Philosophical perspectives on artificial reasoning systems

The philosophy department of TU Dortmund is organising a one-day workshop exploring the philosophical dimensions of (limitations of) reasoning in large language models. Early LLMs were notorious for frequently generating outputs that, to users, appeared to contain basic reasoning errors. However, in recent years, researchers and AI developers have invested much effort in improving the common-sense and formal reasoning capacities of transformer-based models through reinforcement learning and similar techniques. This has led to a generation of 'large reasoning models’ that perform well on various benchmarks of reasoning problems. Still, questions remain about how we should interpret these capacities. Do large reasoning models really engage in reasoning, or merely in a statistical approximation of reasoning? What accounts of reasoning are available in the philosophical literature, and are these accounts applicable to LLMs? Do LLMs fare differently with respect different kinds of reasoning (e.g. common-sense reasoning, practical reasoning or counterfactual reasoning)? What are the risks of over-attributing reasoning capacities to LLMs? This workshop brings together philosophers who work on these and related questions.

Speakers:

Gregor Betz, Karlsruhe Institute of Technology

Jaroslav Malik, University of Hradec Králové

Mirco Sambrotta, Slovak Academy of Sciences

Cordelia Berz, LMU Munich

Leonard Dung, Ruhr-University Bochum

Thom-Felix Thormann, Ruhr-University Bochum

Joris Graff, TU Dortmund

Attendance is open to anyone interested, but we would like to ask attendants to register in advance by sending an email to [email protected]. Registration is possible until October 26th. Upon registering, the schedule will be provided via email.