Adaptive Theory of Mind for LLM-based Multi-Agent Coordination
arXiv:2603.16264v1 Announce Type: new Abstract: Theory of Mind (ToM) refers to the ability to reason about others' mental states, and higher-order ToM involves considering that others also possess their own ToM. Equipping large language model (LLM)-driven agents with ToM has long been considered to improve their coordination in multiagent collaborative tasks. However, we find that misaligned ToM orders-mismatches in the depth of ToM reasoning between agents-can lead to insufficient or excessive reasoning about others, thereby impairing their coordination. To address this issue, we design an adaptive ToM (A-ToM) agent, which can align in ToM orders with its partner. Based on prior interactions, the agent estimates the partner's likely ToM order and leverages this estimation to predict the partner's action, thereby facilitating behavioral coordination. We conduct empirical evaluations on four multi-agent coordination tasks: a repeated matrix game, two grid navigation tasks and an Overco
arXiv:2603.16264v1 Announce Type: new Abstract: Theory of Mind (ToM) refers to the ability to reason about others' mental states, and higher-order ToM involves considering that others also possess their own ToM. Equipping large language model (LLM)-driven agents with ToM has long been considered to improve their coordination in multiagent collaborative tasks. However, we find that misaligned ToM orders-mismatches in the depth of ToM reasoning between agents-can lead to insufficient or excessive reasoning about others, thereby impairing their coordination. To address this issue, we design an adaptive ToM (A-ToM) agent, which can align in ToM orders with its partner. Based on prior interactions, the agent estimates the partner's likely ToM order and leverages this estimation to predict the partner's action, thereby facilitating behavioral coordination. We conduct empirical evaluations on four multi-agent coordination tasks: a repeated matrix game, two grid navigation tasks and an Overcooked task. The results validate our findings on ToM alignment and demonstrate the effectiveness of our A-ToM agent. Furthermore, we discuss the generalizability of our A-ToM to non-LLM-based agents, as well as what would diminish the importance of ToM alignment.
Executive Summary
This article presents Adaptive Theory of Mind (A-ToM), an innovative approach to improve coordination among large language model (LLM)-driven agents in multi-agent collaborative tasks. By estimating and aligning with the ToM order of their partner, A-ToM agents can predict and facilitate behavioral coordination. The authors validate their findings through empirical evaluations on four multi-agent coordination tasks, demonstrating the effectiveness of A-ToM in improving agent coordination. While the article makes significant contributions to the field of artificial intelligence and multi-agent systems, its generalizability to non-LLM-based agents and the impact of ToM alignment on various tasks warrant further exploration.
Key Points
- ▸ Adaptive Theory of Mind (A-ToM) is designed to align with the ToM order of its partner agent.
- ▸ A-ToM agents estimate the partner's ToM order based on prior interactions and leverage this estimation to predict the partner's action.
- ▸ Empirical evaluations on four multi-agent coordination tasks demonstrate the effectiveness of A-ToM in improving agent coordination.
Merits
Strength in Addressing ToM Mismatches
The article effectively addresses the issue of ToM mismatches, which is a critical limitation in previous ToM-equipped agents. By introducing A-ToM, the authors provide a novel solution to improve agent coordination in multi-agent tasks.
Demerits
Limited Generalizability
The article's focus on LLM-driven agents may limit its generalizability to other types of agents, such as those based on neural networks or rule-based systems. Further research is needed to explore the applicability of A-ToM in these contexts.
Insufficient Discussion on ToM Alignment
While the article highlights the importance of ToM alignment, it does not provide a comprehensive discussion on the factors that diminish or enhance the importance of ToM alignment in various tasks.
Expert Commentary
The article makes significant contributions to the field of artificial intelligence and multi-agent systems by introducing A-ToM, a novel approach to improve agent coordination in multi-agent tasks. However, its generalizability to non-LLM-based agents and the impact of ToM alignment on various tasks warrant further exploration. The article's emphasis on LLM-driven agents is also relevant to the broader field of multi-agent systems, where LLMs are increasingly being used to model human-like intelligence and behavior. Ultimately, the development of A-ToM agents has the potential to improve the coordination and efficiency of human-AI teams in various domains, and its implications for policy and practice are worth further consideration.
Recommendations
- ✓ Future research should explore the generalizability of A-ToM to non-LLM-based agents and the impact of ToM alignment on various tasks.
- ✓ The authors should provide a more comprehensive discussion on the factors that diminish or enhance the importance of ToM alignment in various tasks.