TL;DR
UK Research and Innovation has gone further than any other British funder in examining whether AI could help judge grant applications, saying last week it would explore “the safe use of AI-assisted assessment” to shorten processing times. Asked by Research Professional News, no other member of the eight-strong Research Funders Policy Group reported looking beyond eligibility checks. Automation is edging into the machinery that distributes UK public research money.
The loop that prompted it
That policy group shifted its stance in June, permitting AI in grant processing. UKRI is the one that acted on it.
The driver is volume, and the volume has an obvious cause. Stephen Curry, strategy director at RoRI (Research on Research Institute), observed that AI is itself inflating application numbers, so “it seems inevitable that funders would in turn resort to AI to deal with the resultant workload”. Machines writing the bids, machines reading them.
His caveat was firm: whatever UKRI deploys to speed assessment must be “demonstrably robust and trustworthy”.
What UKRI has actually committed to
Less than the headline implies, so far. The agency intends to look backwards before it looks forwards, running a trial alongside the UK Metascience Unit that measures an AI system’s verdicts on historic bids against the outcomes human assessors actually reached. It has also pledged to keep human judgement central to funding decisions.
A UKRI spokesperson framed the trade-off carefully: AI brings efficiency gains but “risks increasing demand on an already highly competitive and strained system”, and progress will need coordinated work across funders, institutions and applicants.
Everyone else is waiting
Two charities in the group, Cancer Research UK and the British Heart Foundation, each ruled it out for the moment. The Royal Society has yet to decide. Nuffield expects to revise its own position before the year ends.
Leverhulme, which sits outside the group, draws the sharpest line: its policy bars the technology from condensing a bid or producing a review outright, permitting it only to polish how a human-written assessment reads. The Royal Academy of Engineering is working further up the pipeline instead, on eligibility checks and on identifying who to invite to apply.
One point commands consensus: confidential applications should not be fed into generative AI tools.
Looking forward
Researchers already report not knowing what AI use they must declare. Once assessors use it too, disclosure becomes a two-sided question, and applicants will reasonably want to know what read their bid.