自主型人工智能犯罪的刑事归因与归责

Criminal Attribution and Responsibility of Autonomous Artificial Intelligence Crimes

摘要: 人工智能技术的迭代升级催生出自主型人工智能犯罪的复杂归责难题，风险社会下人工智能侵害人类法益的行为越来越频繁，基于其自主决策具备特有的算法黑箱与多端介入性，导致在侵害结果发生时，适用传统刑法因果关系理论判定责任归属陷入适用困境。本文通过分析条件说、相当因果关系说和双重筛选条件说在人工智能自主决策犯罪归因中的局限性，提出三层筛选穿透的归因模型和分层刑事归责的建构路径。查明算法中自主决策与结果侵害间的刑事意义关联和相关主体的互动联系、提取重叠因果中的强黏性标准，与制度层面的自主人工智能犯罪的法律搭建和过错责任划分结合、辅之以多端化复合治理模式，实现技术治理与法律规制的深度融合。

Abstract: The iterative upgrading of artificial intelligence (AI) technology has given rise to the complex attribution of responsibility for autonomous AI crimes, and AI infringes on human legal interests more and more frequently in a risky society. Based on its autonomous decision-making, it has its own algorithmic black box and multiple interventions, which leads to a dilemma of applying traditional criminal law causation theories in determining responsibility attribution in the case of infringing results. This paper analyzes the limitations of the condition theory, equivalent causation theory and double screening condition theory in the attribution of artificial intelligence autonomous decision-making crime, and proposes the attribution model of three-layer screening penetration and the construction path of hierarchical criminal attribution. It identifies the criminal significance of the connection between autonomous decision-making in the algorithm and the resultant infringement and the interaction of the relevant subjects, extracts the strong viscous criteria in the overlapping causation, combines with the legal construction of autonomous AI crimes and the fault-based division of responsibility at the institutional level, and is supplemented by the multi-terminalized composite governance model, so as to realize the deep fusion of technological governance and legal regulation.