解读《生成式人工智能服务管理暂行办法》第9条第1款

Interpret Article 9, Paragraph 1 of the Interim Measures for the Administration of Generative Artificial Intelligence Services

摘要: 《生成式人工智能服务暂行办法》第九条第1款明确规定：“提供者应当依法承担网络信息内容生产者责任，履行网络信息安全义务。涉及个人信息的，依法承担个人信息处理者责任，履行个人信息保护义务。”通过对该条款以及相关规定进行对比，可以发现义务主体与责任主体不一致，由此对生成式人工智能服务提供者承担网络信息内容生产者责任的合理性问题进行反思。要求某一主体承担某一责任，前提需要该主体有相应义务。从服务提供者的义务来源即其与生成式人工智能技术的密切关系出发，分析生成式人工智能技术原理与特点，阐明内容的生成实质上受到数据、算法、人机交互等众多因素的共同影响，从技术层面指出立法者将承担内容侵权责任的主体归为生成式人工智能服务提供者是基于其对数据收集与算法设计具有较强控制力。然而，仅就这两方面的理由没有满足由服务提供者承担内容生产者责任的充分必要条件，并没有全面考虑在生成内容的过程中还涉及到服务使用者的影响以及服务提供者对最终的内容生成起到的控制力有限的情况。对该条款的解读还应当回到法律文义解释上，这要求必须透彻理解“生成式人工智能服务提供者”与“网络信息内容生产者责任”两个关键词。相关规定对“服务提供者”的定性不同，定性决定主体的法律地位，进而决定相关主体承担何种法律责任。因此，生成式人工智能服务提供者的法律地位的认定是重中之重，《生成式人工智能服务暂行办法》将其定性为“内容生产者”，但是由于生成内容的控制力方面二者具有显著不同，使得将二者等同缺乏合理依据，进而要求生成式人工智能服务提供者承担内容生产者责任缺乏正当性，立法者在制度设计上忽视其他主体尤其是生成式人工智能服务使用者通过“指令”对内容生成发挥的作用，进而只要求服务提供者承担全部内容侵权责任缺乏合理依据。

Abstract: Article 9, Paragraph 1 of the Interim Measures for Generative Artificial Intelligence Services clearly stipulates: “Providers shall assume the responsibility of producers of network information content in accordance with the law and fulfill the obligation of network information security. Where personal information is involved, it shall assume the responsibility of a personal information processor in accordance with the law and fulfill the obligation to protect personal information.” By comparing this provision with the relevant regulations, it can be found that the subjects of obligation and responsibility are inconsistent, thereby reflecting on the rationality of generative artificial intelligence service providers bearing the responsibility of producers of online information content. To require a certain subject to undertake a certain responsibility, it is necessary that the subject has corresponding obligations. Starting from the source of service providers' obligations, that is, their close relationship with generative AI technology, analyze the principles and characteristics of generative AI technology, and clarify that the generation of content is essentially influenced by many factors such as data, algorithms, human-computer interaction, etc. From a technical perspective, it is pointed out that legislators attribute the subject of content infringement liability to generative AI service providers based on their strong control over data collection and algorithm design. However, the reasons for these two aspects alone do not meet the necessary and sufficient conditions for a service provider to assume the responsibility of a content producer, and do not fully consider the influence of the service user involved in the process of generating content and the limited control that the service provider has over the final content generation. The interpretation of this provision should also return to the literal interpretation of the law, which requires a thorough understanding of the two key terms “generative AI service provider” and “network information content producer responsibility”. The relevant provisions have different characterizations of “service providers”, which determine the legal status of the subject and thereby determine what kind of legal liability the relevant subject bears. Therefore, the determination of the legal status of generative AI service providers is of Paramount importance. The Interim Measures for Generative AI Services categorizes it as a “content producer”, but due to the significant differences in the control of generated content, there is no reasonable basis for equating the two, and thus it is not justified to require generative AI service providers to assume the responsibility of content producers. Legislators’ institutional design neglects the role of other subjects, especially users of generative AI services, in generating content through “instructions”, and thus only requires service providers to bear full liability for content infringement.