生成式人工智能训练数据侵权风险与法律应对  

Risks of Infringement of Generative Artificial Intelligence Training Data and Legal Countermeasures

作　　者：郭德忠[1] 张云蔚 GUO De-zhong;ZHANG Yun-wei(School of Law,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]北京理工大学法学院,北京102401

出　　处：《湘潭大学学报（哲学社会科学版）》2024年第5期78-86,共9页Journal of Xiangtan University:Philosophy And Social Sciences

摘　　要：生成式人工智能模型训练数据的来源、使用和输出均有一定的不可控性,故其在训练中的各个环节和不同阶段均存在侵权风险。传统法律制度中的法定许可、强制许可无法适用于大规模训练数据获取和使用情形,合理使用条款虽涉及特定情形下的责任豁免例外,但并未对训练数据行为本身进行有效规制。“三步检验法”需要在司法实践中进一步探索各要素的具体认定标准和裁量规则。域外“转化性使用”标准虽可资借鉴,但其标准的具体考量因素亟待甄别。有必要完善相关领域的法律体系及其配套措施,促进人工智能技术发展,保障版权人的合法权利,进而协调好二者之间的利益关系。Potential infringement risks arise at various stages of the training process due to the unpredictability inherent in the sources,usage,and outputs of training data for generative AI models.The acquisition and use of large-scale training data do not fall under the scope of statutory and compulsory licenses within traditional legal systems.Although exemptions from liability are covered in specific cases by fair use provisions,these do not effectively regulate the behavior of training data itself.Further exploration in judicial practice is required to determine specific criteria and discretionary rules for each element of the“Three-Step Test”.While the extraterritorial“transformative use”standard can serve as a reference,the specific considerations for this standard urgently need to be identified.It is essential to enhance the legal system and its supporting measures in related fields to foster the development of AI technology and to protect the legitimate rights of copyright holders,thereby balancing the interests of both parties.

关 键 词：生成式人工智能 训练数据 合理使用 转化性使用 利益平衡 

分 类 号：D923.99[政治法律—民商法学]