Multi-agent systems suffer from harmful homogenization effects
Homogenization in Multi-Agent Systems
Machine Learning
Summary
Multi-agent systems consist of many individual agents working together to solve complex tasks. The authors show that when agents interact closely, they tend to become too similar, a problem called homogenization. This similarity can cause hidden errors to become widespread, biased decisions to stick around, and unfair treatment in peer reviews. The authors find that simple fixes like mixing different models don't fully solve these problems, highlighting a need for better approaches to keep agents diverse and avoid these issues.
What this means in practice
- •For software developers: Identify and reduce correlated errors in AI code generation by monitoring agent behavior similarity over interactions.
- •For recruitment teams: Detect persistent biased influence in multi-agent hiring systems even after biased agents are removed to improve fairness.
Authors
Prakhar Ganesh, Kyra Wilson, Luca Zappella, Barry-John Theobald, Nicholas Apostoloff, Lucas Monteiro Paes, Nivedha Sivakumar
Abstract
Multi-agent systems (MAS) leverage interactions between agents to perform complex tasks. Despite their success, we show that these interactions can also lead to homogenization, i.e., agents converging to similar behaviors. Homogenization in MAS can reduce agent diversity and reinforce shared failures. In this paper, we operationalize homogenization using three metrics: conformity to the majority, polarization towards extremes, and growing inertia against changes over subsequent interactions. We evaluate homogenization in MAS for code generation, hiring, and scientific peer review. Across these tasks, we show that homogenization translates to concrete downstream risks: in code generation, it hides and amplifies correlated errors which can create systemic vulnerabilities; in hiring, it allows the influence of biased agents to persist long after their removal; and in peer review, it creates uneven evaluation standards across research areas. Our results establish homogenization as a failure mode of MAS, demonstrating that MAS evaluations must move beyond aggregate performance to carefully analyze interaction dynamics. Finally, we show that simple approaches to increase diversity---leveraging sampling stochasticity and mixed-models MAS---fail to reduce homogenization risks, highlighting the need for strategies to effectively leverage agent diversity.