Forcing different AI models to talk to each other is the only way to stop them from blindly agreeing on everything.
Multi-agent simulations often suffer from an artificial consensus where all agents quickly reach the same conclusion. This study shows that using the same model for every agent causes them to agree regardless of their assigned roles or values. By mixing different architectures, researchers were able to preserve healthy disagreement and better simulate real-world debates. Homogeneous swarms are essentially echo chambers that reflect the underlying model biases. For realistic results, developers must build diverse AI teams using a variety of different model types. Agreement in AI is often a sign of architectural failure, not logical success.
Preserving Disagreement: Architectural Heterogeneity and Coherence Validation in Multi-Agent Policy Simulation
arXiv · 2604.26561
Multi-agent deliberation systems using large language models (LLMs) are increasingly proposed for policy simulation, yet they suffer from artificial consensus: evaluator agents converge on the same option regardless of their assigned value perspectives. We present the AI Council, a three-phase deliberation framework, and conduct 120 deliberations across two policy scenarios to test two interventions. First, architectural heterogeneity (assigning a different 7-9B parameter model to each value per