强人工智能刑事主体资格的法理批判与归责路径完善

Jurisprudential Critique of the Criminal Subject Qualification of Strong Artificial Intelligence and Improvement of Attribution Paths

摘要: 本文旨在批判性审视赋予强人工智能刑事主体资格的理论主张，并从法理层面系统论证其不适宜性。依据其自主性、学习能力与决策不可预测性划分为不同层级以构建分析框架。在对学界“肯定说”的核心论据进行了充分地吸纳与细致的辩驳，指出其无论在法哲学基础、法律拟制类比还是风险规制功能上均存在难以克服的缺陷。强人工智能本质上缺乏基于社会实践的主体性与刑法所要求的自由意志及刑事责任能力，对其施加刑罚将导致报应与预防目的双重落空。因此，本文主张应坚持人类中心主义的刑法立场，否定其刑事主体资格，并转而构建一个针对不同层级AI风险特征的、以研发者、生产者、使用者为核心的多层次归责体系，并辅以社会化风险分担机制，以此形成更具现实指导意义的刑事规制路径。

Abstract: This paper aims to critically examine the theoretical propositions for granting criminal subject qualification to strong artificial intelligence and to systematically demonstrate its inappropriateness from a jurisprudential perspective. By categorizing strong AI into different levels based on its autonomy, learning capability, and decision-making unpredictability, an analytical framework is constructed. Through fully absorbing and meticulously refuting the core arguments of the “affirmative view” in academia, this paper points out its insurmountable deficiencies in legal-philosophical foundations, analogies of legal fiction, and the functionality of risk regulation. Essentially, strong AI lacks the subjectivity based on social practice, as well as the free will and criminal capacity required by criminal law. Imposing punishment on it leads to the dual ineffectiveness of both retributive and preventive purposes of punishment. Therefore, this paper argues for adhering to an anthropocentric standpoint in criminal law and denying criminal subject qualification to strong AI. Instead, it advocates for constructing a multi-layered attribution system tailored to the risk characteristics of different AI levels, centered on developers, producers, and users, supplemented by socialized risk-sharing mechanisms, thereby forming a more practically instructive criminal regulatory path.