AI can help HE assess research more efficiently – Report

National guidelines and regulations are needed on the use of generative AI (GenAI) by universities in the United Kingdom to assess the quality of research for the allocation of public funding, found a study published on 1 December.

Measures must also be taken to prevent the tools being used for research assessment from falling into the hands of external parties, including ‘tech bros’, who might capitalise on the information and data contained in academic research, the report’s author said.

The study, led by the University of Bristol and funded by Research England, is the first to examine how higher education institutions are using AI to evaluate research quality and how AI might be used to save universities time and money.

It focused on the Research Excellence Framework (REF), whose outcomes influence how £2 billion (US$2.7 billion) in public funding is allocated for research. The REF is criticised for being overly burdensome and costly for universities.

The report, REF-AI: Exploring the potential of generative AI for REF2029, found some enthusiasm but also some scepticism among academics and professionals working in the sector about how AI should be used, highlighting the need for better oversight.

Lead author Richard Watermeyer, professor of higher education at the University of Bristol, said: “GenAI could be a game-changer for national-level research assessment, helping to create a more efficient and equitable playing field.

“Although there is a lot of vocal opposition to the incorporation of it into the REF, our report uncovers how GenAI tools are nevertheless being widely – if currently, quietly – used, and that expectation of their use by REF panellists is high.”

The last REF happened in 2021, and, following a review, changes to guidance for the next one, REF2029, are expected to be announced shortly. The total costs of REF2021 were estimated to be around £471 million (US$628), with an average spend of £3 million in each participating higher education institution. The costs for REF2029 are expected to be much higher.

What academic and professional staff think

The study included a survey of nearly 400 academics and professional services staff and asked how they felt about GenAI tools being used for various aspects of REF2029.

Watermeyer told University World News that there is never a time in higher education institutions “when academics are not, in some shape or form, thinking about the next exercise and the review process.

“Universities need to consider how AI can alleviate some of the work and what job it can do,” he said. “Can it help to review, or at least help to score research against the evaluation criteria set out with the REF guidelines for output and impact, for example?

“Another dimension is how we can use the tools to do a more accurate job in helping to predict how the panellists, who do the formal assessments, will judge our research submissions.

“I don’t think there was a university that we consulted whose staff didn’t think that the panellists themselves wouldn’t be utilising the [AI] tools because of the sheer enormity of the task and because it was good value for money and saved time.

“Do we really want to take away our best academics and scholars from their current research to review work they’ve already completed?”

He added that AI might even allow research evaluation to take place on an ongoing basis, providing a current and up-to-date view of the UK’s research ecosystem.

However, Watermeyer said, there existed variations in the way AI was being used within and between institutions for the purpose of REF evaluations.

Some universities were using the tools to gather evidence of their research impact in the wider world and to help craft related stories. In others, there was evidence of in-house tools being developed to streamline REF processes or GenAI being used to review, assess and score their research.

Pro vice-chancellors’ views

The study used focus groups and interviews with 16 university pro vice-chancellors.

Some were positive and regarded AI usage as inevitable: “This is the future…We need to lead into it…We’ve got to understand it and experiment a bit”. And, “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”.

However, other feedback expressed doubt and mistrust and raised the possible limitations of AI: “I do just wonder if we’re just in a bit of an AI bubble at the moment, where we think everything’s going to be driven by AI, and suddenly over the next six, seven years, actually, we’re going to have a much greater clarity of what the limitations of AI are.”

Another respondent commented: “We don’t trust them enough in order to really start backing away from conventional tools and supplementing them with AI tools…there are still very large numbers of administrators and researchers who have had no real experience of AI yet.”

Watermeyer said: “Our findings show that opposition to the tools is concentrated among certain academic disciplines, chiefly arts and humanities and social sciences, and non-users. Professional services staff tend to be much more enthusiastic about the potential of GenAI for REF, as are those in post-92 institutions, which have considerably less resources to devote to REF processes.”

He told University World News that the study signalled the “first step to pull the use of AI out of the shadows”, even though concerns remained about aspects such as ethics and transparency.

“We know AI is being used for this purpose, but it’s not being used in regulated ways, and there is no clear-cut governance around it,” he added.

“Institutions are largely operating in the dark, and while they might have codes of practice statements, there is no requirement for them to disclose their AI use.

“What we really need now is engagement within the sector, with funders, regulators, the research councils and all key stakeholders. This is really quite urgent.

“We can’t roll back from this technology. We can’t put it back in a box or switch it off. So, there is a need to consider whether we need a standardised tool that every institution can plug into.”

Eliminating need for consultants

Watermeyer said the use of AI would remove the need for adjunct industries, such as consultants, “who feed off universities” during the REF process. The money saved could be reinvested into research. “But you would need acceptance and trust in the tools, and we don’t have that yet,” he added.

This was something the sector needed to develop and own. “There is a real concern that if this were to fall into an external party, including the tech bros, then questions around trust and acceptance become harder, when there are already low levels of this within some institutions.

“There’s also a moral dimension here of organisations capitalising on things like data privacy and security, exploitation and manipulation of data.”

The report made a host of recommendations, including that all universities should establish and publish a policy on the use of GenAI for research purposes, encompassing REF; that relevant staff should receive full training on the responsible and effective use of AI tools; and that appropriate security and risk management measures should be implemented.

It also called for robust national oversight, comprising sector-wide guidance on usage for REF29 and a comprehensive REF AI Governance Framework. To help achieve equitable access to the technology among all higher education institutions, it advised that a shared, high-quality AI platform for REF could be developed and made accessible to all institutions.

Co-author Tom Crick, professor of digital education at Swansea University, said: “The current disparate use of these tools needs clearer co-ordination into a standardised framework through which the sector can adopt and develop open, transparent and responsible practice.

“This will be critical to mitigating inequitable practice by higher education institutions and across disciplines, and for even possibly levelling the playing field for this and future national research assessment exercises.”