MetaMind: Modeling Human Social Thoughts with Metacognitive Multi-Agent Systems

TL;DR

MetaMind introduces a cognitively inspired, three-agent metacognitive framework that decomposes social reasoning into Stage 1 Theory-of-Mind hypothesis generation, Stage 2 norm-aware refinement, and Stage 3 validated response generation. By interleaving hypothesis formation, normative constraints, and self-validation, MetaMind achieves state-of-the-art results on ToMBench, social cognition benchmarks, and open-ended social simulations, effectively narrowing the gap between LLMs and human social reasoning. The architecture is model-agnostic and demonstrated across diverse backbones, with ablation studies confirming the necessity of all stages. These findings advance AI toward robust, culturally sensitive, and empathetic social interactions, with broad implications for dialogue systems and socially aware AI.

Abstract

Human social interactions depend on the ability to infer others' unspoken intentions, emotions, and beliefs-a cognitive skill grounded in the psychological concept of Theory of Mind (ToM). While large language models (LLMs) excel in semantic understanding tasks, they struggle with the ambiguity and contextual nuance inherent in human communication. To bridge this gap, we introduce MetaMind, a multi-agent framework inspired by psychological theories of metacognition, designed to emulate human-like social reasoning. MetaMind decomposes social understanding into three collaborative stages: (1) a Theory-of-Mind Agent generates hypotheses about user mental states (e.g., intent, emotion), (2) a Moral Agent refines these hypotheses using cultural norms and ethical constraints, and (3) a Response Agent generates contextually appropriate responses while validating alignment with inferred intent. Our framework achieves state-of-the-art performance across three challenging benchmarks, with 35.7% improvement in real-world social scenarios and 6.2% gain in ToM reasoning. Notably, it enables LLMs to match human-level performance on key ToM tasks for the first time. Ablation studies confirm the necessity of all components, which showcase the framework's ability to balance contextual plausibility, social appropriateness, and user adaptation. This work advances AI systems toward human-like social intelligence, with applications in empathetic dialogue and culturally sensitive interactions. Code is available at https://github.com/XMZhangAI/MetaMind.

Figures (4)

• Figure 1: MetaMind multi-agent framework. The architecture comprises three collaborative agents—Theory-of-Mind Agent, Moral Agent, and Response Agent—working in a staged metacognitive loop. The ToM Agent generates hypotheses about latent mental states, which are refined by the Moral Agent using cultural/ethical constraints. The Response Agent synthesizes contextually appropriate outputs while validating them with inferred intent.
• Figure 2: MetaMind improves Theory-of-Mind reasoning performance across LLMs. Each pair compares base model accuracy (gray) with MetaMind-enhanced accuracy (purple) on ToMBench. MetaMind consistently boosts ToM reasoning across both open-source and proprietary LLMs, highlighting its generality and effectiveness. See detailed performance in Appendix \ref{['appendix:tombench']}.
• Figure 3: (a) Comparison of original LLMs and human capabilities. (b) Comparison between MetaMind-enhanced LLM performance against human capabilities.
• Figure 4: Accuracy landscape over $(\lambda,\beta)$ for the three most competitive $k$.