Transfer of Learned Opponent Models in Zero Sum Games

Abstract

AbstractHuman learning transfer abilities take advantage of important cognitive building blocks such as an abstract representation of concepts underlying tasks and causal models of the environment. One way to build abstract representations of the environment when the task involves interactions with others is to build a model of the opponent that may inform what actions they are likely to take next. In this study, we explore opponent modelling and its transfer in games where human agents play against computer agents with human-like limited degrees of iterated reasoning. In two experiments, we find that participants deviate from Nash equilibrium play and learn to adapt to their opponent’s strategy to exploit it. Moreover, we show that participants transfer their learning to new games. Computational modelling shows that players start each game with a model-based learning strategy that facilitates between-game transfer of their opponent’s strategy, but then switch to behaviour that is consistent with a model-free learning strategy in the latter stages of the interaction.