Grounding Natural Language for Multi-agent Decision-Making with Multi-agentic LLMs

2025-08-10

AI Summary

This work addresses core challenges in multi-agent decision-making—semantic inconsistency, inefficient coordination, and policy conflict—by proposing a large language model (LLM)-based multi-agent collaboration framework. The framework employs natural language as a unified semantic medium, achieving inter-agent semantic grounding and policy alignment through interpretable prompt engineering, a hierarchical memory architecture, multimodal perception alignment, and lightweight instruction fine-tuning. Unlike conventional symbolic communication requiring predefined protocols, our approach enables emergent, protocol-free intent understanding and joint policy generation. Ablation studies on canonical social dilemma benchmarks—including the Prisoner’s Dilemma and Public Goods Game—demonstrate consistent superiority over baseline methods across cooperation rate, task completion rate, and policy stability. Results validate natural language as a robust, general-purpose interface for scalable and adaptive multi-agent coordination.

Technology Category

• Multiagent Systems: Coordination and Collaboration
• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions

Abstract

Language is a ubiquitous tool that is foundational to reasoning and collaboration, ranging from everyday interactions to sophisticated problem-solving tasks. The establishment of a common language can serve as a powerful asset in ensuring clear communication and understanding amongst agents, facilitating desired coordination and strategies. In this work, we extend the capabilities of large language models (LLMs) by integrating them with advancements in multi-agent decision-making algorithms. We propose a systematic framework for the design of multi-agentic large language models (LLMs), focusing on key integration practices. These include advanced prompt engineering techniques, the development of effective memory architectures, multi-modal information processing, and alignment strategies through fine-tuning algorithms. We evaluate these design choices through extensive ablation studies on classic game settings with significant underlying social dilemmas and game-theoretic considerations.

Problem

• Extend LLMs for multi-agent decision-making integration
• Design systematic framework for multi-agentic LLMs
• Evaluate framework in game settings with social dilemmas

Innovation

• Integrating LLMs with multi-agent decision-making algorithms
• Advanced prompt engineering and memory architectures
• Multi-modal processing and fine-tuning alignment strategies