Sequential Theory of Mind Modeling in Team Search and Rescue Tasks

Abstract

The ability to make inferences about other’s mental states is referred to as having a Theory of Mind (ToM). Such ability is fundamental for human social activities such as empathy, teamwork, and communication. As intelligent agents being involved in diverse human-agent teams, they are also expected to be socially intelligent to become effective teammates. In this paper, we propose a computational ToM model which observes team behaviors and infer their mental states in a simulated search and rescue task. The model structure consists of a transformer-based language module and an RNN-based sequential mental state module in order to capture both team communication and behaviors for the ToM inference. To provide a feasible baseline for our ToM model, we present the same inference task to human observers recruited from Amazon MTurk. Results show that our proposed computational model achieves a comparable performance with human observers in the ToM inference task.