📄 中文摘要
随着 AI 相关事件的频繁发生，安全失效和恶意行为的滥用问题日益严重。在复杂情境中，识别导致不良结果的因素，即因果选择问题，是确立责任的关键第一步。研究通过人类实验考察了在 AI 系统导致有害结果时，公众对因果责任的看法，包括因果链结构中的因果性、责备、可预见性和反事实推理。结果显示，当 AI 的代理性为中等（人类设定目标，AI 决定手段）或高（AI 设定目标和手段）时，参与者更倾向于归因于更大的因果责任。
人类对 AI 在代理、滥用和不对齐中的因果归属
📄 English Summary
Human Attribution of Causality to AI Across Agency, Misuse, and Misalignment
AI-related incidents are increasingly frequent and severe, encompassing safety failures and misuse by malicious actors. In such complex scenarios, identifying the elements that caused adverse outcomes, known as the problem of cause selection, is a critical first step in establishing liability. This research conducts human experiments to investigate public perceptions of causal responsibility in causal chain structures when AI systems are involved in harmful outcomes. The findings reveal that when AI agency is moderate (humans set the goals while AI determines the means) or high (AI sets both the goals and the means), participants attribute greater causal responsibility to the AI systems involved.