生成式人工智能的法律规制——从技术与人机关系出发  

Basic Issues and Core Mechanisms of Legal Regulation of Generative Artificial Intelligence——From the Perspective of Technology and Human-Machine Relationship

作　　者：姜沣格[1] 于海防[2] JIANG Feng-ge;YU Hai-fang

机构地区：[1]山东工商学院法学院,烟台264005 [2]烟台大学法学院、烟台大学数字法治研究中心,烟台264005

出　　处：《求是学刊》2025年第2期123-137,共15页Seeking Truth

基　　金：山东省社会科学规划专项“网络格式缔约中意思表示真实性的保障研究”。

摘　　要：生成式人工智能在智能水平、自主性等方面获得巨大提升,基于对其技术与人机关系特性以及法律问题成因的分析可以发现,生成式人工智能在承继以往人工智能宏观、微观法律问题之外,又在技术层面、事实层面、价值层面产生大量新问题。生成式人工智能法律规制应当确立局部替代人类、整体增强人类的价值理念,以监管机制为中心,分阶段采取强度不同的监管模式,并将人类价值观做技术嵌入。沿用《数据安全法》上的重要程度、危害程度作为人工智能分类分级的标准,建立分类分级机制以及技术标准机制。在此基础上,根据生成式人工智能的技术流程,建立针对起始阶段的研发机制、针对中间阶段的应用机制以及针对最终阶段的责任机制,并配以强制责任保险与社会救助基金机制。Generative Artificial Intelligence(AI)has achieved significant improvements in aspects such as intelligence level and autonomy.Based on the analysis of the characteristics of technology and human-machine relationship as well as the causes of the legal issues,it can be found that generative AI has generated a large number of new problems at the technical,factual,and value levels in addition to inheriting the previous macro and micro legal issues of AI.The legal regulation of generative AI should establish the principle of replacing human beings partly and enhancing human beings integrally.With the supervision mechanism as the center,the regulation should adopt supervision modes of different intensities in stages and embed human values technically.Following the importance degree and harm degree of the Data Security Law as the criteria for classification and grading of AI,the classification and grading mechanism as well as the technical standard mechanism should be established.On this basis,according to the technological process of generative AI,a research and development mechanism for the initial stage,an application mechanism for the intermediate stage,and a liability mechanism for the final stage are established,supplemented by the mechanisms of com-pulsory liability insurance and social assistance fund.

关 键 词：大语言模型 生成式人工智能 人机关系 法律规制 人工智能法 

分 类 号：D92[政治法律—法学]