[3Xin2] Poster session 1

[3Xin2-98] Role estimation using information in trust and distrust of utterance of AIwolf

〇Tomohiro Miyagawa¹ (1.Univ. of Kogakuin)

Keywords:AI, Simulation, Agent, AIWolf

AIwolf, a game of incomplete information, relies heavily on the estimation of other players’ pole, which are hidden information, to win or lose. The basic method of estimating role information is to use the utterances of other players, but it is difficult to determine whether or not to believe the utterance because they may contain lies. Therefore, this study focuses on the “amount of role information contained in utterances” and obtains the amount of role information in utterance that are believed or not believed from statistical data. The difference between the average amount of role information obtained from the statistical data and the amount of role information obtained during the actual game was used to estimate the role, and the win rates were compared. As a result, we created an agent to estimate the role using the game logs, and compared the win rate between the two agents of believing and disbelieving in the utterance.