FST JOURNAL

On Thursday 12^th March 2026, a group of scientists, policy makers, and industry experts gathered to identify practical actions that could strengthen UK Research in an era of rising AI technology and rapid technological change.

The roundtable was organised by Elsevier in collaboration with The Foundation for Science and Technology.  Meeting at the Royal Society, the group started the first in a series of international discussions exploring how artificial intelligence (AI) is reshaping research systems worldwide.

Elsevier shared findings from its ongoing Researcher of the Future initiative, based on insights from more than 3,200 researchers across 113 countries. The session examined the UK-specific data set, discussed institutional and cultural responses, and explored how AI might influence productivity, creativity, research integrity, and policy frameworks. 

Attendees included:

• Tracey Brown, Director, Sense about Science
• Gavin Costigan, (Former) Chief Executive, Foundation for Science and Technology
• Helen Coulshed, Associate Vice Dean Education and Senior Lecturer in Chemistry Education, King’s College London
• Camilla D'Angelo, Policy Manager, CASE
• Esra Erkal, Executive Vice President, Global Communications, Elsevier
• Catriona Firth, Associate Director, Research Environment, Research England
• Pallab Ghosh, Science Correspondent, BBC
• Laura Hassink, Managing Director for STM Journals, Elsevier
• Sudesh Kumar, President of AIRTO and Vice-President Health, University of Warwick and AIRTO 
• Sarah Main, Vice President Academic and Government Relations UK, Elsevier
• Geoffrey Neale, Lecturer in Multifunctional Composite Structures, Cranfield University
• Paul Nurse, President, Royal Society
• Lawrie Phipps, Director of External Policy Development, Jisc
• Hetan Shah, Chief Executive, British Academy
• Vivienne Stern, Chief Executive, Universities UK
• Dominic Tate, Deputy Director of Library and University Collections, University of Edinburgh 
• Lesley Thompson, Vice President, Funders, Global Business Development, Elsevier
• David Willetts, Chair, Foundation for Science and Technology
• James Wilsdon, Exec Director, Research on Research Institute, UCL

Adoption is rising, but confidence in lagging

Elsevier’s most recent findings from their ‘Researcher of the future’ initiative highlighted that around half (52%) of UK researchers now use AI tools in their work, mirroring global adoption levels. Uptake has grown sharply (by over 20 percentage points in a year), but UK researchers are notably more cautious than their international peers. This caution reflects a culture that focuses on integrity and quality. UK researchers are notably more cautious when it comes to trust, ethics, and the role of AI in advancing discovery. According to the report’s findings, 11% of UK researchers say AI is trustworthy, and only 5% believe AI tools are ethically developed. 

Roundtable participants agreed that AI should be used as a productivity tool, not a creative partner in research. AI is widely valued for efficiency gains—for example, data analysis, and automating repetitive processes. However, participants noted that AI often produces the “obvious answer” quickly, which is useful but may limit deeper or more novel inquiry. The group stressed that AI can answer questions, but cannot reliably ask the right ones, which is central to scientific breakthroughs. Concerns were raised that AI-generated material can appear plausible but lacks originality or true insight, potentially contributing to a homogenisation of research outputs. Importantly, this dynamic risks reinforcing conservative thinking and reducing the diversity of research voices and ideas.

AI caution as a competitive strength?

"Skepticism has value if it helps us build better and more responsible systems" - Roundtable participant

Although it was noted that caution around AI could hinder innovation, several voices argued that the UK’s careful, ethics-led approach could become a distinctive advantage internationally, positioning it as a hub for high-integrity, quality-focused research.

For example, rather than chasing scale and speed, the UK could emphasise “responsible excellence”—balancing innovation with rigorous oversight. However, achieving this would require sustained investment in skills development, data quality, and public trust in research.

Training emerged as a top priority, extending beyond students to faculty, supervisors, and research professionals. Only 13% of UK researchers report receiving sufficient training in AI tools—much lower than the global average which stands at 27%. A recurring theme was the absence of structured institutional support for researchers using AI. Participants noted that there is a growing “assessment gap”, where AI use is disclosed but evaluators lack the capability to judge it effectively. Governance arrangements and guidance also remain inconsistent across institutions.

There was concern that the overuse of AI (e.g. for literature reviews) may undermine researcher development, especially for early-career researchers. Participants said that the goal is to enable researchers to use AI thoughtfully—to prompt deeper analysis and reflection, not replace it.

Some individuals around the table said that researchers are not just under-trained; they are concerned about retrospective accountability (being judged later for current AI use. Many researchers are also learning through experimentation, leaving them uncertain about ethical boundaries, data management, and disclosure expectations. Agreement was heard around the table for centralised, cross-institutional guidance rather than fragmented local policies. They saw libraries and research support offices as critical infrastructure for research literacy and expertise for critical thinking, ethical use, and data governance.

Trust, Integrity, and Data Bias

Trust and research integrity were central concerns. Participants noted risks from biased or incomplete datasets, opaque AI models, and unverified outputs. There were calls for transparency about how AI is used in research design, data analysis, and writing and the discussion highlighted the need to differentiate between types of AI—ranging from assistive tools to autonomous systems—with distinct ethical and governance implications.

Several participants around the table said that real-world risks were already visible (e.g. biased health outcomes from unrepresentative datasets). There was agreement that greater openness about negative or null results was important to counteract publication bias and strengthen research credibility.

It was noted that broader research system challenges including funding constraints, shrinking time for research, and incentives that reward volume and polish—shape how AI is being used. Participants said that this creates a system-level challenge, not just an individual behaviour issue, requiring redesign of evaluation processes.

For example, funders are already seeing sharp increases in application volumes, likely driven by AI-assisted proposal writing. AI enables higher throughput of proposals and publications, potentially making it harder to identify high-quality, original research.

Participants warned that this could lead to an oversupply of “well-written” work with limited novelty, while genuine breakthroughs risk being crowded out.

Next steps: Action and Policy

The roundtable concluded with a shared sense of urgency to move from uncertainty to clear cross-sector principles. It was repeatedly stressed that AI is not a single category and includes very different tools and capabilities. Therefore, policy and governance should be use-case specific, rather than a one-size-fits-all.

 Recommended areas for action included:

• Need for flexible frameworks that evolve with technology, rather than static regulation.
• Integrating training on AI ethics and data management at all career stages.
• Creating mechanisms for sharing negative findings and improving data transparency.
• Encouraging collaboration across academia, funders, and publishers to ensure consistent standards.
• Supporting research on AI’s social, cultural, and ethical impacts alongside technical developments.

Conclusion

The discussion reflected both optimism and unease about AI’s growing role in research. UK researchers are adopting AI at pace but remain cautious—an attitude that could become a strength if it leads to robust, trustworthy, and creative science.

The path ahead will depend less on promoting AI adoption and more on building competence, confidence, and clarity: ensuring that tools accelerate high-quality research without eroding the values and rigour that underpin the UK’s scientific reputation.