Opponent Modelling in Strategic Interaction

在重复的多智能体游戏中，如扑克或金融市场，智能体（人类或人工）可以学习对手的策略，并相应地调整自己的策略。传统上，博弈论忽略了这种学习或对手建模，专注于单次博弈的均衡策略。在这项研究中，我们将专注于多智能体游戏中的对手建模。参与者反复与不同的计算机代理人进行合作和竞争游戏，这些代理人实施差异化策略。我们的目标是引出参与者的策略是否会受到他们认为他们面临的对手的知识的影响，以及这种知识在多大程度上会在游戏中转移。随后的目标是测试参与者如何实施这些模型的各种假设，例如通过观察过去的戏剧，刻板印象或观点。

In repeated multi-agent games such as Poker or Financial Markets, agents (human or artificial) can learn about the strategies of their opponents and adapt their own strategies accordingly. Traditionally, game theory has ignored such learning or opponent modelling, focusing on equilibrium strategies for single-shot games. In this study, we will focus on opponent modelling in multi-agent games. Participants play cooperative and competitive games repeatedly with distinct computer agents that implement differentiated strategies. We aim to elicit whether participant's strategies are affected by knowledge of the opponent they believe they are facing, and to what extent is this knowledge transferred across games. A subsequent goal is to test various hypotheses on how these models are implemented by participants, such as through observation of past plays, stereotyping or perspective taking.