Monte Carlo Tree Search player for Mai- Star and Balance Evaluation

Abstract

The paper presents Artificial Intelligence (AI) players developed for the game Mai-Star's characters. The Monte Carlo Tree Search algorithm simulates player characters, geishas. Mai-Star is a card game for three to six players. Each player can have multiple cards at one moment. Almost every card has an associated effect, depending on the rules adopted. The game has Geisha that allow players to make some special actions called abilities. The AI player uses the Single Observer Information Set Monte Carlo Tree Search (SO-ISMCTS) algorithm. We developed a Mai-Star computer game with an interface to interact with, for human competitive testing. We also implemented a random AI player. Using statistical evaluations of the generated AI players, we find that the game is unbalanced for any one-versus-one match-up. It further suggested that one of the geishas named Oboro has features that give her an advantage over other geishas for winning. Oboro was re-played against the other Geisha after decreasing her abilities (a nerf), which confirms how amazing her powerset was as Oboro's winning rate decreased from ninety-four percent to a still astonishing seventy-one percent.