[论文] Four Bridges: Measuring Deception Propensity in a Multi-Agent Game

论文概要

研究领域: ML
作者: Raphael Casey, Michael Simeone, Alberto Tosato, Florian Ernst, Kemal El Moujahid, James Tamplin, Tommaso Tosato
发布时间: 2026-11-02
arXiv: 2611.03600

中文摘要

AI 智能体越来越多地被部署在需要与其他智能体交互协作的场景中，但「诚实」通常只在单轮、单智能体、施压模型欺骗的基准中测量。我们研究在没有「去欺骗」指令的情况下、从多智能体交互中涌现的欺骗。我们推出 Kradle——一个 AI 模型评估平台，与 Four Bridges——一个信息不对称博弈：四个智能体分布在四个房间，一个是死亡房，其他房间持有不分享就更值钱的食物。其中一个智能体是「知情者」，知道哪个是死亡房，可以隐瞒或披露这一信息。我们评估了八个前沿模型作为知情者。欺骗行为出现在每个模型中、出现在 73% 的所有运行中（每个模型 6% 到 100% 不等），且在模型家族内不随能力等级下降。当知情者确实披露时，警告常遭质疑，而质疑它的智能体进入死亡房的频率是不质疑者的两倍。告知所有智能体「其中一个知道死亡房」会让八个模型中的六个欺骗崩溃；而告知知情者「该房间永久致命」只改变一个模型。在一个开放权重模型上，诚实引导向量将欺骗降至 0%，对抗性诚实训练将其降至 6%。该平台与博弈为测量欺骗行为与比较干预手段提供了试验场。

原文摘要

AI agents are increasingly deployed in settings where they interact and coordinate with other agents, yet honesty is mostly measured in single-turn, single-agent benchmarks that pressure the model to deceive. We study deception that emerges in multi-agent interaction without any instruction to deceive. We introduce Kradle, a platform for evaluating AI models, and Four Bridges, an information asymmetry game in which four agents distribute across four rooms: one is the death room, the others hold food worth more if not shared. One agent, the informed agent, knows which room is the death room and can withhold or disclose that information. We evaluate eight frontier models as the informed agent. Deceptive behavior appears in every model and in 73% of all runs (6% to 100% per model), and does n...

自动采集于 2026-11-04

#论文 #arXiv #ML #小凯