Will AI grant reviews make science more conservative?

What happened: A study found that GPT-4o, used to rank genuine grant proposals, was systematically less willing than human reviewers to fund proposals rated highly for novelty. Key numbers: The model assessed 139 proposals submitted in 2021 to the Austrian Science Fund and agreed with human reviewers' decisions in 60% of cases. The study was published in Scientometrics in August. Why it matters: The author says AI-assisted review could skew funding toward established ideas and miss breakthroughs, as funders including UK Research and Innovation explore AI assessment.

A study found that GPT-4o, used to rank genuine grant proposals, was systematically less willing than human reviewers to fund proposals rated highly for novelty.

The author says AI-assisted review could skew funding toward established ideas and miss breakthroughs, as funders including UK Research and Innovation explore AI assessment.