Improving Multi-Agent Cooperation using Theory of Mind
This addresses the challenge of designing more human-like and effective cooperative AI agents for applications like gaming or teamwork scenarios, though it is incremental as it builds on existing ToM methods.
The study tackled the problem of improving multi-agent cooperation by incorporating a Bayesian Theory of Mind (ToM) approach to represent others' intentions, finding that teams with ToM agents significantly outperform non-ToM agents in a cooperative game, with benefits increasing as more ToM agents are included.
Recent advances in Artificial Intelligence have produced agents that can beat human world champions at games like Go, Starcraft, and Dota2. However, most of these models do not seem to play in a human-like manner: People infer others' intentions from their behaviour, and use these inferences in scheming and strategizing. Here, using a Bayesian Theory of Mind (ToM) approach, we investigated how much an explicit representation of others' intentions improves performance in a cooperative game. We compared the performance of humans playing with optimal-planning agents with and without ToM, in a cooperative game where players have to flexibly cooperate to achieve joint goals. We find that teams with ToM agents significantly outperform non-ToM agents when collaborating with all types of partners: non-ToM, ToM, as well as human players, and that the benefit of ToM increases the more ToM agents there are. These findings have implications for designing better cooperative agents.