Proposta de modelo NEAT para jogar One Night Ultimate Werewolf

Abstract

The use of Artificial Intelligence approaches as NPCs in games is a very common practice, as they seek to convey the impression to players that these characters are somewhat autonomous. One of the approaches used is the technique called NEAT, which consists of making use of artificial neural networks together with genetic algorithms to manage the topology, connections, and weights of a network in an adaptive way. This work presents the proposal to create an NPC for games in a subcategory of board games, those based on bluff and incomplete information. The game used as a case study is One Night Ultimate Werewolf, a social deduction game, so that information is incomplete for players, and part of them must use the bluff in order to confuse other players. The objective is to evaluate the possibility of modeling the behaviors of this type of game for the application of NEAT.