**The Unlikely Harmony of AI Rivals: Why the US and China are Collaborating Behind the Scenes**
When it comes to artificial intelligence, the US and China are often seen as bitter rivals, each trying to outdo the other in the development of cutting-edge algorithms and models. But, in a fascinating twist, the world’s two AI superpowers are working together in cutting-edge research to an astonishing extent.
According to a recent analysis of over 5,000 AI research papers presented at the prestigious Neural Information Processing Systems (NeurIPS) conference, a remarkable 3% of the papers (141 out of 5,290) have collaboration between authors from US and Chinese institutions. This trend is consistent, with 134 out of 4,497 papers in 2024 involving authors from both countries.
But that’s not all – the analysis also reveals how algorithms and models developed in one country are being shared and adapted across the Pacific. The transformer architecture, developed by a team of researchers at Google and now widely used in the industry, is featured in 292 papers with authors from Chinese institutions. Meanwhile, Meta’s Llama family of models was a key ingredient in the research presented in 106 of these papers. And, the increasingly popular large language model Qwen from Chinese tech giant Alibaba appears in 63 papers involving authors from US organizations.
Jeffrey Ding, an assistant professor at George Washington University who tracks China’s AI landscape, isn’t surprised by the level of teamwork. “Whether policymakers on either side like it or not, the US and Chinese AI ecosystems are inextricably entwined – and both benefit from the association,” he says.
The analysis may simplify the extent to which the US and China share ideas and expertise, but many Chinese-born researchers study in the US, forging bonds with colleagues that last a lifetime. And, as Katherine Gorman, a spokesperson for NeurIPS, puts it, “NeurIPS itself is an example of international collaboration and a testament to its significance in our field. Collaborations between students and advisors often continue long after the student has left their university. You can see these kinds of signs that indicate cooperation throughout the field in many places, including professional networks and past collaborators.”
So, while the US and China may be at odds in many areas, when it comes to AI research, they’re still working together.
**How I Analyzed the Papers**
I wanted to see if I could use AI to help analyze the NeurIPS papers, so I used OpenAI’s code-writing model, Codex, to do the legwork for me. After writing a script to retrieve all the papers, I used the model to dig into each one and do some analysis. This involved having Codex write a script to search for US and Chinese institutions in the author field of each paper.
It was an interesting experiment, and it showed me the potential for coding models to automate helpful chores. There’s a lot of panic about AI replacing coding jobs, but this is something that I wouldn’t have had the time or funds to build otherwise. The process involved a lot of trial and error, and you have to be very careful, as AI models can make surprisingly silly errors even when they’re being quite smart. I had to make sure that each report included a method for me to review the results, and I checked as many as possible manually.
This is just an excerpt from my AI Lab newsletter. If you want to read more, check out the full newsletter here.
