Entertaining Computer Go Players

2016 Fiscal Year Final Research Report

• Project/Area Number: 26330417
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Entertainment and game informatics 1
• Research Institution: Japan Advanced Institute of Science and Technology
• Principal Investigator: Ikeda Kokolo 北陸先端科学技術大学院大学, 先端科学技術研究科, 准教授 (80362416)
• Research Collaborator: Simon Viennot 北陸先端科学技術大学院大学, 先端科学技術研究科, 助教
• Project Period (FY): 2014-04-01 – 2017-03-31
• Keywords: 接待碁 / エンターテイメント / 指導碁 / 機械学習 / 楽しませる / モンテカルロ木探索 / 自然言語処理 / 手加減

Outline of Final Research Achievements

Making "strong" computer Go players is not so difficult task now, then our next target is to make "entertaining" computer Go players. For this purpose, we tried 6 components, 1) to estimate opponent's strength and preference, 2) to lose intentionally by bad moves, 3) to play natural moves at the same time, 4) to show various strategies, 5) to play/resign at good timing, and 6) to speak with natural language. By using mainly machine learning methods and Monte Carlo tree search method, some progresses for all except 5) have been achieved in these given 3 years. This result is valuable because this is also related to "educational" computer Go players.

Free Research Field

ゲーム情報学