TL;DR

Ottoline Leyser, who ran UK Research and Innovation until June 2025, has told Research Professional News that AI should have only a limited role in deciding what gets funded, because allocating public money means judging what will “enrich humanity”, and that needs people. Her intervention, published on Wednesday 14 October, comes as UKRI trials AI-assisted scoring against human decisions and after a UKRI-funded network reversed an AI triage that had screened out about half of a call’s applications. The academics quoted alongside her agree on one point: whoever adopts AI sifting first, funder or university, forces everyone else to follow.

Leyser’s argument

Leyser, who now holds the regius chair of botany at Cambridge, says AI appeals to those who want assessment to be wholly objective, and that wish is misplaced: “the idea of a fully objective assessment that delivers what is needed is a non-starter”. A healthy portfolio funds projects that matter for many different reasons, which requires “subjective decisions”. AI might help check aspects of a proposal, she allows, “but ultimately we are allocating public funds to support work that will enrich humanity in many different ways. We need humans to decide what is enriching.”

Where UKRI stands

UKRI said in September that it would make grant assessment quicker using automation alongside staff expertise, and its chief executive, Ian Chapman, has promised a system that is “faster, fairer and ready for the future” with expert human judgement at its core. Its policy bars generative AI from conducting application assessment. It is, however, running a retrospective trial alongside the UK Metascience Unit, comparing AI scores with what human reviewers decided.

The arms-race worry

Several researchers told RPN the pressure now sits with universities, which funders are asking to cap how many bids their staff submit. Stephen Curry of the Research on Research Institute expects universities to turn to AI for that sifting, while warning that current tools cannot judge proposals in a way researchers trust and that a university caught doing it faces reputational risk. Nottingham’s Adam Golberg says internal demand management depends on the broad trust of researchers, which AI use would undermine. UCL’s Robert Insall predicts an “arms race”: if UKRI’s assessment uses AI, universities will tune bids to please it.

Mike Thelwall of Sheffield, whose preprint found AI could play a useful role, takes the opposite view, arguing institutional experiments are lower-risk than national ones. Plymouth’s John Terry sets out the test any sift should pass: “scepticism first, then validation against human judgment at scale, then checks for gaming and bias”.

Looking forward

In our view the research debate mirrors this week’s police watchdog warning about AI use without clear policy: the risk is tools arriving before the rules. UKRI has not said when the Metascience trial will report, and that timetable now matters to every research office weighing its own AI sift.