Date: Wednesday, 31.01.24 15:20-17:00 CET
Location: Building S1|15 Room 133
Abstract:
The basic tenet of embodied cognition is that (human) cognition operates in a physical and social world, not as arbitrary computation over arbitrary symbols. There is, however, little agreement as to what that entails exactly, leading to various discussions and disagreements in various fields. One particular point of debate is whether embodiment entails that one has to reject cognitivism, representationalism and computationalism, or just some subset thereof. Meanwhile, work in AI currently largely focuses on computational approaches that are entirely disembodied, with clear success but also clear limitations. In robotics, most approaches to intelligent behaviour also remain computationalist by necessity.
The main purpose of this talk is to discuss how a body can be relevant beyond trivial roles as a data source and sink even if one wants to adhere to computationalism otherwise. We will briefly talk about symbol grounding, especially in the context of large language models, and about how representational frameworks can be shaped by the body based on simple demonstrations in a perception-action learning framework.
