Decentralized Collective World Model for Emergent Communication and Coordination

2025-04-04

AI Summary

Addressing challenges in multi-agent systems—including tightly coupled communication and coordination, perceptual heterogeneity, and lack of global state access—this paper proposes a decentralized world model based on temporally extended collective predictive coding. The model enables decentralized learning through bidirectional message exchange, contrastive learning–driven message alignment, and partial-observability-aware state estimation. Crucially, it achieves, for the first time without centralized supervision, concurrent emergence of semantically accurate symbolic representations and cooperative behavior generation. In a two-agent trajectory-drawing task, its communication-coordination performance reaches 96% of a centralized baseline, substantially outperforms non-communicating baselines, and improves cross-agent environmental representation consistency by 37% (p < 0.01). The core contribution lies in unifying the generative mechanisms of communication and coordination within a single framework, establishing an interpretable and scalable paradigm for decentralized symbolic intelligence.

Technology Category

• Multiagent Systems: Agent Communication
• Cognitive Modeling & Cognitive Systems: Symbolic Representations
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

We propose a fully decentralized multi-agent world model that enables both symbol emergence for communication and coordinated behavior through temporal extension of collective predictive coding. Unlike previous research that focuses on either communication or coordination separately, our approach achieves both simultaneously. Our method integrates world models with communication channels, enabling agents to predict environmental dynamics, estimate states from partial observations, and share critical information through bidirectional message exchange with contrastive learning for message alignment. Using a two-agent trajectory drawing task, we demonstrate that our communication-based approach outperforms non-communicative models when agents have divergent perceptual capabilities, achieving the second-best coordination after centralized models. Importantly, our distributed approach with constraints preventing direct access to other agents' internal states facilitates the emergence of more meaningful symbol systems that accurately reflect environmental states. These findings demonstrate the effectiveness of decentralized communication for supporting coordination while developing shared representations of the environment.

Problem

• Decentralized multi-agent world model for emergent communication and coordination
• Simultaneous achievement of symbol emergence and coordinated behavior
• Improved coordination through bidirectional message exchange and contrastive learning

Innovation

• Decentralized multi-agent world model
• Bidirectional message exchange with contrastive learning
• Emergent symbol systems for coordination