审计多智能体审议中的置信度路由、校准与承诺
原标题：What Confidence Routing Is Actually Doing: Auditing Routing, Calibration, and Commitment in Multi-Agent Deliberation
AI 摘要
该研究审计了多智能体审议中常见的「置信度路由」协议：让各智能体报告置信度，由最高分者发言。作者将审计拆分为路由（是否选对候选）、校准（置信度是否像概率）和承诺（被选智能体是否公开说出赢得回合的答案）三个可分别测量的维度。在 4,181 条 gpt-oss-120b 奥数轨迹上，置信度能区分对错（AUROC 0.72）但严重过度自信（79% 置信 vs 52% 准确率）；等渗校准可将 ECE 从 0.278 降至 0.008，却无法恢复区分度。路由效果依赖设定，Gemma 单元中原始置信度 argmax 甚至低于随机选择；被选智能体在 20.4% 的有效配对中修改了投票答案。结论是部署前必须分别测量这三个维度。
正文节选
What Confidence Routing Is Actually Doing: Auditing Routing, Calibration, and Commitment in Multi-Agent Deliberation Abstract A common multi-agent design asks agents to report confidence and lets the highest-scoring agent speak next, implicitly using one scalar both to route the conversation and to estimate uncertainty. We audit this confidence-routed broadcast protocol by separating three trace-level questions: whether it selects the right candidate (routing), whether reported confidence behave