̽Ƶ

Most academics strongly opposed to using AI in REF 2029

While new technologies seen as ‘game changer’ for national-level research assessment, study finds vehement opposition, particularly among humanities scholars

Published on
December 1, 2025
Last updated
December 1, 2025
Patrick Jack
Dr Who (Tom Baker) looks concerned as a cyberman grabs his shoulders. To illustrate that most academics strongly opposed to using AI in REF 2029.
Source: Frank Barratt/Getty Images

The vast majority of academics do not want artificial intelligence used to assess the next Research Excellence Framework (REF), according to a new report.

Ahead of changes to guidance for the REF 2029 expected to be announced this month, the study also found that more senior university staff are generally more supportive of using AI and not succumbing to a “moral panic” around its use.

The report, led by the University of Bristol and funded by Research England, found that some universities are already using generative AI to assess the quality of their research.

But it showed that there was wide variation in how it was being used, with some universities using AI tools to gather evidence of real-world impact, and others building new tools to streamline REF processes or assess their research.

̽Ƶ

ADVERTISEMENT

In a survey of almost 400 academics and professional services staff conducted as part of the study, the majority of respondents strongly disagreed with all aspects of using AI in the REF.

Two-thirds strongly disagreed with the idea that universities should use it to support internal assessment of REF research outputs, and three-quarters strongly disagreed with its use by REF panels in assessing outputs.

̽Ƶ

ADVERTISEMENT

A further 86 per cent disagreed with AI being used to support the assessment of impact case studies by REF panels.

With funding required for REF 2029 likely to be even higher than the £471 million spent in 2021, lead author Richard Watermeyer, professor of higher education at Bristol, said AI had the potential to alleviate some of the burden.

“GenAI could be a game-changer for national-level research assessment, helping to create a more efficient and equitable playing field.”

Some respondents in the report highlighted the advantage of using AI tools to handle the “drudge dimensions” of some REF preparations, and in reducing the huge burden placed on academics in reviewing outputs for REF institutional selections.

However, Watermeyer said GenAI offers no complete solution and acknowledged the “vocal opposition” the survey revealed to the incorporation of it into the REF.

̽Ƶ

ADVERTISEMENT

“It could also create new bureaucratic challenges of its own, including establishing new requirements and protocols for its appropriate use.”

The report found different views among the 16 pro vice-chancellors it interviewed, with some urging caution amid an “AI bubble” until it becomes clearer what the limitations of the technology are, and others concerned around how much they can trust AI.

But another said: “I do think that just to put our heads in the sand and say it’s not going to happen or not on our watch I think is very limiting of what the future might look like…I think there’s a lot of moral panic.”

̽Ƶ

ADVERTISEMENT

Watermeyer said opposition to AI is concentrated among certain academic disciplines, in particular arts and humanities and social sciences, while professional services staff tend to be much more enthusiastic about its potential.

Steven Hill, director of research at Research England, said the findings offer “both a caution and a call to action”.

“It warns against haste and complacency alike, while inviting the sector to lead with principle, collaboration, and well-informed critique. With the right safeguards, the integration of GenAI can help us uphold excellence, fairness, and trust in the assessment of UK research.”

Authors recommended that all universities should establish and publish a policy on the use of GenAI for research purposes, that staff should receive full training on the responsible and effective use of AI tools, and for robust national oversight.

̽Ƶ

ADVERTISEMENT

The majority of interviewees cautioned that that without a standardised tool across the sector, the use of GenAI in REF preparations will “bake in structural inequalities for poorer resourced institutions”. So the report also called for a shared, high-quality AI platform for the REF to be developed and made accessible to all institutions.

patrick.jack@timeshighereducation.com

Read more about
Read more about: 
Research
UK research policy
Research Excellence Framework (REF)
Artificial intelligence in higher education

Register to continue

Why register?

  • Registration is free and only takes a moment
  • Once registered, you can read 3 articles a month
  • Sign up for our newsletter
Please
or
to read this article.

Related articles

A robot game advert. The amazing REF robot “always gives the right answer”, with a robot pointing to scores of 1 to 4. To illustrate whether generative AI can improve the REF.

