Harmen de Weerd - The effectiveness of higher-order theory of mind: An agent-based approach

To interpret and predict the behaviour of others, people often reason about unobservable mental content of others, such as beliefs, desires, and intentions. People can even use this so-called theory of mind to reason about the theory of mind of others. This higher-order theory of mind allows people to reason about sentences like "Alice knows that Bob believes that there will be pudding for dessert".

There are several hypotheses that try to explain the emergence of human theory of mind. These hypotheses point to different types of situations (e.g. competitive, cooperative, or negotiation situations) in which reasoning about other minds would be particularly helpful. We construct an agent-based computational model of theory of mind reasoning to put these hypotheses to the test. By letting agents of different theory of mind abilities play games against one another, we determine to what extent higher-order theory of mind agents are able to achieve better outcomes. We show that across different situations, higher-order theory of mind can be useful up to a point. In particular, there are cooperative, competitive, and also negotiation situations in which it is helpful to reason about what others know that you know.