arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.16424cs.AIcs.MA
收藏

可验证的语义用于智能体间通信

Verifiable Semantics for Agent-to-Agent Communication

Microsoft AI(微软人工智能);Wabash College(瓦巴什学院)

Philipp Schoenegger, Matt Carlson, Chris Schneider, Chris Daly

AI总结:

本文提出基于刺激-意义模型的认证协议,通过验证共享事件和术语的语义一致性,减少智能体间分歧。核心 guarded 推理降低分歧72-96%,语言模型验证中减少51%。

摘要:

多智能体AI系统需要一致的通信,但缺乏验证智能体共享术语理解的方法。自然语言可解释但易产生语义漂移,而学习协议高效但不透明。本文提出基于刺激-意义模型的认证协议,通过测试共享可观察事件并验证术语的实证分歧是否低于统计阈值。在此协议中,限制推理到认证术语(核心 guarded 推理)可保证分歧有界。还提出了检测漂移(重认证)和恢复共享词汇(重新协商)的机制。在不同语义分歧程度的模拟中,核心 guarded 减少分歧72-96%。在细调语言模型验证中,分歧减少51%。该框架为可验证的智能体间通信提供了第一步。

英文摘要:

Multiagent AI systems require consistent communication, but we lack methods to verify that agents share the same understanding of the terms used. Natural language is interpretable but vulnerable to semantic drift, while learned protocols are efficient but opaque. We propose a certification protocol based on the stimulus-meaning model, where agents are tested on shared observable events and terms are certified if empirical disagreement falls below a statistical threshold. In this protocol, agents restricting their reasoning to certified terms ("core-guarded reasoning") achieve provably bounded disagreement. We also outline mechanisms for detecting drift (recertification) and recovering shared vocabulary (renegotiation). In simulations with varying degrees of semantic divergence, core-guarding reduces disagreement by 72-96%. In a validation with fine-tuned language models, disagreement is reduced by 51%. Our framework provides a first step towards verifiable agent-to-agent communication.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