AI-related articles were released between 2017 and 2022. (This total, and the other Research Almanac-derived findings in this post, are based on articles with English titles or abstracts in our Merged Academic Corpus; they omit articles published solely in Chinese and non-public research. For further details and caveats, see the Almanac documentation.)
AI research grew 121% overall between 2017 and 2022.
Different AI subtopics account for notably different shares of AI research overall, and are growing at different rates:
About 400,000 articles - 32% of the AI total - were about computer vision. Computer vision research grew 121% between 2017 and 2022.
About 135,000 articles - 11% of the AI total - were about natural language processing. NLP research grew 104% between 2017 and 2022.
About 185,000 articles - 15% of the AI total - were about robotics. Robotics research grew 54% between 2017 and 2022 - notably slower than the AI field as a whole, computer vision, and NLP.
Finally, research into AI safety grew a massive 315% over this period, but from a much smaller base. According to Research Almanac data, safety research comprises only 2% of all research into AI.
Country trends
18% of the AI-related articles in the Research Almanac dataset had American authors. 25% had Chinese authors, and 17% had European authors. (Note that some articles lack information about author nationality, and articles published solely in Chinese are omitted, which could affect the numbers for Chinese authors.)
Looking only at highly cited articles, America claims the top spot from China. 36% of top-cited AI articles (defined as the 10% of articles in each publication year with the most citations) had American authors, compared to 31% with Chinese authors and 19% with European authors.
👉 To view the next five leading countries in AI research and see how global authorship has evolved over time across all countries, visit the "Countries" section in the Research Almanac.
In the aggregate, Chinese researchers seem to focus more on robotics and especially computer vision compared to U.S. researchers. U.S. researchers have a slight lead in NLP publication activity and a larger lead in AI safety work (though safety research is a very small "slice of the pie" for both the United States and China).
Top organizations
The five biggest producers of AI research articles are all Chinese research institutions, led by the Chinese Academy of Sciences, with nearly 20,000 articles released between 2017 and 2022. (Again, we count only English-language articles - Chinese organizations' counts would likely be higher still if articles in Chinese were included.) American and Chinese organizations round out the top ten.
When only highly cited articles are counted, American institutions claim a larger share of the leaderboard, though the Chinese Academy of Sciences still leads overall.
👉 To view the top ten companies active in AI research, visit the "Patents and industry" section in the Research Almanac.
Top research clusters
To round out our analysis, let’s use the Map of Science's recently revamped subject search to identify some especially hot areas of AI research.
Map of Science users often want to learn about the "top" areas of research on a particular subject. There could be many ways to define "top" research, of course, but ETO analysts often use a combination of the Map's growth and scale concepts to begin the analysis. With this approach, we look for research clusters in the Map that are both unusually large and growing unusually fast - a rare pairing.
Let's see how this works for AI in particular. Starting from the Map's list view, I'll filter down to research clusters with a growth rating of 90 or higher (that is, clusters with a higher proportion of recently published articles than 90% or more of all clusters in the Map) and with at least 2,000 new articles in the last five years. (You could experiment with different thresholds on each filter; here, I'm choosing very high thresholds in order to identify truly exceptional clusters.)
Out of nearly 86,000 clusters in the Map, only 39 clusters meet these filters as of the publication of this post - a tiny slice of the broader research landscape. Next, I'll use the subjects menu to identify the clusters with lots of AI-related research.
21 clusters remain. That's an interesting finding in itself: over half of the top research clusters in the Map of Science are AI-related (using our particular definition of "top").
What exactly are these clusters about? With the Map's key concepts feature and a few minutes spent browsing the twenty top AI clusters' detail views, we can start getting a rough lay of the land:
To those who follow AI research trends, these 20 top clusters form a microcosm of the field as a whole. Several critical strands of research appear to be represented:
Research on basic AI approaches and capabilities, with a massive and quickly growing body of work on large language models (#1407) unsurprisingly leading the way. Other clusters in the top 20 focus on high-level AI approaches such as federated learning (#9948), a set of techniques for coordinating complex AI-related algorithms across many computers, and diffusion models (#7917), a type of AI model architecture often used for multimedia.
Research on applying AI to real-world needs and problems. Some of the Map's most prominent AI clusters focus on using machine learning to diagnose plant diseases (#5167), another on helping autonomous vehicles detect nearby objects (#25074), another on using AI to monitor lithium-ion batteries (#21023), another on AI models that can generate and understand computer code (#26579), and so on - the spectrum of emerging AI applications is as broad as science itself.
Research on AI safety and security - in a nutshell, identifying the many surprises and downsides modern AI continues to reveal, and working to mitigate them. Several of the top AI clusters in our results tackle well-known challenges in this domain, including gender bias in large language models (#44973), manipulation of facial recognition systems (#40959), and AI models inadvertently leaking potentially confidential training data to attackers (#40161)
Research on society's response to AI - for example, strategies for building AI literacy as part of K-12 education (#27139) or integrating AI into business processes (#74631).
As next steps, you might use the Map to identify key countries and organizations in these different areas of AI research - or run a similar "top clusters" analysis for subjects other than AI. How about chemistry, radiology, or robotics?
As always, we're glad to help you get the most out of the Map of Science and our other resources. Visit our support hub to contact us, book live support with an ETO staff member or access the latest documentation for our tools and data. 🤖