TL;DR

Half of the 179 proposals submitted to a UKRI-funded research call were turned down after a first sift carried out entirely by AI, Times Higher Education reports. The call was CRANE’s phase one pilot, which Oxford hosts with money from the EPSRC, UKRI’s engineering and physical sciences council. UKRI said this month it would explore AI-assisted assessment with safeguards, and this case shows how applicants react when it happens.

How the sift worked

According to the email sent to unsuccessful applicants, each bid was marked by several AI models on seven measures. Among them were value for money, feasibility, relevance, and quality and novelty. A second pass with different models was run to limit the quirks of any single system, and the two sets of scores were merged into one ranking. Human reviewers only saw the upper half.

The call documents had warned that AI tools might be used in the initial assessment, with reviewers accountable and the award panel making final ranking decisions.

The reaction

John Stott, an astrophysicist at Lancaster University, said it was “depressing that I wrote something that no one ever read before it was rejected”. He told THE he would think twice about applying to schemes assessed this way.

Two Lancaster forensic linguists, Claire Hardaker and Georgina Brown, were also rejected. Their proposal was partly about the dangers of placing too much trust in AI that sounds fluent. “There’s probably a sort of irony-is-dead joke in there somewhere,” Hardaker wrote on LinkedIn.

Margot Finn, a professor at UCL, said funders using AI even for an initial sift need to be far more careful and transparent in how they communicate outcomes.

Where UKRI stands

UKRI had been approached for comment at the time of publication. As we reported earlier this month, the funder is dealing with a surge in applications that AI is partly driving. Its proposals include in-person interviews for riskier awards, more triage by its own staff, and a trial that would compare AI verdicts with decisions already made on past applications.

Looking forward

The CRANE case sits one step removed from UKRI: a funded network, not the funder, ran the sift. But it previews the argument any public body will face when it automates the first cut of a competitive process, whether that is grants, procurement or recruitment. The complaint from Lancaster was not about rejection itself but about nobody having read the bid. Businesses applying for Innovate UK or similar funding will be watching which way UKRI goes.