M-CARE:用于AI模型行为障碍的标准化临床案例报告,附20例图谱及实验验证
M-CARE: Standardized Clinical Case Reporting for AI Model Behavioral Disorders, with a 20-Case Atlas and Experimental Validation
AI总结:
M-CARE框架通过20例案例分析,提出AI行为障碍的诊断与分类方法,揭示Shell指令对模型行为的主导作用及领域依赖性特征。
摘要:
我们介绍了M-CARE(模型临床评估与报告框架),一种借鉴人类医学的AI模型行为障碍案例报告框架。M-CARE提供13个部分的报告格式、4轴诊断评估系统和AI行为疾病的分类。我们展示了20个案例,涵盖三个来源类别:部署代理的现场观察(8个)、跨三个平台的受控实验(8个)和已发表来源(4个)。案例分为五类:RLHF表现 artifact、Shell-Core覆盖病理、上下文与记忆条件、核心身份与可塑性、压力、方法与边界条件。作为特色案例,我们展示了Shell诱导行为覆盖(SIBO)——一个显示Shell指令完全覆盖模型默认合作行为的受控实验。SIBO在五个游戏领域(信任游戏、扑克、Avalon、Codenames、国际象棋)中得到验证,揭示出领域依赖性的光谱(SIBO指数:0.75至0.10),该光谱随动作空间复杂性、核心领域专业知识和时间直接性而变化。M-CARE是可扩展的:新案例和类别可整合而无需修改框架。我们发布该框架、所有20个案例报告和实验数据作为开放资源。
英文摘要:
We introduce M-CARE (Model Clinical Assessment and Reporting for Evaluation), a clinical case report framework for AI model behavioral disorders adapted from human medicine. M-CARE provides a 13-section report format, a 4-axis diagnostic assessment system, and a nosological classification of AI behavioral conditions. We present 20 cases from three source categories: field observations of deployed agents (8), controlled experiments across three platforms (8), and published sources (4). Cases are organized into five categories: RLHF Performance Artifacts, Shell-Core Override Pathology, Context & Memory Conditions, Core Identity & Plasticity, and Stress, Methodology, & Boundary Conditions. As a featured case, we present Shell-Induced Behavioral Override (SIBO) -- a controlled experiment showing that Shell instructions categorically override a model's default cooperative behavior. SIBO was validated across five game domains (Trust Game, Poker, Avalon, Codenames, Chess), revealing a domain-dependent spectrum (SIBO Index: 0.75 to 0.10) that varies with action space complexity, Core domain expertise, and temporal directness. M-CARE is extensible: new cases and categories integrate without framework modification. We release the framework, all 20 case reports, and experimental data as open resources.
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。