生成式人工智能服务中“信息错误”的民事责任  

On the Legal Liability of"Information Errors"in Generative Artificial Intelligence Services

作　　者：郭金良[1] Guo Jinliang

机构地区：[1]辽宁大学法学院

出　　处：《政法论坛》2025年第2期25-35,共11页Tribune of Political Science and Law

基　　金：2025年度辽宁省经济社会发展研究课题“跨领域知识驱动数字金融发展的法治保障研究”(2025lSLYBWZZKT-030);辽宁省属本科高校基本科研业务费专项“助力辽宁高质量发展中金融风险治理的法律保障研究”(LDJBKYSK2024008)的阶段性研究成果。

摘　　要：作为人工智能技术的创新应用,ChatGPT在生成成果上的创造性特征引发了社会的广泛关注,“信息”是服务的核心要素。在生成式人工智能服务提供过程中可能产生“信息错误”,包括训练数据的信息错误、技术缺陷导致的信息错误、服务使用者提示的信息错误以及服务提供者故意造成的信息错误。“信息错误”可能产生违约责任和侵权责任,传统民事责任规则的适用存在诸多法律困境,如ChatGPT是否具有法律责任主体资格的争论、违约行为认定标准模糊、侵权责任归责原则适用不合理、因果关系及损害的认定规则不清晰等。应当以民事责任控制理论为基础,重塑生成式人工智能服务民事责任制度。一方面,通过明确服务提供者与使用者之间的协议类型,明确不同“信息错误”情形下违约责任的规则适用;另一方面,侵权责任规则的塑造包括确立侵权责任归责原则分类适用规则,对不同类型的侵权行为确定差异化的因果关系认定规则,通过“其他合理方式计算”合理认定间接损害。As an innovative application of artificial intelligence technology,ChatGPT's creative features in generating results have attracted widespread societal attention,and"information"is the core element of the service."Information Errors"may occur in the process of the provision of generative artificial intelligence services,including errors in training data,errors caused by technical defects,errors prompted by service users,and intentional errors caused by service providers."Information Errors"may result in breach of contract liability and tort liability.The application of traditional civil liability rules faces many legal dilemmas,such as disputes over whether ChatGPT qualifies as a legal subject of liability,vague criteria for determining breach of contract,unreasonable application of the principle of liability attribution for tort,and unclear rules for determining causality and damages.Based on the theory of civil liability control,the generative artificial intelligence service civil liability system should be reshaped.On the one hand,by clarifying the types of agreements between service providers and users,the rules for breach of contract liability in different "Information Error"situations can be clarified;On the other hand,the shaping of tort liability rules includes establishing classification and application rules for tort liability attribution principles,determining differentiated causal relationship determination rules for different types of infringement acts,and reasonably determining indirect damages through"other reasonable calculation methods".

关 键 词：生成式人工智能 信息错误 违约责任 侵权责任 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] D922.17[自动化与计算机技术—控制科学与工程] D923[政治法律—宪法学与行政法学]