Improving the winning percentage of the Werewolf team through collusion strategies

Abstract

In recent years, the number of systems using artificial intelligence (AI) is increasing. Even in Chess, Shogi, and Go, which are perfect information games, AI has been used to beat masters. Furthermore, AI has been actively researched in the imperfect information games Poker and Mahjong, and has reached the point where it can be played on par with humans. On the other hand, the werewolf game focused on in our research is an incomplete information game, and it is also necessary to aim for victory as a team using asymmetric information. In this paper, we report that the winning percentage of werewolf team can be improved by taking a team strategy in a 15-player werewolf game.