There is a large and widening gap between research activity in China and similar activity in the US, according to a report from principals at Air Street Capital and Plural. Chinese institutions have produced 4.5 times as many papers than American institutions since 2010, according to the report. The total of AI-focused papers produced in China exceeds those of the US, India, the UK, and Germany combined.

The report's authors say this gap has implications for security and geopolitics, including surveillance, autonomy, scene understanding, and object detection.

The report says that AI-driven scientific research continues to lead to breakthroughs, “but major methodological errors like data leakage need to be interrogated further, leading to a growing reproducibility crisis in AI-based science driven in part by this data leakage.”

The report cites evidence of new, independent research labs rapidly open-sourcing the closed-sourced output of major labs, noting that “despite the dogma that AI research would be increasingly centralized among a few large players, the lowered cost of and access to compute has led to state-of-the-art research coming out of much smaller, previously unknown labs.”

“Safety is (also) gaining awareness among major AI research entities,” the authors write, “with an estimated 300 safety researchers working at large AI labs, compared to under 100 last year, This is a promising sign when it comes to AI safety becoming a mainstream discipline.”

The spike in interest in AI continues on a daily basis. One of the latest developments is a report that Sound Capital, an Los Angeles-based venture fund headed by the actor Ashton Kutcher is planning to invest $200 million in AI ventures.

The question for the global tech industry is how the gap in research between China and the US will play out in investments in companies within the respective countries, and to what degree the gap will further exacerbate tensions between Washington DC and Beijing over sophisticated software development and the chips that process the software's output.