Could generative AI improve the REF?

The task of reading and rating the thousands of outputs submitted to the UK’s Research Excellence Framework is notoriously Herculean. Could AI ease the burden – or would its use undermine the whole point of having REF panels? As Jisc consults on that question, four writers offer their views

By Caroline Ball
5 June
Man searching for money in a maze, as an illustration of the difficulties involved in securing funding through the Research Excellence Framework.

Where now for the REF?

The Research Excellence Framework has been postponed for three months ‘to take stock and ensure alignment with the UK government’s priorities and vision for higher education’. But how radical should the changes be? Should there be any at all? Is it time to start again? We present five very different views

By Cat Davies
26 September

Reader's comments (20)

#1 Submitted by graff.... on December 1, 2025 - 1:04am
Is it the "report" or this write up that misses the fundamental issue? It is NOT "using AI" yes or no but how and in what ways AI is used knowledgeably and responsibly. It is hard to use any device now without "using" some form of AI.
#2 Submitted by ... on December 2, 2025 - 7:01am
If it saves money and thus reduces the number of colleagues at risk of redundancy (myself included!) over the next few years then I am all for AI being used. One suspects a great deal of REF is now political in any case. I wonder if the 400 academics were asked the question: "If the use of AI in REF would help prevent you from being made redundant in the next few years would you support it's use?" how many would respond in the affirmative?
#3 Submitted by ... on December 2, 2025 - 7:08am
I just wondered who the 400 academics selected were and how they were selected? Are they colleagues already involved in the exercise for example? Well they are not going to vote for AI are they? They have an interest in maintaining the status quo.
#4 Submitted by ... on December 1, 2025 - 7:12am
"With funding required for REF 2029 likely to be even higher than the £471 million spent in 2021". Well that is the question. VCs are constantly complaining about the underfunding of the sector with the decline of the fees in real terms and the supposedly diminishing recovery of research overheads. But no-one really seems to be raising the issue of this huge sum of half a billion quid spent by the sector on what is after all not core teaching and research but a simply quality assessment. Could it not go to UKRI to boost research grant awards? TEF doesn't cost anything near this much and teaching is the major source of university funding. Much of this spending is going not on the actual implementation of the excercise, UKRI spends around £17 million on that I think, but on the vast, surro8=nding, collateral paraphenalia of expensively staffed subject panels, internal research offcers and administration (the endless mocks REFS etc) involving a whole cadre of dedicated people. If AI (and it may be a big if) can assist in reducing this paraphenalia or simplifying it somewhat, then we really should be using it to do this urgently. REF has become a bureacracy within a bureaucracy, a firm within a firm, as they say. Of course there will be huge resistence to the use of AI from those who operate the current system for various reasons (some more compelling than others), but when we are told that the sector may be losing c. 10,000 jobs year on year, as reported in THES, then this extraordinary level of funding for non-core activity is verging on obscenity, in my view, especially if the funding is uiltimately deriving from the fees levied on individual students (now over £10k).
#5 Submitted by ... on December 2, 2025 - 8:26am
REF2021 cost £471m over 7 years, and distributed more than £15bn in QR funding. Perhaps you think that £15bn should have been distributed on a whim, no-strings-attached?
#6 Submitted by p.... on December 2, 2025 - 9:16am
"Perhaps you think that £15bn should have been distributed on a whim, no-strings-attached?". Obviously not, clear reductio ad absurdum here. But maybe there is a more rational and much less wasteful way of doing this, especially in the current financial crisis with VCs predicting jobs losses of 10k year on year, costing less than £471 million (a big chunk of the £15 bn to be dispersed) but more rigorous than on simply "a whim" ? The use of AI might be one way of bringing the cost down? Even at half the cost of c. £250 million it would still be very expensive, but that would be a start? Indeed perhaps you think the exercise should cost even more? £471 million is the costing for REF2017 which was substantially more than it predecessor, we have not yet got a realistic estimate for REF2029 but it's bound to be substantially more. What's your notional view of what this exercise should reasonably cost (in either real terms or as percentage of the £15bn to be dispersed), or do you think there should be no limit? I don't think anyone is really in control of this process at the moment.
