Group size effects and collective misalignment in LLM multi-agent systems

2025-10-25

AI Summary

Prior studies on LLM-based multi-agent systems often compare only fixed group sizes, neglecting the continuous effects of scale variation on collective dynamics. This limits understanding of how population size fundamentally shapes emergent bias and coordination failure. Method: We propose a scalable multi-agent experimental framework grounded in coordination games, integrating large-scale agent-based simulations with mean-field theoretical analysis. Contribution/Results: We identify, for the first time, that inter-agent interaction can amplify, induce, or override individual preferences—revealing model-dependent dynamical phase transitions at critical scale thresholds. Empirically, beyond the supercritical scale, system evolution becomes deterministic, enabling precise characterization of attraction basins underlying competitive equilibria. Our findings establish group size itself as a pivotal dynamical parameter—not merely a design choice—thereby introducing a new paradigm for designing and analyzing controllability in multi-agent systems.

Technology Category

• Multiagent Systems: Mechanism Design
• Game Theory and Economic Paradigms: Coordination and Collaboration
• Machine Learning: Scalability of ML Systems

Application Category

• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Multi-agent systems of large language models (LLMs) are rapidly expanding across domains, introducing dynamics not captured by single-agent evaluations. Yet, existing work has mostly contrasted the behavior of a single agent with that of a collective of fixed size, leaving open a central question: how does group size shape dynamics? Here, we move beyond this dichotomy and systematically explore outcomes across the full range of group sizes. We focus on multi-agent misalignment, building on recent evidence that interacting LLMs playing a simple coordination game can generate collective biases absent in individual models. First, we show that collective bias is a deeper phenomenon than previously assessed: interaction can amplify individual biases, introduce new ones, or override model-level preferences. Second, we demonstrate that group size affects the dynamics in a non-linear way, revealing model-dependent dynamical regimes. Finally, we develop a mean-field analytical approach and show that, above a critical population size, simulations converge to deterministic predictions that expose the basins of attraction of competing equilibria. These findings establish group size as a key driver of multi-agent dynamics and highlight the need to consider population-level effects when deploying LLM-based systems at scale.

Problem

• Explores how group size affects collective bias in LLM multi-agent systems
• Analyzes non-linear dynamics and model-dependent regimes in agent interactions
• Investigates population-level convergence and attraction basins in large groups

Innovation

• Systematically explores multi-agent dynamics across group sizes
• Develops mean-field analytical approach for population-level predictions
• Reveals non-linear group size effects on collective bias formation