Human-Scale Game Theory in Imperfect Information Settings

不完美信息环境下的人性博弈论

• 批准号: 298239-2012
• 负责人: Bowling, Michael
• 金额: $ 3.06万
• 依托单位: University of Alberta
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2016
• 资助国家: 加拿大
• 起止时间: 2016-01-01 至 2017-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=594882
• 关键词: Human; Scale; Game; Theory; Imperfect

Decision making in the presence of other agents is a critical challenge for deploying artificially intelligent agents in real-world settings. Real world settings -- ranging from negotiation to security to interactive entertainment -- also typically involve imperfect information, where agents are missing key pieces of information to make their decisions. Game-theoretic techniques for decision-making under such circumstances have recently seen considerable improvement. However, they are still not adequate for human-scale problems. For example, advancements in general-purpose equilibrium algorithms for sequential games have made it possible to solve games with trillions of game states (up from millions only 5 years ago), yet the smallest variants of well known games played by humans (e.g., poker) are still far much larger. While my research group has had some recent success at playing complex games at a human level, the techniques employed necessarily give up theoretical guarantees when applied to such large domains. Without guarantees, though, it is hard to predict the success of the approaches for other applications, or guide a practitioner in how best to exploit improvements in computing resources.

在其他代理商在场的情况下，决策是在现实世界中部署人工智能代理的一个关键挑战。现实世界中的环境 - 从谈判到安全到交互式娱乐 - 通常还涉及不完美的信息，而代理缺少关键信息来做出决定。在这种情况下，用于决策的游戏理论技术最近已有很大的改进。但是，它们仍然不足以解决人类规模的问题。例如，连续游戏的通用均衡算法的进步使得以数万亿的游戏状态（仅在5年前就高于数百万人）可以解决游戏，但是人类（例如，扑克）玩过的最著名游戏的最小变体仍然大得多。尽管我的研究小组最近在人体层面上玩复杂的游戏方面取得了一些成功，但使用的技术在适用于如此大的领域时必须放弃理论保证。但是，没有保证，很难预测其他应用程序的成功，或者指导从业者如何最好地利用计算资源的改进。