Abstract: Artificial intelligence legislation in China involves four major controversies: whether, when, in what form to legislate, and how to balance security and development.This article argues for the prompt adoption of comprehensive AI legislation.Existing fragmented regulation and concerns about systemic coherence do not negate its necessity; theories of legal infrastructure and second-order decision-making further demonstrate its institutional value.As no ideal legislative moment exists, China should follow an approach of “legislate first, improve later” to address the Collingridge dilemma.AI legislation should also shape technological development toward human-centered and socially beneficial ends.With the rise of AI agents, the conventional risk-regulation paradigm has become insufficient.Legislation should therefore extend from risk control to relationship construction by establishing basic rules for human-agent, agent-agent, and composite relationships, thereby developing an AI legal framework suited to China's circumstances.

Key words: Artificial Intelligence, Comprehensive Legislation, Legislative Timing, Security and Development, Relational-construction Paradigm

摘要： 我国人工智能立法始终面临是否立法、何时立法、以何种形式立法以及立法如何处理安全与发展关系四重争议,这直接关涉人工智能立法的功能定位与范式选择。本文主张尽快推进我国人工智能的综合性立法,对现行人工智能分散治理充分性的肯定、对体系融贯的质疑均不足以否定综合性立法需求;与此同时,法律基础设施理论与二阶决策理论,从功能与结构两个维度共同表明了综合性立法是匹配人工智能治理复杂性的必要方案。就立法时机而言,人工智能立法不存在绝对最佳时点,应当基于“先立后优”思路破解“科林格里奇困境”。就立法价值而言,立法与创新并非零和博弈,法律应当积极发挥对人工智能发展方向的形塑功能,我国人工智能立法不能只是技术管制法或产业促进法,而应是一部以人为本的发展法。就立法范式而言,智能体技术的发展使既有风险规制范式面临根本挑战,立法应在风险规制基础上推进至关系建构层面,对人智关系、智智关系与复合关系作出基础性制度安排,构建具有中国特色的人工智能法律体系。

关键词: 人工智能, 综合性立法, 立法时机, 安全与发展, 关系建构范式