MetaMind: General and Cognitive World Models in Multi-Agent Systems by Meta-Theory of Mind
arXiv:2603.00808v1 Announce Type: new Abstract: A major challenge for world models in multi-agent systems is to understand interdependent agent dynamics, predict interactive multi-agent trajectories, and plan over long horizons with collective awareness, without centralized supervision or explicit communication. In this paper, MetaMind, a general and cognitive world model for multi-agent systems that leverages a novel meta-theory of mind (Meta-ToM) framework, is proposed. Through MetaMind, each agent learns not only to predict and plan over its own beliefs, but also to inversely reason goals and beliefs from its own behavior trajectories. This self-reflective, bidirectional inference loop enables each agent to learn a metacognitive ability in a self-supervised manner. Then, MetaMind is shown to generalize the metacognitive ability from first-person to third-person through analogical reasoning. Thus, in multi-agent systems, each agent with MetaMind can actively reason about goals and b
arXiv:2603.00808v1 Announce Type: new Abstract: A major challenge for world models in multi-agent systems is to understand interdependent agent dynamics, predict interactive multi-agent trajectories, and plan over long horizons with collective awareness, without centralized supervision or explicit communication. In this paper, MetaMind, a general and cognitive world model for multi-agent systems that leverages a novel meta-theory of mind (Meta-ToM) framework, is proposed. Through MetaMind, each agent learns not only to predict and plan over its own beliefs, but also to inversely reason goals and beliefs from its own behavior trajectories. This self-reflective, bidirectional inference loop enables each agent to learn a metacognitive ability in a self-supervised manner. Then, MetaMind is shown to generalize the metacognitive ability from first-person to third-person through analogical reasoning. Thus, in multi-agent systems, each agent with MetaMind can actively reason about goals and beliefs of other agents from limited, observable behavior trajectories in a zero-shot manner, and then adapt to emergent collective intention without an explicit communication mechanism. Extended simulation results on diverse multi-agent tasks demonstrate that MetaMind can achieve superior task performance and outperform baselines in few-shot multi-agent generalization.
Executive Summary
The article introduces MetaMind, a novel world model for multi-agent systems that leverages a meta-theory of mind framework to enable agents to reason about goals and beliefs of other agents. Through self-reflective, bidirectional inference, agents learn metacognitive abilities and generalize them to third-person reasoning, allowing for zero-shot adaptation to emergent collective intentions without explicit communication. Simulation results demonstrate superior task performance and few-shot multi-agent generalization.
Key Points
- ▸ MetaMind leverages a novel meta-theory of mind framework
- ▸ Agents learn metacognitive abilities through self-reflective, bidirectional inference
- ▸ MetaMind enables zero-shot adaptation to emergent collective intentions without explicit communication
Merits
Improved Multi-Agent Generalization
MetaMind achieves superior task performance and outperforms baselines in few-shot multi-agent generalization
Demerits
Limited Scalability
The approach may struggle with scalability in complex, real-world multi-agent systems
Expert Commentary
The introduction of MetaMind marks a significant advancement in the field of multi-agent systems, as it enables agents to reason about goals and beliefs of other agents in a self-supervised manner. The use of a meta-theory of mind framework allows for a more nuanced understanding of interdependent agent dynamics, which is crucial for achieving collective awareness and long-term planning. However, further research is needed to address potential scalability issues and explore the broader implications of MetaMind in real-world applications.
Recommendations
- ✓ Further investigation into the scalability of MetaMind in complex, real-world multi-agent systems
- ✓ Exploration of potential applications and implications of MetaMind in various domains, including policy and regulation