The Axiom of Consent: Friction Dynamics in Multi-Agent Coordination
同意公理:多智能体协调中的摩擦动力学
专题命中 AI治理与伦理 :alignment(abstract);分类 cs.CY
AI总结 提出一个基于同意公理的协调摩擦形式化框架,定义核心三元组(对齐、利益、熵)和摩擦泛函,并通过多智能体强化学习实验发现线性对齐依赖被U型关系证伪,进而提出二次型改进。
Comments 94 pages; includes machine-checked Lean 4 proofs. v2: empirical-companion references reconciled to the companion's final findings; the MARL factorial appendix is split out to the companion paper; several formal statements corrected and honestly scoped; reference list repaired