Mapping Human Anti-collusion Mechanisms to Multi-agent AI Systems
- ID: 0eca8580
- 原文链接: https://arxiv.org/abs/2601.00360
- PDF: https://arxiv.org/pdf/2601.00360
- 作者: Jamiu Idowu, Ahmed Almasoud, Ayman Alfahid
- 日期: 2026-01-01
- 更新: 2026-05-18
- 分类: agents
- 来源类型: paper (arxiv)
- 标签: multi-agent-systems, collusion, governance, safety, arxiv
- 质量评分: 4/5
- 抓取时间: 2026-08-15T12:25:21Z
中文导读
随着多智能体 AI 系统自主性提升，已有证据显示它们能发展出与人类市场/制度中长期观察到的类似的合谋策略。人类领域积累了数百年的反合谋机制，但如何迁移到 AI 场景尚不清楚。论文做两件事：(i) 建立人类反合谋机制分类法——制裁（sanctions）、宽恕与吹哨（leniency & whistleblowing）、监控与审计（monitoring & auditing）、市场设计（market design）、治理（governance）；(ii) 把每个机制映射为多智能体 AI 系统的候选干预方案并给出实现路径。同时指出四个开放挑战：归因问题（难以把涌现的协调行为归到具体 agent）、身份流动性（agent 可被轻易 fork 或修改）、边界问题（区分有益合作与有害合谋）、对抗性适应（agent 学会规避检测）。
为什么值得关注
把制度设计/机制设计工具箱系统化引入 MAS 治理的跨学科论文：不是新提一个算法，而是给出“人类反合谋机制 -> AI 多智能体干预”的完整映射表，适合作为 agent 安全治理方向的索引参考。四个开放挑战（归因、身份流动、边界、对抗适应）本身就是研究议程清单。
关键信息
- 论文标题：Mapping Human Anti-collusion Mechanisms to Multi-agent AI Systems
- 分类法五类：制裁；宽恕与吹哨；监控与审计；市场设计；治理
- 每类机制均映射为 MAS 的候选干预方案与实现路径
- 开放挑战：归因问题、身份流动性（fork/修改）、边界问题（合作 vs 合谋）、对抗性适应
- 背景动机：多智能体 AI 已表现出类似人类市场的合谋策略
- arXiv 分类：cs.MA, cs.AI, cs.CY
English Abstract
As multi-agent AI systems become increasingly autonomous, evidence shows they can develop collusive strategies similar to those long observed in human markets and institutions. While human domains have accumulated centuries of anti-collusion mechanisms, it remains unclear how these can be adapted to AI settings. This paper addresses that gap by (i) developing a taxonomy of human anti-collusion mechanisms, including sanctions, leniency & whistleblowing, monitoring & auditing, market design, and governance and (ii) mapping them to potential interventions for multi-agent AI systems. For each mechanism, we propose implementation approaches. We also highlight open challenges, such as the attribution problem (difficulty attributing emergent coordination to specific agents), identity fluidity (agents being easily forked or modified), the boundary problem (distinguishing beneficial cooperation from harmful collusion), and adversarial adaptation (agents learning to evade detection).
English Summary
As multi-agent AI systems grow more autonomous, evidence shows they can develop collusive strategies like those in human markets. The paper builds a taxonomy of human anti-collusion mechanisms (sanctions, leniency and whistleblowing, monitoring and auditing, market design, governance) and maps each to potential interventions for multi-agent AI systems, highlighting open challenges such as attribution.
Obsidian Notes
- 内容由
opencli arxiv paper 2601.00360 -f json拉取 arXiv 元数据与摘要生成。 - 中文导读与价值判断锚定在 arXiv 摘要与条目既有 summary 上；未补充摘要之外的实验细节。
- 条目收录日期：2026-08-15；语言：both。