Safeguarding integrity in AI-enabled research
While AI can be a force for good in accelerating research, strong guardrails are important for upholding ethics, transparency and integrity. A THE webinar, held in partnership with Elsevier, brought together a panel of experts to explore challenges of maintaining research integrity and establishing governance frameworks in the era of AI.
Thorsten Joachims, vice-provost for AI strategy at Cornell University in the US and director of the Cornell AI Initiative, spoke about the disruptions higher education is currently facing. “One of them is technological change and AI, and the other one is the erosion of public trust, and that is both in research as well as in our educational mission,” said Joachims.
“AI can severely speed up and strengthen the already-happening motions that attack trust in science,” said Anita de Waard, vice-president of research collaborations at Elsevier. Adopting reproducibility norms and sharing research data can enable a culture of transparency and verification.
“We’re all seeking to build environments in which trustworthy research can be conducted,” said Tilo Böhmann, vice-president for research at the University of Hamburg in Germany. “Reproducibility is one of the key ingredients for trust in science.”
However, institution-wide AI policies only set the baseline for research integrity. “You have to support disciplines to develop their norms and best practices around AI,” Joachims said. Fostering communities of practice and open dialogue is crucial as disciplines engage with AI differently.
“We need to provide platforms and hubs where people can connect, learn and discover,” Böhmann agreed. He noted that while guidelines for governing AI use in teaching and learning are widely adopted at universities, implementing them within research remains complex due to diverse disciplinary needs.
“We have to create new trust-generating mechanisms,” said Joachims. While AI can undermine research processes, there is potential for AI to generate trust, such as by checking long appendices and verifying information in papers, he added.
The panellists spoke about the evolving tension around AI use. “You want to enable people to work effectively with AI, but you also don’t want to damage their capability to reason, to learn to reason and to learn to do tasks independently,” said de Waard. As the global AI race gains momentum, innovative AI research may also need to be protected. “This is another tension that will continue to get stronger and something where agreed-upon guardrails across the scholarly ecosystem would be tremendously helpful.”
RELX, the parent company of Elsevier, has created AI principles focusing on core values such as transparency, explainability and lack of bias. “With every piece of software that we develop as part of our tool suite LeapSpace and other products, every data scientist and software developer needs to showcase that they are in compliance with these development principles,” said de Waard.
The panel:
- Tilo Böhmann, vice-president for research, University of Hamburg
- Anita de Waard, vice-president of research collaborations, Elsevier
- Thorsten Joachims, vice-provost for AI strategy, Cornell University and director, Cornell AI Initiative
- Alistair Lawrence, head of branded content, Times Higher Education (chair)
