Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cognitive Insights and Stable Coalition Matching for Fostering Multi-Agent Cooperation

Jiaqi Shao, Tianjun Yuan, Tao Lin, Bing Luo

TL;DR

This work investigates how Theory of Mind (ToM) affects cooperation in large-language-model (LLM) powered multi-agent systems and finds that ToM depth can sometimes impair collaboration. It introduces a ToM-aware team-formation framework that forms belief-aligned, stable coalitions by optimizing social welfare under stability constraints and a belief-action alignment metric, while allowing specialization-driven adaptation. The mechanism relies on iterated belief updates via a ToM function $f_{ToM}$, a stability notion against blocking coalitions, and an epsilon threshold to prune weak alignments. Across programming, debate, and reasoning tasks, experiments show that belief-aligned teams, especially with higher-ToM, achieve improved task performance and longer team lifetimes, demonstrating effective, human-like coordination in complex tasks. The findings suggest practical implications for cooperative AI in domains requiring robust, interpretable, and scalable multi-agent collaboration, while acknowledging ethical considerations and the need for safeguards in belief modeling.

Abstract

Cognitive abilities, such as Theory of Mind (ToM), play a vital role in facilitating cooperation in human social interactions. However, our study reveals that agents with higher ToM abilities may not necessarily exhibit better cooperative behavior compared to those with lower ToM abilities. To address this challenge, we propose a novel matching coalition mechanism that leverages the strengths of agents with different ToM levels by explicitly considering belief alignment and specialized abilities when forming coalitions. Our proposed matching algorithm seeks to find stable coalitions that maximize the potential for cooperative behavior and ensure long-term viability. By incorporating cognitive insights into the design of multi-agent systems, our work demonstrates the potential of leveraging ToM to create more sophisticated and human-like coordination strategies that foster cooperation and improve overall system performance.

Cognitive Insights and Stable Coalition Matching for Fostering Multi-Agent Cooperation

TL;DR

This work investigates how Theory of Mind (ToM) affects cooperation in large-language-model (LLM) powered multi-agent systems and finds that ToM depth can sometimes impair collaboration. It introduces a ToM-aware team-formation framework that forms belief-aligned, stable coalitions by optimizing social welfare under stability constraints and a belief-action alignment metric, while allowing specialization-driven adaptation. The mechanism relies on iterated belief updates via a ToM function , a stability notion against blocking coalitions, and an epsilon threshold to prune weak alignments. Across programming, debate, and reasoning tasks, experiments show that belief-aligned teams, especially with higher-ToM, achieve improved task performance and longer team lifetimes, demonstrating effective, human-like coordination in complex tasks. The findings suggest practical implications for cooperative AI in domains requiring robust, interpretable, and scalable multi-agent collaboration, while acknowledging ethical considerations and the need for safeguards in belief modeling.

Abstract

Cognitive abilities, such as Theory of Mind (ToM), play a vital role in facilitating cooperation in human social interactions. However, our study reveals that agents with higher ToM abilities may not necessarily exhibit better cooperative behavior compared to those with lower ToM abilities. To address this challenge, we propose a novel matching coalition mechanism that leverages the strengths of agents with different ToM levels by explicitly considering belief alignment and specialized abilities when forming coalitions. Our proposed matching algorithm seeks to find stable coalitions that maximize the potential for cooperative behavior and ensure long-term viability. By incorporating cognitive insights into the design of multi-agent systems, our work demonstrates the potential of leveraging ToM to create more sophisticated and human-like coordination strategies that foster cooperation and improve overall system performance.
Paper Structure (29 sections, 6 equations, 2 figures, 13 tables, 1 algorithm)

This paper contains 29 sections, 6 equations, 2 figures, 13 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-agent Cooperation.
  4. Motivation
  5. Formulation
  6. ToM Formulation for LLM Agents
  7. Team Formulation
  8. Team Formation Framework.
  9. ToM-Aware Team Formation
  10. Team Formation Mechanism
  11. Adaptation for Specialized Agent Abilities
  12. Experiment
  13. Main Results
  14. Case Study: Debate
  15. Evaluations on Reasoning Tasks
  16. ...and 14 more sections

Figures (2)

  • Figure 1: Illustration for the MAS with ToM agents. (Left) The MAS for the iterative programming tasks consists of one project manager (PM) with ToM ability and four Engineers. The cooperation involves: ❶PM updates its beliefs and takes actions based on its ToM reasoning; ❷PM observes Engineers' actions; ❸PM think and reflect other's actions and update beliefs; ❹PM provides instructions to Engineers. (Right) The ToM cognitive thinking process of PM agent involves recursive belief updates and decision-making based on the inferred beliefs. We also employ different ToM settings for the PM and engineers as detailed in \ref{['app:analysis_tom']}.
  • Figure 2: Comparative analysis of belief alignment across ToM levels in ( MBPP ) at round 5 under "with random team formation" and "our proposed team formation" settings.