Estonia’s Research Council is comparing AI assessments of previously reviewed research projects with the evaluations made by human experts. The trial is intended to test the technology and its risks before any decision about using AI in real funding evaluations, while the University of Tartu is asking how fairness, transparency and accountability will be protected.
Watch Desk analysis
What happened
The Estonian Research Council, known as ETAG, told ERR News that it is testing large language models on projects that have already undergone human evaluation. The council says principal investigators checked with their teams and no one objected to their projects being included. It is using selected platforms in closed environments, after finding that systems developed specifically for Estonia performed considerably worse.
The University of Tartu’s vice rector for research, Mari Moora, says the university has not been given enough detail about the experiment’s purpose, how AI relates to expert review, or how its results will be assessed. She also raised questions about confidentiality, bias, equal treatment and how an AI-influenced decision could be challenged. ETAG says examining such risks is part of the trial. It hopes to draw its first conclusions in about a month, before Christmas. Read ERR News’ account.
Why it matters
Research grants help decide which ideas and fields receive support. Even a trial using previously assessed projects raises practical questions: can a model’s assessment be compared fairly with expert judgement, and can researchers understand or challenge an outcome if AI later influences a funding decision?
ETAG’s account gives the trial a more specific scope than a plan to hand grant decisions to a chatbot. The council says it is comparing outputs against existing human evaluations and has not yet decided to use AI in real-world assessments. That distinction matters, as does whether the comparison can reveal uneven treatment across research fields and unconventional proposals.
Our read
This is a worthwhile experiment only if it tests more than whether a model can produce a plausible-looking assessment. A comparison with previous expert reviews is a sensible starting point; it is not, by itself, proof that AI would make funding decisions fairer or more reliable. The council’s promised first conclusions should say what was tested, what the model got wrong and how researchers’ work was protected. Otherwise, “closed environment” and “human benchmark” are reassuring labels doing a lot of lifting.
What to watch
- What ETAG publishes about the platforms, evaluation criteria and comparison with expert assessments.
- Whether the first findings show differences across fields or types of research proposal.
- How the council would handle confidentiality, human oversight and appeals before using AI in live funding decisions.
Discussion spark: Should a research council use AI in grant assessments if it can show the system matches expert reviews, or should funding decisions remain entirely human-led?
Sources and evidence
- Source update (17 November 2026, 07:22 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.