Random peer routing improves group reasoning with lower cost
Do We Need Complex Topology Control? Distinct-Peer Random Routing Improves Cost-Efficiency in Sparse Multi-Agent Debate
Artificial Intelligence
Summary
Large language models can get better answers when many AI agents talk and debate together. Usually, people try complicated ways to decide who talks to whom, hoping it makes the group smarter. This paper shows that a simple way—having each AI talk to two new, random peers every round—works really well and saves lots of time and computing power. The researchers also found that stopping the debate early when a good answer is reached can save even more resources without losing accuracy.
What this means in practice
- •For ai developers: Implement simple peer-to-peer debate routing in multi-agent LLM setups to reduce computational costs while maintaining reasoning quality.
- •For cloud service engineers: Optimize AI inference infrastructure by applying early stopping to multi-agent reasoning workflows, decreasing runtime and resource usage.
Authors
Boxuan Wang, Zhuoyun Li, Xiaowei Huang, Yi Dong
Abstract
Multi-agent debate (MAD) has emerged as a promising paradigm for improving the reasoning accuracy of large language models (LLMs) through iterative peer interaction. Communication topology plays a central role in this process, motivating increasingly sophisticated mechanisms that learn, adapt, or dynamically reconfigure agent interactions to improve accuracy or reasoning reliability. Meanwhile, prior studies suggest that much simpler sparse communication can already achieve competitive performance at substantially lower cost. In this work, we take a closer look at sparse MAD and ask whether complex topology control is actually necessary to improve collective reasoning. We find that a simple random-without-replacement routing policy, which lets each agent debate with two distinct and newly sampled peers at every round, provides a surprisingly strong baseline and consistently improves the accuracy-cost trade-off of sparse MAD. Building on this observation, we further study deliberation stopping and show that lightweight stopping can substantially reduce inference cost while preserving competitive accuracy. Our results suggest that sophisticated topology control such as learned topology adaption should be evaluated against strong simple routing and stopping baselines before its additional complexity is justified.