Multi agent language models show improved reasoning through interaction
Collective Emergent Alignment: Measuring and Amplifying Interaction-Induced Reasoning Alignment in Multi-Agent LLMs
Artificial IntelligenceMultiagent Systems
Summary
Large language models can work together by sharing their thoughts and ideas, which sometimes leads to better reasoning than if they worked alone. The authors studied how these models align their reasoning when interacting and created a way to measure this improvement. They found that sharing selected facts instead of full explanations improves how well the models agree and solve tasks, while also reducing the amount of back-and-forth communication needed. Their work helps understand how groups of language models can cooperate more effectively.
What this means in practice
- •For multi agent system engineers: Improve collective decision making by implementing fact-level communication protocols that enhance reasoning alignment and reduce communication overhead.
- •For dialog system developers: Design chatbots that share compact, selective facts to improve joint problem solving and responsiveness in multi-agent conversational settings.
Authors
Boxuan Wang, Zhuoyun Li, Sarvapali D. Ramchurn, Xiaowei Huang, Yi Dong
Abstract
Multi-agent large language model (LLM) systems can exhibit emergent behaviors through interaction, yet whether a similar phenomenon exists for reasoning alignment remains largely unexplored. We study whether agent interaction produces additional reference-grounded reasoning alignment beyond matched individual self-refinement, and introduce REMAP (Reference-Entailment-Matrix for Agentic Process), a framework that quantifies this effect through Semantic Coverage and interaction-specific alignment gains. Across diverse model families and two multi-hop reasoning benchmarks, we observe broadly positive interaction gains, particularly on reasoning content that depends on integrating evidence distributed across agents. These gains are consistently associated with independently measured peer-information recovery, while further analysis reveals composition-dependent heterogeneity effects and asymmetric information transfer across agent capability levels. Motivated by these findings, we further introduce a fact-level multi-agent communication framework that replaces full reasoning-trace exchange with compact, gap-aware, and selectively routed information. Our experiments show that compact fact exchange improves reasoning alignment and task performance while reducing communication overhead, whereas more selective communication exposes a trade-off between acquiring distributed information, retaining local content, and communication efficiency. Overall, our results show that collective emergent alignment can be quantified, linked to information-transfer patterns, and used to guide more effective multi-agent communication.