Surveying Computational Theory of Mind and a Potential Multi-agent Approach

Theory of Mind (ToM) is a capability of humans to attribute mental states to other agents allowing us to reason on the agent’s intentions and predict their future behavior. The complexity of ToM paves a way for an open arena of development for autonomous, predictive capabilities. While we see specific use-cases of ToM in recent computational studies, we must be aware of its over-arching role as a cognitive method for understanding social behavior to better guide the work towards a higher potential. We survey a few dimensions of exploration for the field to offer a way forward including ideas from the theory of child development, game theory, and generalizability, to list a few. Our own work focuses on expanding to multiple, strategic agents where we attempt to understand non-cooperative, even adversarial, behaviors in dynamic environments. We explore and aggregate elements from several contemporary ToM models, update our experimental groundwork, and begin testing the extents of their capabilities.