[4I1-OS-26a] ゲームを題材とした高度コミュニケーション(1/2)

[4I1-OS-26a-05] Other's mental model inference for action prediction and consistent cooperative behavior

〇Tomohiro Hasegawa¹, Kazunori Terada¹, Celso M. de Melo², Francisco C. Santos³ (1. Gifu University, 2. US Army Research Laboratory, 3. INESC-ID and Instituto Superior Tecnico, Universidade de Lisboa)

Keywords:mental model inference, 2x2 matrix game, Social Value Orientation (SVO), prisoner's dilemma, stag hunt

In a non-zero-sum society, where cooperative behavior can lead to higher profits but also to being exploitation, it is an important task to avoid exploiters and find cooperators. The mind is not directory observable, but if we can infer, it will be helpful to distinguish the cooperators from the exploiters. In the present study, we examined whether people model their counterparts as a rational agent with mental states in a game situation, and whether they predict their counterpart's actions and make decisions in novel situations based on the mental state inference. We also compared the results with those of modeling by action frequency. The results showed that although the participants predicted the counterparts' behavior in a novel situation based on the frequency model when the action frequency was available, they predicted the behavior based on the inferred mental state when the action frequency was not available, and generated consistent cooperative behavior.