China-US divide extends to AI, both in research and real life, study finds

Stark separation of Chinese and American citation networks between 2012 and 2021, says paper by student researchers in the UK and US

Anecdotal evidence also suggests Chinese and Western academics will not eat or talk together at machine learning conferences

Reading Time:3 minutes

American institutions cite Chinese papers less often than European ones, study shows. Photo: Shutterstock

Published: 8:03pm, 6 Dec 2022Updated: 8:33pm, 6 Dec 2022

Chinese and American institutions are collaborating less on machine learning, not only in terms of academic citations but social interaction as well, researchers in Britain and the US have found.

The findings come as a surprise because analysis by two doctoral students of articles submitted to a noted machine learning conference showed the US had the highest number accepted, followed by China, and it is generally accepted that there is a lot of give-and-take between the top two.

The students, Zhao Bingchen at the University of Edinburgh and Gu Yuling with the Allen Institute for Artificial Intelligence in Seattle, said the rankings might hide growing divisions.

Zhao and Gu analysed the citation data from last December’s NeurIPS (Neural Information Processing Systems), a noted academic conference on AI. They found that Chinese institutions habitually under-cited work from the United States and Europe, and vice versa. Under-citation indicates a deliberate attempt to avoid highlighting related work.

Their paper was released last month on the arXiv website, an open-access repository of academic preprints before peer review, and will be discussed virtually at a NeurIPS 2022 workshop on Friday.

To measure the segregation quantitatively, Zhao and Gu built a citation graph based on all AI-related papers from 2012 to 2021 published on the NeurIPS website.

The citation data was acquired through the AI search engine SemanticScholar developed at the Allen Institute for AI, with corresponding author information obtained through the data mining tool AMiner, created by Tsinghua University in China.

Select Voice

Choose your listening speed

Get through articles 2x faster

1.25x

250 WPM

Slow

Average

Fast

00:0000:00

1.25x