Belief Coevolution in a Social Network of Generalist and Specialist Large Language Models

2026-07-29

AI Summary

This study addresses the unclear mechanisms of belief formation and propagation among large language models (LLMs) in multi-agent environments by proposing the CoevolveSim framework. Through agent-based simulations involving both generalist and specialist LLMs interacting within social networks, the work systematically investigates how domain expertise, role assignment, and network topology influence belief dynamics. Introducing fine-tuned specialist LLMs for the first time, the study reveals that such models can double consensus shifts and induce asymmetric influence patterns. It further demonstrates that role-based prompting alone is insufficient to replicate realistic belief diffusion, necessitating explicit model heterogeneity. Based on 1,280 controlled experiments, results show that specialist models significantly amplify consensus bias, while network structure and assigned roles affect individual beliefs but exert limited impact on collective consensus; accurate prediction of belief evolution in heterogeneous populations requires joint modeling of agent identity and belief composition.

Abstract

Large language models (LLMs) are increasingly deployed in multi-agent environments. However, the processes by which beliefs form and propagate among interacting LLMs remain poorly understood. We introduce CoevolveSim, a framework for studying belief diffusion within networked LLM populations. CoevolveSim allows us to isolate and study three factors: domain specialization, social-role assignment, and social network structure. Within this framework, generalist and specialist LLM agents exchange and revise beliefs. In each round, an LLM agent observes a summary of its neighbors' beliefs before updating its own. We run 1,280 controlled simulations spanning four scenarios, two network structures, and 20 medical-indication statements. We find that persona-style role assignment and network structure reshape individual belief revision but have minimal effect on population-level consensus. In contrast, introducing (finetuned) specialist LLMs more than doubles the shift in consensus and gives rise to consistent asymmetries in exerted influence. We further show that simple persistence-based opinion-dynamics models reproduce collective outcomes in all-generalist LLM populations, whereas heterogeneous LLM populations require population-level belief composition to reproduce consensus and agent identity to predict individual belief transitions. Our results indicate that realistic simulation of belief diffusion in multi-agent LLM systems requires a diverse set of underlying LLMs, not persona prompting alone.

Problem

• belief diffusion
• large language models
• multi-agent systems
• social network
• specialization

Innovation

• belief coevolution
• specialist LLMs
• multi-agent simulation
• opinion dynamics
• heterogeneous LLM populations