On the origins of theory of mind and misbeliefs – a game theoretical perspective

Abstract

Within AI research, proposals have been launched that argue for the creation of a new science of cooperative AI to address the problems that may arise from the diverse ecologies of AI systems that interact in complex ways with other AI and humans. One element of this new science would be the realization of AI that understands the motivations and ambitions of the other AI and humans in the mix. Such an understanding requires AI to have a Theory of Mind (ToM), i.e., they need to recognize that the other entities have cognitive and emotional states and use that information to make decisions. Many have argued that current LLM-based AI already have that capacity, yet there is no consensus that this is due to actual social reasoning.

In this seminar, Prof. Tom Lenaerts will not address whether the former is true or not, but will present a minimal model that aims to explain when ToM agents are favoured in an ecosystem of interacting agents and what the consequences are. To this end, the authors developed an Evolutionary Game Theoretical model in which a finite population of individuals use strategies that incorporate (or not) a ToM, modelled using level-k recursive reasoning, to infer a best response to the anticipated behaviour of others within the context of the centipede game. They calibrated their results to the existing experimental knowledge on how humans behave in this game. They find that strategies incorporating a ToM evolve and prevail under natural selection, provided the agents make cognitive errors and a temptation for higher future gains is in place. They found furthermore that such non-deterministic reasoning co-evolves with an optimism bias, favouring the selection of a new equilibrium configuration in the centipede game, which was not anticipated to date. This work reveals not only a unique perspective on the evolution of bounded rationality but also a co-evolutionary link between the evolution of ToM and the emergence of misbeliefs.

Bio

Tom Lenaerts is full professor in the Computer Science department at the Université Libre de Bruxelles (ULB), where he is co-heading the Machine Learning Group, focussing on AI, Game Theoretical modelling and Computational Biology. He holds a partial affiliation as research professor with the Artificial Intelligence Lab of the Vrije Universiteit Brussel and is affiliated researcher at the Center for Human-Compatible AI of UC Berkeley. He was board member, vice-chair and finally chair of the Benelux Association for Artificial Intelligence between 2016 and 2024 and Director of the Interuniversity Institute of Bioinformatics in Brussels between 2017 and 2021. He currently is the Academic Director of FARI, the Brussels AI for Common Good institute and national contact point for the CAIRNE hub in Brussels. He has been publishing in a variety of interdisciplinary domains on AI and Machine Learning, involving topics related to optimization, multi-agent systems, collective intelligence, evolutionary game theory, computational biology and bioinformatics.