Data Brief

Comparing U.S. and Chinese Contributions to High-Impact AI Research

Ashwin Acharya

and Brian Dunn

January 2022

In the past decade, Chinese researchers have become increasingly prolific authors of highly cited AI publications, approaching the global research share of their U.S. counterparts. However, some analysts question the impact of Chinese publications; are they well respected internationally, and do they cover important topics? In this data brief, the authors build on prior analyses of top AI publications to provide a richer understanding of the two countries’ contributions to high-impact AI research.

Download Full Report

Executive Summary

While the large and growing number of Chinese artificial intelligence publications is well known, the quality of this research is debated. Some observers claim that China is capable of producing a high quantity of AI publications, but lags in original ideas and impactful research.1 Even Chinese researchers occasionally criticize their country’s academic system for its lack of innovation in AI.2 In recent years, however, quantitative analyses have found that Chinese AI publications are increasingly influential.3

AI is an economically and strategically important emerging technology, and the Chinese government has promoted domestic AI progress for years. Chinese and U.S. strengths in AI development will have ramifications for the two countries’ relative capabilities in areas ranging from science and medicine to battlefield applications. Further, Chinese researchers’ ability to produce impactful AI advances reflects on the more general question of whether Beijing can foster impactful innovation—a capability sometimes called into question by U.S. and European observers.4

This brief provides a data-driven comparison of U.S. and Chinese AI research, examining both publications that are highly cited and those published in top AI conferences.5

We find that:

  • Chinese researchers’ output of highly cited AI publications is increasingly competitive with the work of their U.S. counterparts. Over the past decade, Chinese researchers have published a growing share of the world’s top-5-percent AI publications, rising from half of U.S. output in 2010 to parity in 2019.
  • Top Chinese publications are often cited outside of China, although China still lags behind the United States in international citations. Highly cited Chinese publications receive 35 percent of their citations from non-Chinese sources, and their citation count from international sources has steadily increased over time. However, U.S. publications maintain a lead over Chinese ones in international citations, reflecting the United States’ closer ties to other leading AI producers.
  • China contributes an increasing share of publications at 13 top AI conferences, while the U.S. share of publications at these conferences is stagnant. Between 2010 and 2019, China’s share of these publications grew from 13 percent to 31 percent, while the U.S. share fell from 55 percent to 51 percent.
  • A notable share of both U.S. and Chinese researchers’ high-impact AI publications were U.S.-Chinese collaborations. For example, such collaborations accounted for 24 percent of both countries’ highly cited AI publications in 2019.
  • Some research clusters in CSET’s Map of Science contain far more top AI publications than others. These clusters’ topics reflect some areas of interest for Chinese and U.S. researchers.
    • Clusters with a disproportionate share of China’s highly cited and top-venue publications include publications on general-purpose computer vision research, as well as applications of AI to surveillance and industry.  
    • Clusters with a disproportionate share of the United States’ highly cited and top-venue publications cover algorithmic innovations in deep learning, such as transformers and deep reinforcement learning, as well as AI ethics and safety research.
  • The United States and China combined publish about 65 percent of highly cited AI research. U.S. allies, particularly the European Union and the Five Eyes countries, also make significant contributions to AI research.6
Download Full Report
Comparing U.S. and Chinese Contributions to High-Impact AI Research
  1. For example, a McKinsey report claims that “China lags behind the United States and the United Kingdom in terms of fundamental research that advances the field of AI.” Dominic Barton, Jonathan Woetzel, Jeongmin Seong, and Qinzheng Tian, “Artificial Intelligence: Implications for China” (McKinsey & Company, April 2017), https://www.mckinsey.com/~/media/mckinsey/featured percent20insights/China/Artificial percent20intelligence percent20Implications percent20for percent20China/MGI-Artificial-intelligence-implications-for-China.ashx
  2. For example, several Chinese researchers recently wrote that “although aggregate AI research outputs (e.g., scientific publications, patents) are rising rapidly in China, truly original ideas and breakthrough technologies are lacking.” Daitian Li, Tony W. Wong, and Yangao Xiao, “Is China Emerging as the Global Leader in AI?,” Harvard Business Review, February 18, 2021, https://hbr.org/2021/02/is-china-emerging-as-the-global-leader-in-ai.
  3. Jiangjiang Yang and Oren Etzioni, “China is closing in on the US in AI research,” Allen Institute for AI (Medium), May 11, 2021, https://medium.com/ai2-blog/china-is-closing-in-on-the-us-in-ai-research-ea5213ae80df; Dewey Murdick, James Dunham, and Jennifer Melot, “AI Definitions Affect Policymaking” (Center for Security and Emerging Technology, June 2020), https://cset.georgetown.edu/wp-content/uploads/CSET-AI-Definitions-Affect-Policymaking.pdf.
  4. Robert D. Atkinson and Caleb Foote, “Is China Catching Up to the United States in Innovation?” (Information Technology & Innovation Foundation, April 2019), https://projects.iq.harvard.edu/files/innovation/files/2019-china-catching-up-innovation.pdf.
  5. Our analysis is not limited to publications in academic journals and conferences; it also includes preprints on the ArXiv repository, which private AI labs often use to report their latest innovations. For example, most of the publications linked on OpenAI’s publications page are ArXiv preprints. Such preprints can still appear in our subset of highly cited AI publications. For example, OpenAI’s ArXiv preprint “Deep Double Descent: Where Bigger Models and More Data Hurt” appears in the CSET merged corpus as one of the most highly cited AI publications of 2019, placing in the highest percentile for computer science publications in that year. Preetum Nakkiran, “Deep Double Descent: Where Bigger Models and More Data Hurt,” arXiv preprint arXiv:1912.02292 (2019), https://arxiv.org/abs/1912.02292.
  6. We refer to the Five Eyes countries, excluding the United States, as CANZUK. This group includes Canada, the United Kingdom, Australia, and New Zealand. In this brief, European Union refers to the 27 member states of the after the departure of the United Kingdom: Austria, Belgium, Bulgaria, Croatia, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, and Sweden.