USA and China is in many ways its biggest rival in artificial intelligence, with companies racing to outdo each other in algorithms, models and specialized silicon. And yet, the world’s artificial intelligence superpowers continue to collaborate to a surprising degree when it comes to cutting-edge research.
WIRED analysis of more than 5,000 artificial intelligence research papers presented last month at the industry’s premier conference, Neural Information Processing Systems (NeuroIPS) reveals a significant scope of cooperation between American and Chinese laboratories.
The analysis found that 141 of the 5,290 total articles (about 3 percent) involved collaborations between authors affiliated with U.S. institutions and authors affiliated with Chinese institutions. Collaboration between the United States and China also appears to be fairly consistent, with 134 of 4,497 total articles in 2024 featuring authors from institutions in both countries.
WIRED also examined how algorithms and models developed in one country are shared and adapted across the Pacific. The transformer architecture, developed by a team of researchers at Google and now widely used in the industry, was described in 292 papers by authors from Chinese institutions. Meta’s Lama family of models was a key component of the research reported in 106 of these articles. Meanwhile, Chinese tech giant Alibaba’s increasingly popular Qwen large-language model appears in 63 articles featuring authors from U.S. organizations.
Jeffrey Ding, an assistant professor at George Washington University who tracks the artificial intelligence landscape in China, says he’s not surprised by the level of teamwork. “Whether policymakers on both sides like it or not, the U.S. and Chinese AI ecosystems are inextricably linked and both benefit from this arrangement,” Ding says.
The analysis undoubtedly simplifies the extent to which the United States and China share ideas and talent. Many researchers born in China study in the U.S., forming lifelong bonds with colleagues.
“NeurIPS itself is an example of international collaboration and a testament to its importance in our industry,” Katherine Gorman, spokeswoman for NeurIPS, said in a statement. “Collaboration between students and advisors often continues long after the student leaves the university. These types of signals indicate collaboration in many places, including professional networks and former colleagues.”
The latest issue of WIRED explores the many ways China is shaping this century. But with U.S. politicians and tech executives using concerns about China’s economic rise as justification for abandoning regulations and driving staggering investment, our analysis is a good reminder that the world’s two AI superpowers still have a lot to gain from working together.
A note on methodology
To analyze NeurIPS articles, I used Codex, OpenAI’s code writing model. After writing a script to download all the articles, I used the model to explore and analyze each of them. This required Codex to write a script to search for US and Chinese institutions in the author field of each article.
The experiment provided a fascinating insight into the potential of coding models to automate useful tasks. There’s a lot of panic about AI replacing coding tasks, but it’s something I wouldn’t normally have the time or budget for. I started by writing scripts and having Codex modify them before simply asking Codex to do the analysis itself. This involved the model importing Python libraries, testing various tools, and writing scripts before creating reports for review. This process involved a lot of trial and error, and you have to be especially careful because AI models make surprisingly stupid mistakes, even when they are quite sharp. I needed to make sure each report had a way to review the results, so I manually checked as many reports as possible.
This is the release Will Knight AI Lab Newsletter. Read previous newsletters Here.