Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large Language Models Miss the Multi-Agent Mark

Emanuele La Malfa, Gabriele La Malfa, Samuele Marro, Jie M. Zhang, Elizabeth Black, Michael Luck, Philip Torr, Michael Wooldridge

TL;DR

This position paper argues that current MAS LLM research misapplies multi-agent system theory by neglecting core MAS principles in four areas: social agency, environment design, coordination/communication, and emergence measurement. It advocates for native social pre-training of LLMs, open multimodal and memory-enabled environments, asynchronous and standardized communication protocols, and rigorous, MAS-based metrics for emergent behaviours. By outlining concrete research directions and emphasizing precise terminology, the authors aim to ground MAS LLM work in established MAS theory to improve robustness, safety, and interoperability. The work highlights the practical impact of integrating MAS concepts to avoid reinventing solutions and to better harness open-ended, multi-agent interactions in LLM frameworks.

Abstract

Recent interest in Multi-Agent Systems of Large Language Models (MAS LLMs) has led to an increase in frameworks leveraging multiple LLMs to tackle complex tasks. However, much of this literature appropriates the terminology of MAS without engaging with its foundational principles. In this position paper, we highlight critical discrepancies between MAS theory and current MAS LLMs implementations, focusing on four key areas: the social aspect of agency, environment design, coordination and communication protocols, and measuring emergent behaviours. Our position is that many MAS LLMs lack multi-agent characteristics such as autonomy, social interaction, and structured environments, and often rely on oversimplified, LLM-centric architectures. The field may slow down and lose traction by revisiting problems the MAS literature has already addressed. Therefore, we systematically analyse this issue and outline associated research opportunities; we advocate for better integrating established MAS concepts and more precise terminology to avoid mischaracterisation and missed opportunities.

Large Language Models Miss the Multi-Agent Mark

TL;DR

This position paper argues that current MAS LLM research misapplies multi-agent system theory by neglecting core MAS principles in four areas: social agency, environment design, coordination/communication, and emergence measurement. It advocates for native social pre-training of LLMs, open multimodal and memory-enabled environments, asynchronous and standardized communication protocols, and rigorous, MAS-based metrics for emergent behaviours. By outlining concrete research directions and emphasizing precise terminology, the authors aim to ground MAS LLM work in established MAS theory to improve robustness, safety, and interoperability. The work highlights the practical impact of integrating MAS concepts to avoid reinventing solutions and to better harness open-ended, multi-agent interactions in LLM frameworks.

Abstract

Recent interest in Multi-Agent Systems of Large Language Models (MAS LLMs) has led to an increase in frameworks leveraging multiple LLMs to tackle complex tasks. However, much of this literature appropriates the terminology of MAS without engaging with its foundational principles. In this position paper, we highlight critical discrepancies between MAS theory and current MAS LLMs implementations, focusing on four key areas: the social aspect of agency, environment design, coordination and communication protocols, and measuring emergent behaviours. Our position is that many MAS LLMs lack multi-agent characteristics such as autonomy, social interaction, and structured environments, and often rely on oversimplified, LLM-centric architectures. The field may slow down and lose traction by revisiting problems the MAS literature has already addressed. Therefore, we systematically analyse this issue and outline associated research opportunities; we advocate for better integrating established MAS concepts and more precise terminology to avoid mischaracterisation and missed opportunities.

Paper Structure

This paper contains 12 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Social Intelligent Agents: LLM Agents Lack Native Social Behaviour
  3. Environment Design: MAS LLMs Environments are LLM-centric
  4. Coordination and Communication Issues in MAS LLMs
  5. MAS LLMs do not Quantify the Emergence of Complex Behaviours
  6. Alternative Views
  7. Conclusion
  8. Categories of Environments
  9. Papers that Leverage or Study Asynchronicity in MAS LLMs
  10. Frameworks and Architectures for Asynchronous Multi-Agent LLMs.
  11. Applications of Asynchronous Multi-LLM Systems.
  12. Papers on Emergent Behaviours and Their Measurement in MAS LLMs

Figures (1)

  • Figure 1: We categorise the environments of approximately $100$ MAS LLMs Benchmarks & Evaluations papers published between 2023 and 2025. For those papers that describe or allow inference of their environment characteristics (Section \ref{['sec:2']}), we present the data using wheel plots. See the papers list in Appendix \ref{['a:survey-environment']}.