Human learning transfer takes 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 role in learning transfer by letting human participants play different games against the same computer agent, who possesses human-like theory of mind abilities with a limited degree of iterated reasoning. We find that participants deviate from Nash equilibrium play and learn to adapt to the opponent's strategy to exploit it. Moreover, we show that participants transfer their learning to new games and that this transfer is moderated by the level of sophistication of the opponent. Computational modelling shows that it is likely that players start each game using a model-based learning strategy that facilitates generalisation and opponent model transfer, but then switch to behaviour that is consistent with a model-free learning strategy in the later stages of the interaction.

To what extent do human reward learning and decision-making rely on the ability to represent and generate richly structured relationships between options? We provide evidence that structure learning and the principle of compositionality play crucial roles in human reinforcement learning. In a new multi-armed bandit paradigm, we found evidence that participants are able to learn representations of different reward structures and combine them to make correct generalizations about options in novel contexts. Moreover, we found substantial evidence that participants transferred knowledge of simpler reward structures to make compositional generalizations about rewards in complex contexts. This allowed participants to accumulate more rewards earlier, and to explore less whenever such knowledge transfer was possible. We also provide a computational model which is able to generalize and compose knowledge for complex reward structures. This model describes participant behaviour in the compositional generalization task better than various other models of decision-making and transfer learning.