Social circuits trace message effects to reduce echo chambers in AI teams
Social Circuits behind Multi-agent Echo Chambers
Artificial Intelligence
Summary
When AI agents talk to each other to solve problems, they sometimes create echo chambers where they all end up making the same mistakes. This paper introduces a way to track exactly how one message changes the internal thinking of another agent. The authors show how these changes can be used to pick better messages, improving group decisions while using fewer words. This method helps understand and guide AI communication more effectively.
What this means in practice
- •For chatbot developers: Improve multi-chatbot coordination by selecting messages that positively influence internal reasoning and avoid reinforcing shared errors.
- •For customer support teams: Enhance automated agent collaboration in support systems to make faster, more accurate decisions with fewer interactions.
Authors
Chuiyang Meng, Wenlu Yu, Ming Tang, Cheng Li
Abstract
Language-model agents exchange messages to combine evidence, but their communication can also create echo chambers that reinforce shared errors. However, overall task performance does not explain how a message changes the receiving agent's internal activations and affects its decision. In this work, we introduce Social Circuits, a framework for tracing message effects through receiver activations. We compare the receiver's answers before and after changing a message. Then, we restore selected activations recorded under the original message to determine how much of the message effect these activations reproduce. Based on Social Circuits, we propose Circuit-Guided Deliberation (CGD), which learns to select useful messages using receiver activation changes. We establish when activation replacement preserves receiver decisions and bound the gap between CGD's task performance and the best achievable through message selection. Experiments show that receiver activation changes explain the message effects and guide message selection that improves the task performance. Across three models and four datasets, CGD achieves the highest or joint-highest average accuracy in our main comparisons while generating fewer tokens than multi-agent baselines.