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 J. Wooldridge

Keywords: multi-agent large language models, language models, multi-agent systems

TL;DR: LLMs misuse key Multi-agent system concepts; we call for aligning them more closely with foundational Multi-agent theory.

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.

Lay Summary: Many multi-agent LLM systems (MAS LLMs) are being built and described using terms from classic multi-agent systems (MAS) theory. Still, they often fail to adhere to the foundational principles. These LLM setups frequently lack key agentic qualities: genuine autonomy, meaningful social interaction, structured environments, and clear ways to measure emergent behaviour. This mismatch risks the field unnecessarily resolving old problems already addressed by classic MAS research instead of generating new insights. We pinpoint these shortcomings, explain their significance, and propose ways to integrate better-established MAS concepts into the design and evaluation of LLM-based agent frameworks.

Submission Number: 161