[3H1-GS-10] AI application

[3H1-GS-10-05] Computional model of mind reading based on Bayesian theory of mind in social dilemma.

〇Tomohiro HASEGAWA¹, Kazunori TERADA¹, Francisco C. Santos², Celso M. de Melo³ (1. Gifu University, 2. INESC-ID and Instituto Superior Técnico, Universidade de Lisboa, 3. DEVCOM U.S. Army Research Laboratory)

Keywords:SVO(Social Value Orientation), Game theory, Prisoner's dilemma, Stag-hunt, Bayesian theory of mind

In a nonzero-sum society, where cooperative behavior brings higher benefits but can be exploited, the ability to accurately predict whether a partner will behave in a cooperative (Good: G) or exploitative (Bad: B) manner is important. Although it is known that reputation determines whether a partner is G or B, and contributes to the maintenance of a cooperative society, it is unknown what cognitive-computational mechanisms are used to infer G or B in iterated interactions. We believe that people make Bayesian inferences about their partner's SVO based on the observed situation and behavior, using a generative model of a rational agent based on the partner's Social Value Orientation (SVO) as a likelihood function, and predict the partner's behavior based on this inference. In this paper, we conducted a computer simulation using Bayesian theory of mind, a situational action frequency model, and an action frequency model to predict the opponent's behavior in a 5-round repeated prisoner's dilemma game following a 10-round repeated deer hunting game, and compared the results with those of a participant experiment.