Design of an Artificial Game Entertainer by Reinforcement Learning

摘要

Games are often used for performance evaluation of artificial intelligence (AI) methods. Most AI studies using the games aim to design a computer player which plays a game better than humans. On the other hand, video game companies develop one-to-one games such as Go and Reversi and design computer opponents which entertain human players to have them play the games a lot. However, it takes much time to design such computer opponents. In this paper, we propose a reinforcement learning method for automatically designing an AI player entertaining the human players, especially those who are not good at playing games, in the one-to-one games. There are several ways to entertain them. One of the ways is to use a computer opponent which is neither too strong nor too weak, and the proposed method designs such an artificial game entertainer. The performance of the proposed method is evaluated through numerical experiments.