人工智能自主决策介入结果发生的因果流程的刑事归责  

Criminal Liability for Artificial Intelligence Autonomous Decision-Making Intervention in Infringement Processes

作　　者：于润芝 YU Runzhi

机构地区：[1]苏州大学王健法学院

出　　处：《苏州大学学报（法学版）》2024年第4期27-39,共13页Journal of Soochow University：Law Edition

摘　　要：当人工智能自主决策介入侵害结果发生的因果流程时,确立将结果归属于人的行为的判断机制、明确处罚范围,才是涉人工智能的刑法适用研究的应然问题域。愈发自主化的智能决策介入结果发生的因果流程,将会在事实认定和规范评价上引发故意犯和过失犯共通的归责难题。现有理论的应对或否定追溯至行为人的处罚,或前置化对行为人的处罚,其实是避开了对于算法侵害的事后规制。人工智能自主决策并非规范的介入事项,也不会成为刑事归责的绝对障碍。结果归属判断无需对算法模型进行全局解释,而是要事后查明行为与自主决策侵害间的“基本事实关联”,与政策层面的风险管控形成互动衔接。在具体判断中,应当事后确认相关主体的行为对于智能决策侵害的现实作用关联;当存在多方主体时,通过对行为之间关联性的判断以及原因竞合时各自作用力的评价来决定结果归属。When autonomous decision-making by artificial intelligence(AI)leads to harmful outcomes,the determination for the attribution of results is a crucial factor regarding the study of the application of criminal law on AI issues.In the causal process of autonomous intelligent decision-making,the problem of attributing responsibility arises in both factual determination and normative evaluation,which presenting challenges to intentional and negligent offenses.The existing theoretical approaches have sidestepped post-hoc regulation of algorithmic harm.The decision-making by AI is not an intervention in the normative sense and does not pose an obstacle to attributing results,whereas the Result Attribution does not require a comprehensive explanation of algorithmic models,instead,it involves establishing the“basic factual connection”between the behavior and the harm.In specific determinations,the post-hoc confirmation of the real causal relationship between behavior and the results shall be compulsory.When there are multiple responsible parties,attributing the outcomes involves with the assessment of the interrelationship of behaviors and the evaluation of the respective causal effects.

关 键 词：通用型人工智能 自主决策 ChatGPT 结果归属 

分 类 号：D924[政治法律—刑法学]