Osborne's Oxford lab pitches Bayesian tools for AI governance

The Bayesian Governance Lab at Oxford, led by M. A. Osborne, opens with a methodological claim on its own page: AI policy disagreement is a problem of uncertainty, and uncertainty can be measured.

"We treat that disagreement as a problem of uncertainty, and uncertainty as something to be measured and reasoned about rather than asserted," the lab writes. It sits inside the Oxford Martin AI Governance Initiative and the university's Machine Learning Research Group, and describes itself as a small, intentional fellowship applying Bayesian inference, Gaussian processes, probabilistic numerics, and Bayesian optimization to the assurance, monitoring, and governance of AI systems.

Osborne has carried that framing into government. The page says he provided oral evidence to the House of Commons Science, Innovation and Technology Committee on AI governance and contributed to the Committee's interim report identifying twelve governance challenges. His advisory footprint also touches the Cabinet Office, HM Treasury, and European policy forums.

Recent output spans technical and policy work: a 2025 NeurIPS paper on scalable valuation of human feedback for model alignment (Fujisawa, Adachi, Osborne); a 2024 Brown Journal of World Affairs reappraisal of generative AI and the future of work (Frey & Osborne); and a 2024 arXiv paper, "Governing Through the Cloud: The Intermediary Role of Compute Providers in AI Regulation," by Heim, Fist, Egan and colleagues.

Two of the AI researchers we track have already shared the page.