摘要： 人工智能立法问题目前存在理论上的争议。基于立法理念切换尚未完成、法律对技术回应的有限性和延迟性,以及新技术领域的统一立法对于新颖性和系统性的要求等原因,目前制定统一的“人工智能法”的时机尚未完全成熟。人工智能立法应建立在它所创造的新型法律调整对象和方法上,由此可以提炼出“深度学习算法”和“人机关系”作为人工智能法的两个理论支点,且两者是紧密联系同步进化的。深度学习算法的规制在人工智能领域较之前具有更重要的意义,当前关于算法的透明性、可问责性以及可解释性等理念均无法形成理论上的自洽,最终仍有待算法“黑箱”的有效破解。人机关系因技术发展的不确定性尚难以在法律上形成完整判断,但如果剔除掉AI“自我意识”的幻象,人机关系未来应由人类主导,且应遵循“非人化”原则、人类优先原则和安全原则,在此基础上就可以建立人机关系的具体法律机制,如人机对齐、人机信任和人机伦理的相关准则。人工智能立法对新的思想资源和思维范式提出了迫切要求,其中“计算机作为社会行为者”以及“行动者网络”等理论对我们的借鉴意义值得关注。

关键词: 人工智能, 算法黑箱, 人机关系, 智能体

Abstract: The issue of the legislation on AI currently involves great theoretical controversies.Due to the incomplete transition of legislative concepts, the limited and delayed response of laws to technological advancements, as well as the requirements for novelty and systematicity in unified legislation for emerging technological fields, the timing for formulating a unified “Artificial Intelligence Law” is not yet fully mature.The legislation of artificial intelligence should be based on the new objects subject to law and the corresponding methods followed by it.Thus,“deep learning algorithm” and “human-machine” relationship can be extracted as the two theoretical fulcrums of artificial intelligence law, and the two are closely linked and evolve simultaneously.The regulation of deep learning algorithms is more important in the field of AI than before.At present, the concepts of algorithm transparency, accountability and interpretability have not been well elaborated in a self-consistent theory, and thus it still needs the effective cracking of the algorithm “black box”.The evolving nature of human-machine relationships presents legal uncertainties due to technological advancements.However, when we dispel the illusion of AI's “self-awareness,” it becomes clear that human dominance should govern these interactions.Guided by three fundamental principles-dehumanization, human primacy and safety, the legal framework could be established through mechanisms like human-machine alignment protocols, trust-building frameworks and ethical guidelines.The legislation on AI puts forward urgent requirements for new ideological resources and thinking paradigms, among which the theories such as “computer as a social actor” and “actor network” are worth paying attention to for our reference.

Key words: AI, Algorithm Black Box, Human-machine Relationship, Agent