About

Hi. I’m Tom. I like thinking about how we can make AI safe and beneficial.

I am a DPhil (PhD) student in the Bayesian Governance Lab at the University of Oxford, and am supervised by Prof. Maike Osborne. I am part of the AIMS CDT.

My research focuses on understanding the behaviour of autonomous AI agents, and maintaining human agency + control when interacting with them.

I previously spent time researching mechanistic interpretability and trying to understand how agents think.

Please reach out! My email is: tdb47@robots.ox.ac.uk.

Writing

• Interpreting Emergent Planning in Model-Free Reinforcement Learning — a summary of my ICLR 2025 Oral paper.