#7 Submitted by ... on December 2, 2025 - 10:06am
I love the idea that because we are paying this in installments over seven years (as with the previous and the one before that) then it somehow represents value for money risible as a justification. And, in my view though I may be mistaken, there is widespread scepticism about the exercise and its ideological and instrumnetalist nature. Those conerns were exacerbated with the addition of Impact as a category, the insistence on open access publication, and then the imposition of EDI categories as critieria for excellence! The only people I know who defend it are those who receive some form of clear benefit from its operation. But to my mind, it has become a gravy train for some.
#8 Submitted by oliver.... on December 1, 2025 - 8:05am
Teaching may be the current source of funding, but research is what builds universities' reputations. Teaching is a function of this. The model has clearly failed anyway, and AI slop won't help.
#9 Submitted by ... on December 1, 2025 - 9:20am
Yes exactly and with it goes the "prestige" that the students seek, so REF preparation spending actually pays off with recruitment for some at least? But you could build a major new state of the art regional hospital for £500 million. AI might help reduce costs. I donpt have too much faith in the current set up which, in my view, is largely political rather than an objective assessment of research quality.
#10 Submitted by p.... on December 1, 2025 - 7:49pm
So REF 2029 will cost probably somewhere over half a billion in real terms? The last REF 2021 cost about £4.7 bn in real terms (a huge rise on the previous one), so the sector will have spent in real terms in this decade probably £1 billion on the REF? And we are told that there will be maybe 10,000 job losses year or year in HE? Is someone having a laugh here? Have I got this wrong?
#11 Submitted by p.... on December 1, 2025 - 7:53pm
"At £471m, or around £67m a year over the seven year REF cycle, its annualised cost (of REF2021) was greater than the annual income of the University of Chichester. The septennial fee is more than the total annual funding distributed by OfS for student access and success. If Research England, its counterparts in the nations, and UK universities wished to get into football they could buy a West Bromwich Albion sized club every year instead of running REF" [or a state of the art major regional Hospital to serve the NHS and the public.
#12 Submitted by ... on December 1, 2025 - 8:43pm
They talk about the system being marketised and with some justice, but the whole REF establishment has become like some Soviet era centrally-planned, wasteful, bloated, bureaucratic nightmare of which the whole Impact agenda entirely fits perfectly. It acts to perpetrate itself by making more and unecessay work to justify its existemce. The REF admin staff from PVC Research to Subject Panels experts etc, function analogously to the old Soviet Communist Party now as an elite section of the profession, superior and cut off and remote, who only talk to each other in their own and other instritutions, speaking their strange languages about "outputs" and other jargon, but not to the ordinary workers who do the teaching. Is this a fair analogy?
#13 Submitted by ... on December 2, 2025 - 6:52am
Well yes exactly, and the highly weighted EDI sections of the exercise are clearly ideological rather than about research quality. These could be done via AI I would think as little judgment is needed just formulaic responses?
#14 Submitted by ... on December 2, 2025 - 6:55am
If it saves money and thus reduces the number of colleagues at risk of redundancy over the next few years then I am all for AI being used. One suspects a great deal of REF is now political in any case. The sector is a shambes and badly managed. Let's hope Sir Pat and Ed Peck will grab it by the scruff of the neck and impose some sanity.
#15 Submitted by ... on December 2, 2025 - 6:45pm
One problem is that there is little real advocacy for the sector and its problems. The leaders of HEFCE/UKRI and university VCs are very much a club of like-minded people in my experience (we probably all have anecdotes we coukd share), so no-one is really going to call out mistakes, bad policy decisions, and mismanagement. They will blame all worries on 2006 capped fees not keeping up with inflation. In my opinion (and others may well disagree of course), the way RAE/REF burgeoned and expanded into this resource hungry monster is another example of the larger sector mismanagement over the last 10 years. It did not always used to be like this. In the easrly fars of RAE, VCs used to seek legal review in many cases.
#16 Submitted by roy.... on December 2, 2025 - 8:55pm
