Topic: Introspective Machines: Are LLMs Better at Self-Reflection than Humans?
Time: March 11th, 2025, 3:00–5:00 PM
Venue: Room 2401, Guanghua West Main Building
Speaker: Prof. Herman Cappelen, Department of Philosophy, The University of Hong Kong
Host: Prof. Liu Chuang（刘闯）, School of Philosophy, Fudan University
Lecture Abstract:
Can LLMs introspect? If so, can we trust their introspective reports? we gave Claude a very difficult math problem. It came up with an answer and we asked it how confident it was in that answer. Here is the reply: I'd rate my confidence at about 60%. While my reasoning through the symmetry argument feels solid (leading to 1/9), I have some doubts about whether I fully accounted for all the dynamics of the sequential cutting process. The fact that each cut changes the geometry and possibilities for future cuts makes me think there might be subtle effects I missed that could break the apparent symmetry. Note that if this is a true self-report, then Claude can feel that an argument is solid, have doubts, think, and assess its own credences. In other words, we would know that it has cognitive states of the kinds that we humans are familiar with. The conclusion of the paper is an optimistic one: Not only could there already be machine introspection, but it could be better, more reliable, than human introspection (which, arguably, isn’t all that good.)
Short Bio of the Speaker
Herman Cappelen is a Norwegian philosopher. He is currently the Chair Professor of Philosophy at the University of Hong Kong. Cappelen was previously a professor of philosophy at the University of Oslo and the University of St Andrews in Scotland. He works primarily on the philosophy of language, philosophical methodology, philosophy of AI and related areas in epistemology, philosophy of mind, and metaphysics. In 2013, he became editor of the journal Inquiry: An Interdisciplinary Journal of Philosophy.
Short Bio of the Host:
Prof. Liu Chuang (刘闯) is a Distinguished Professor at the School of Philosophy, Fudan University, and the Academic Director of the Institute of Philosophy, Chinese Academy of Sciences. His research interests include philosophy of science, philosophy of physics, methodology of science, and philosophy of intelligence.
