AI Model Matches Human Grant Decisions 60% of Time
A study found GPT-4o agreed with human reviewers on Austrian Science Fund funding decisions in most, but not all, cases.
Updated on Nov. 3, 2026 in Artificial Intelligence
Live Poll
Would you trust AI to help decide which scientific research gets funding?
Researchers tested GPT-4o on 139 grant proposals submitted to the Austrian Science Fund in 2021 and found it agreed with human reviewers' funding decisions in 60% of cases. The study was published in Scientometrics in August.
Why it matters
The findings bear on whether generative AI could support research funding decisions, a process that shapes which science gets resourced. The model was less willing than reviewers to fund proposals rated highly for novelty, raising questions about how such tools treat unconventional ideas.
The model assessed 139 proposals and matched human funding decisions in 60% of cases. Whether that level of agreement is adequate for real funding decisions has not been established.
The players
Diogo Machado
Machado leads the data-science team at Technopolis, a science-policy consultancy in Brighton, UK, and conducted the study.
UK Research and Innovation
UK Research and Innovation is the UK public body that funds research and has announced a controlled trial of AI-assisted grant assessment.
Wellcome
Wellcome is a London-based biomedical research charity that requires its reviewers not to input grant applications into generative AI tools.
The details
Diogo Machado, who heads the data-science team at Technopolis in Brighton, UK, asked GPT-4o to compare proposals in pairs within the same research field and answer with A or B only. Proposals with the most pairwise wins were ranked highest for funding.
Timeline
2021: Grant proposals were submitted to the Austrian Science Fund.
August: The study was published in Scientometrics.
Last month: UK Research and Innovation announced it is exploring AI-assisted grant assessment.
Market Landscape
UK Research and Innovation's announced trial of AI-assisted assessment places this study within a broader move by funders to test automated evaluation of grant proposals. Wellcome's ban on reviewers using generative AI shows that major funders are taking divergent positions on the same technology.
For researchers applying for grants, the findings suggest AI tools may increasingly appear in funding review processes. Funders are setting different rules on reviewer use of generative AI, so applicants and reviewers should check each funder's current policy.
The takeaway
Automated assessment may match human judgment on many decisions yet still undervalue novel ideas, so funders testing these tools should check how they treat unconventional proposals. Researchers should follow funder policies on AI use in review and submissions.
Further reading
For more on the tools reshaping research, see our Artificial Intelligence coverage.
Source note: This article includes information reported by Nature.
Live Poll
Would you trust AI to help decide which scientific research gets funding?
