Multi-Agent Collaboration Mechanisms: A Survey of LLMs

Abstract: With recent advances in LLMs, Agentic AI has become phenomenal in real-world applications, moving toward multiple LLM-based agents to perceive, learn, reason, and act collaboratively. These LLM-based Multi-Agent Systems (MASs) enable groups of intelligent agents to coordinate and solve complex tasks collectively at scale, transitioning from isolated models to collaboration-centric approaches. This work provides an extensive survey of the collaborative aspect of MASs and introduces an extensible framework to guide future research. Our framework characterizes collaboration mechanisms based on key dimensions: actors (agents involved), types (e.g., cooperation, competition, or coopetition), structures (e.g., peer-to-peer, centralized, or distributed), strategies (e.g., role-based or model-based), and coordination protocols. Through a review of existing methodologies, our findings serve as a foundation for demystifying and advancing LLM-based MASs toward more intelligent and collaborative solutions for complex, real-world use cases. In addition, various applications of MASs across diverse domains, including 5G/6G networks, Industry 5.0, question answering, and social and cultural settings, are also investigated, demonstrating their wider adoption and broader impacts. Finally, we identify key lessons learned, open challenges, and potential research directions of MASs towards artificial collective intelligence.

• The paper surveys LLM collaboration mechanisms and introduces a framework to classify multi-agent interactions.
• It analyzes various collaboration types including cooperation, competition, and coopetition along with rule-, role-, and model-based strategies.
• It highlights applications in 5G/6G networks, Industry 5.0, NLP, and social simulations, emphasizing future research in safety and ethics.

Multi-Agent Collaboration Mechanisms: A Survey of LLMs

The paper "Multi-Agent Collaboration Mechanisms: A Survey of LLMs" explores the evolving field of LLM-based Multi-Agent Systems (MASs) and their capabilities to solve complex, real-world tasks collaboratively. This paper surveys the mechanisms that enable collaborations—and provide a framework for understanding these mechanisms more comprehensively.

Introduction

With the increasing capabilities of LLMs in understanding and generating language, researchers are now focusing on using these models in Multi-Agent Systems (MASs). These systems allow multiple LLMs to cooperate, compete, or evolve collectively toward predefined objectives, transitioning from isolated to collaboration-centric models. This paper surveys the existing methodologies in MAS and offers a new framework to guide future research in this dynamic field.

Framework for Multi-Agent Collaboration Systems

The paper introduces a framework that distinguishes collaboration systems based on the actors, types, structures, strategies, and coordination protocols essential for MAS deployment:

1. Actors: Defines the agents involved in the collaboration.
2. Types: Identifies whether the collaboration is cooperative, competitive, or a mix of both.
3. Structures: Outlines the interaction structure (peer-to-peer, centralized, distributed).
4. Strategies: Discusses how roles, rules, or models guide collaboration.
5. Coordination: Describes the protocols and methods for orchestrating seamless interaction between agents.

   Figure 1: Our proposed framework for LLM-based multi-agent collaborative system. Each agent consists of a LLM $m$ as the neural processor, current objective $o$, environment $e$, input perception $x$ and corresponding output/action $y$. The framework's central focus is on collaboration channels $\mathcal{C}$.

Collaboration Types

The paper categorizes collaborations in MAS into three primary types:

• Cooperation: Agents work together to achieve shared objectives. This type of collaboration emphasizes delegation and specialization to enhance overall system performance.
• Competition: Agents pursue individual objectives that may conflict with others', yet can still contribute to a broader competitive collaborative environment, such as in debates.
• Coopetition: A mix of cooperation and competition—agents collaborate on shared tasks while competing in others.

  Figure 2: Illustrative examples of collaboration types, where each agent $a$ is equipped with different tools or capabilities through their system prompt $r$.

Collaboration Strategies

MASs can employ various strategies to foster collaboration:

1. Rule-based: Interactions are governed by predefined rules to ensure predictability and fairness.
2. Role-based: Agents are assigned specific roles that leverage their strengths in completing tasks.
3. Model-based: Agents use probabilistic decision-making to adapt to dynamic environments.

   Figure 3: Different types of collaboration strategies. In the rule-based example, agents debate and participate in majority voting. A software project is an instance of a role-based protocol. In games, agents communicate and perform probabilistic decision-making.

Applications

LLM-based MASs have been applied across various domains, including:

• 5G/6G Networks: Enhancing communication systems with semantic layer processing.
• Industry 5.0: MASs in IoT for automating complex industrial operations and services.
• NLP Applications: The use of MASs can enhance the capabilities of LLMs in generating human-like conversation or responses.
• Cultural and Social Simulations: Simulating complex social dynamics and interactions to advance sociocultural understanding.

  Figure 4: LLM-based multi-agent collaborative system in social and cultural applications.

Conclusion

The integration of LLMs into multi-agent systems holds promise for advancing AI's capabilities in addressing complex, dynamic challenges. The framework introduced in this paper provides a structured approach for understanding and designing these systems. Continued exploration of collaboration types, strategies, and applications will be critical to advancing the field toward developing more intelligent, adaptive, and cooperative AI systems. The paper identifies that future improvement and study will be needed in collective intelligence, safety and ethical considerations, and sustainability challenges in large-scale systems.