2026-04-22T18:19:23Z https://ipsj.ixsq.nii.ac.jp/oai

oai:ipsj.ixsq.nii.ac.jp:00209854 2025-01-19T18:22:08Z 1164:5305:10533:10534

人間にとって手強さを感じさせるガイスターAIの提案 A Geister AI that is tough against Human Players 伊藤, 篤志 伊藤, 毅志 Atsushi, Ito Takeshi, Ito ガイスター 不完全情報ゲームであるガイスターのAIの研究は行われているが，現在のところ最強と言われているAIでも人間と対戦して十分に勝ち越すレベルのものは出来ていない．これは，何回か対戦すると，AIの癖が容易に見破られてしまうことが原因であると思われる．本研究では，この問題を解決するために，ベイズ推定を用いた相手の駒の推定を行い，相手も同様に駒推定を行うことを利用して木探索を行うAIを提案した．この提案AIと過去のAI大会で優秀な成績を収めている2つのAIを実際に人間と対戦させることで，AIと対戦したときに人間が感じる手強さや面白さについて評価させる実験を行った．その結果，提案AIは従来のAIの一つよりは手強く，面白く感じさせることが出来たが，もう一つのAIとは同等の手強さと面白さであることが示された．対戦したAIに対する感想から，提案AIの有効性が示せたプレイヤも多数いたが，一部のプレイヤには見破られてしまった．より見破られにくくするために，探索に用いる評価関数の設計をより工夫した方が良い点が示唆された． In recent years, there has been a lot of research on AI for the imperfect information game Geister, but even the strongest AI has yet to be able to play against a human and win. This is probably due to the fact that the habits of the AI can be easily detected after a few rounds of play. In order to solve this problem, we proposed a tree search AI that uses Bayesian estimation of the opponent's pieces and the self piece estimation as well. The proposed AI and two other AIs that had achieved excellent results in past AI competitions were tested against human players to evaluate the strength and fun of the AI against human players. The results showed that the proposed AI was sufficiently tougher and more interesting than one of the conventional AIs, but was equally tough and interesting as the other AI. Many players were able to demonstrate the effectiveness of the proposed AI, but some players were able to see through it. In order to make it more difficult to be detected, it was shown that the design of the evaluation function used in the search needs to be improved. 情報処理学会 2021-02-26 jpn technical report https://ipsj.ixsq.nii.ac.jp/records/209854 2188-8736 AA11362144 研究報告ゲーム情報学（GI） 2021-GI-45 5 1 5 https://ipsj.ixsq.nii.ac.jp/record/209854/files/IPSJ-GI21045005.pdf application/pdf 593.8 kB 2023-02-26