People

Pierre Beckmann is a PhD student at EPFL and Idiap Research Institute, and a MATS Scholar, with close ties to the Institute of Philosophy at the University of Bern. He studies understanding, reasoning, and world modeling in language models through two complementary approaches: mechanistic interpretability with philosophy, investigating how models implement these capacities with insights from his philosophical background, and philosophy with mechanistic interpretability, drawing on empirical findings to argue about whether, and in what sense, models can be said to have these capacities. His work has appeared in philosophy journals such as Philosophical Studies and at machine learning venues such as NeurIPS. He is a founder of PhilInterp, the Philosophy with Mechanistic Interpretability Network.

Matthieu Queloz is an incoming Swiss National Science Foundation Professor at the University of Zurich, where he leads a five-year research project on AI models' ability to understand and reason about the world. His earlier work examined how concepts are formed, evaluated, and engineered, and his historical research traced early modern mechanistic approaches to cognition and the systematicity of thought. Building on both, he now investigates how different levels of understanding and world models emerge in large language models. His approach bridges philosophy and mechanistic interpretability. He is also interested in how far AI can model normative truths and navigate conflicts between values. He is a member of the Digital Society Initiative and a founder of PhilInterp, the Philosophy with Mechanistic Interpretability Network.

Iwan Williams is a philosopher of mind and artificial intelligence, working as a postdoc at the Center for Philosophy of Artificial Intelligence (CPAI), University of Copenhagen. He also directs the P1 Program on Functions and Representations in AI Systems at the Pioneer Centre for Artificial Intelligence, and in 2026 was a Summer Fellow at the Center for AI Safety (CAIS), San Francisco. He received his PhD in 2021 from Monash University, Melbourne, supervised by Tim Bayne and Jakob Hohwy. His research focuses on the nature and structure of mental representations in minds and machines, and has addressed questions such as “can chatbots make assertions?”, “do LLMs have intentions?” and “how can philosophy help mechanistic interpretability?”

Eliot du Sordet is a PhD student in the Department of Philosophy at the University of Neuchâtel, working at the intersection of philosophy of mind and philosophy of AI. He is particularly interested in whether large language models represent the world and build world models of it, a question he approaches with findings from mechanistic interpretability. His work draws on teleosemantics, the theory on which what a state represents is fixed by the function it was selected to serve, and asks what that theory says about the representations found inside language models. He is also interested in how the move to agentic language models may change which mental capacities we attribute to them. He co-organises The Age of the Bookworm.

Patrick Butlin is a philosopher of mind and cognitive science and Senior Research Lead at Eleos AI Research, a non-profit organisation that investigates the potential for AI moral patienthood. He studies properties that could ground moral patienthood in AI systems, such as consciousness and agency, trying to work out which of these properties matter and whether they are present in current systems. Part of his work involves devising potential mechanistic interpretability experiments, and interpreting results, that could tell on these questions.