Join the GW Philosophy Department for an engaging talk by Rémi Tison, Postdoctoral Researcher at GW, as he explores the congitive abilities of AI and their limitations in understanding human thought in his presentation, "Can ChatGPT Read My Mind? AI Social Cognition and Mindshaping."

Abstract: 

Recent advances in Large Language Models (LLMs) have sparked intense philosophical and scientific debates concerning their cognitive abilities. Despite their impressive achievements, LLMs still fall short of human performance in the domain of social cognition, particularly in the ability to attribute mental states such as beliefs, desires and intentions to others. In this presentation, I argue that many LLM socio-cognitive limitations have a root which is often neglected in the literature. Human social cognition crucially involves normative cognition. Roughly, normative cognition consists in our ability to evaluate behaviors and thoughts as correct or incorrect and to regulate others and ourselves in light of such evaluations. From the perspective presented here, attributing beliefs or intentions to others crucially relies on normative cognition because it not only has the function of predicting and explaining behavior, but also of justifying and rationalizing it with respect to social and rational norms. The normative attitudes which are necessary for these uniquely human socio-cognitive abilities have not yet been implemented in LLMs. I additionally suggest that this is not due to some inherent technological challenge. Rather, proper normative attitudes have not been implemented in artificial agents in general because of ethical and political concerns.