What this article is really saying: 1) the quantity of work involved in the REF is so enormous and so tedious that humans shouldn't have to do it, 2) academics cannot be trusted to perform this work objectively, and 3) the assessment protocols are so poorly constrained that the assessments may need to be tweaked to get sensible results, such that the REF will need the capacity to redo aspects of the assessments via AI. And the best that anyone can come up with to justify this is that it will "save money". There is absolutely no evidence that I am aware of that an AI tool exists that is able to understand creativity or is able to overcome biases programmed into it, certainly not a tool fit for this critically important task, through which thousands of human careers will be affected. Any move in this direction needs to be benchmarked and approved by the people affected, and I am sceptical in the extreme about working academics trusting such a move at present, with good reason.
#17 Submitted by p.... on December 3, 2025 - 7:07am
I think the point about AI saving money is satirical and slightly mischievous. AI would not be able to spot themirody I fancy!
new
#18 Submitted by ... on December 3, 2025 - 8:51am
This is an irrelevance really. To be Devil's Advocate: REF measures several things that are not research quality. It's used as a mechanism to drive modish HE policy which have nothing to do with quality or excellence (impact, open access, EDI etc), it's heavily gamed and thoroughly dominated by the RG in terms of its subject panels who make sure the criteria and operation are congenial for their interests as the most research intensive institutions. And crucially, it's not a funding exercise as some seem to beleive and if the results are a bit awkward, then the actual funding formula just changes or is massaged to allocate the research funding where they want it to go (in terms of discipline and geography) in the first place. It will, of course, always throw up a few exceptional outliers, notable succeses and unexpected failures, to pretend the system is not entirely staged and to 'encourage les autres' and let them know who holds the whip hand, bit apart from that it's very predictable as it meant to confirm the status quo of research funding and show that the current funding patterns are by and large correct and justifiable, The tragedy is that it costs so very much now to implement (in both resource and time) which could be spent elsewhere in these difficult circumstances of job lossess and rising class sizes and, if it's really for show, then AI could do it as what difference would it really make? Of course this is te reason why they will never let AI anywhere near it in the first place and the agenda will be resisted as it shows the absurdity of the system. Am I being too cynical? But it does often strike me as odd that we are trained as academics to analyse structures and belief systems and question and challenge received authority subjecting statements to rational enquiry, yet so many accept unquestioningly the organization, expense, politics and structures of REF as a purely rational, disinterested measurement of "research quality", itself a problematic term.
#19 Submitted by s.... on December 3, 2025 - 9:18am
By definition, if the use of AI is to assess quality then it makes a mockery of the idea of peer review. AI is already being used to crunch data in areas like insurance that are not transparent and hard to challenge. Any honest equalities review would have to acknowledge the potential for unequal treatment. It isn't clear though how they are thinking about using AI. There are a lot of different ways to use AI. It is disappointing that we seem to have less information about potential approaches than I discuss with my students with regard to their papers. Which seems frankly indicative of why a lot of problems are happening in the sector. Carelessness and lack of respect for the work staff actually do.
#20 Submitted by ... on December 3, 2025 - 10:23am
"use of AI is to assess quality then it makes a mockery of the idea of peer review". Well yes exactly, but there are real issues about our unexemined concept of "peer review" and how it operates in the REF. Research entered into REF has already been expert peer reviewed several times over before publication (and the research otself peer reviewed during grant applications). In REF Individual peer reviewers often disagree substantially about the research quality of an output, I am told, before they come to a final rating. And no paper trail of the decisions exists concerning how individual judgments were made, so it is an act of faith in my view or at least partakes of aspects of faith. So there is an enormous amount of subjectivity involved on REF peer review. It would certainly be very interesting to run a pilot (like the expensive Impact pilot of several years ago) using AI prgrammes to assess say a UoA from REF2021 and see how the scores differed if at all. I think that's the only way we could know for sure?

You might also like

Sponsored

Featured jobs

See all jobs
ADVERTISEMENT