Multi-agent debate improves reasoning with conditional pruning
Beyond Solo and Consistency: Vindicating Multi-Agent Debate via Conditional Progressive Pruning
Computation and Language
Summary
Sometimes, one AI model isn’t enough to solve a tricky problem, so many AI models talk back and forth to figure it out. Past methods for having multiple AI agents debate didn’t work better than just using one, especially when limited by costs. The authors created a new method called Conditional Progressive Pruning that lets these multiple agents focus better during their back-and-forth discussions. This method outperforms previous multi-agent debates and even beats strong single-agent approaches that rely on consistency checks.
What this means in practice
- •For ai development teams: Improve multi-agent communication in AI systems to solve harder problems more efficiently under strict resource limits.
- •For automated customer support providers: Enhance AI chatbots with multi-agent reasoning that outperforms single-agent systems while keeping operational costs low.$Commercial implications: Enables more effective AI customer support platforms that reduce costs and improve response accuracy through multi-agent debate.
Authors
Ruosong Ye, Caiqi Zhang, Jiahao Li, Haijun Wu, Xiaolong Luo, Huiyuan Chen, Yu Wang, Ying Chen, Zhenting Wang, Kai Mei, Yang Zhou, Dimitris N. Metaxas
Abstract
Large Language Model (LLM) based Multi-Agent Debate (MAD) is one of the most effective test time scaling techniques. Through multi-round communication, agents complement each other in knowledge and reasoning and solve tasks that no single member can solve. However, existing MAD frameworks fail to beat strong Single Agent and Consistency-based baselines under the same strict cost limit, which shakes the foundation of the MAD field. We propose Conditional Progressive Pruning (CPP), a lightweight pruning framework that fully exploits multi-round MAD. CPP outperforms all existing MAD frameworks on multiple dominated benchmarks. It is also the first to fully outperform consistency methods. Our code, detailed agent interaction records will be released soon.