人工智能可解释性的技术困境与法律消解

The Technical Dilemma of Artificial Intelligence Explainability and Legal Resolution

摘要: 人工智能技术的持续突破，正在以颠覆性的力量重塑全球科技创新与产业经济发展格局。然而，人工智能的迅速发展也带来了诸多风险，其中人工智能的“算法黑箱”现象和不确定性导致人们无法理解人工智能的决策和输出，从而引发信任危机和归责困境，很大程度上阻碍了人工智能产业的发展。本文从人工智能的“算法黑箱”本质出发分析其可解释性问题，分别从技术角度、用户信任和法律价值三个角度论证对人工智能可解释性进行法律规制的必要性和可行性。在明确可解释性和透明度的定义和区别的基础上，将可解释性分为面向用户和专业人员两方面的可解释性，从可追溯、可视化、反事实解释三个角度构建面向用户的可解释性要求，提出根据领域分类的“硬法”“软法”可解释性要求，以法律引导、促进人工智能技术的安全、可靠、可信任的发展。

Abstract: The continuous breakthroughs in artificial intelligence technology are reshaping the global pattern of technological innovation and industrial economic development with disruptive power. However, the rapid development of artificial intelligence has also brought many risks, among which the phenomenon of the “algorithmic black box” and uncertainty lead to people’s inability to understand the decisions and outputs of artificial intelligence, thus triggering a crisis of trust and accountability dilemma, which largely hinders the development of the artificial intelligence industry. This paper analyzes the interpretability issues starting from the essence of the “algorithmic black box” of artificial intelligence, demonstrating the necessity and feasibility of legal regulation of artificial intelligence interpretability from three perspectives: technological perspective, user trust, and legal value. Based on clarifying the definitions and distinctions of interpretability and transparency, interpretability is divided into user-facing and professional interpretability. From the three angles of traceability, visualization, and counterfactual explanation, user-facing interpretability requirements are constructed, proposing “hard law” and “soft law” interpretability requirements categorized by fields to guide and promote the safe, reliable, and trustworthy development of artificial intelligence technology through law.