麻將之人工智慧研究

Abstract

麻將為資訊不明確、多玩家類型的遊戲，除了無法得知另外三家對手的手牌以外，隨著牌局的進行，玩家可能會根據摸牌而轉牌，更增加了遊戲的變化性。

在本篇論文之中，我們根據麻將遊戲的特性來設計AI，除了根據我方的手牌，以減少上聽數為首要準則，決定如何捨牌；另外，隨著牌局進行到中、後期，在無法確定安全捨牌時，我們試圖使用Monte-Carlo方法模擬遊戲進行，避開有可能導致輸牌的捨牌，以追求最終贏牌為目標。

Mohjong is an imperfect information game, and multiplayer game. Each player couldn’t get other players’ tile information. During game processing, each player might change tiles to make better combinations. And it makes game more unpredictable. In this thesis,we analyze Mohjong’s features, and try to design a smart computer program. We try to reduce count-to-listen number to decide which tile be discarded is better. When we detect other player might in listen condition, would start to use Monte-Carlo method to simulate the game to the end. We want to use the method to make a better decision to avoid to lose game.