Chinese scientists are adopting artificial intelligence (AI) tools in research at a faster rate than their counterparts in the US or European Union (EU), according to a recent analysis published by the European Commission.
The resulting working paper noted that the “overall effect of AI on both novelty and impact” was strongest in China, followed by the US, and then the EU.
Titled Artificial intelligence in science: promises or perils for creativity?, it reviewed about 3 million documents, including peer-reviewed journal articles, published between 2000 and 2022 on the platform OpenAlex. Researchers found that the number of papers using AI aids rose sharply around 2010 – largely due to advances in deep learning – and that China had led the trend since 2016.
The number of Chinese scientific papers in its dataset published using AI tools totalled more than 25,000 in 2022, the study’s researchers found. That compared with 15,000 from the EU and 12,000 from the US.
[See more: For the first time ever, there are now more top scientists in China than the US]
As its title suggested, the study highlighted advantages and disadvantages of AI’s use in academia. One of its concluding remarks was that “AI could just as easily hinder scientific progress as accelerate it.”
It described AI as “a powerful ‘method of invention’” across disciplines that could “help pull science out of the productivity slump of recent decades,” predicting that its biggest impact was likely in fields where knowledge was “sparse and complex, meaning more fragmented and disconnected ideas.”
In terms of perils, the study described AI as falling short when it came to the “genuinely creative intuitive thinking” human brains were capable of. It also called AI out on a lack of vision. “The use of AI in science … will move from a luxury to a necessity,” it noted. “But science is not just about answering questions, it is also about asking the right ones.”
The study said that “securing strategic advantage in AI research” was a shared priority for most governments, but warned that very few had concrete measures in place to support AI in scientific research – highlighting a need for improvements in targeted policymaking.