生成式人工智能数据训练中的版权问题研究

Research on Copyright Issues in Generative Artificial Intelligence Data Training

利用作品对生成式人工智能进行数据训练能够促进公共利益的最大化,这与版权法的价值目标相契合。在输入阶段,版权合规型AI通常生成不侵犯版权的内容,此时的数据输入行为具有转换性,根据四要素标准应构成合理使用。版权违规型AI通常生成侵犯版权的内容,此时的数据输入行为构成侵权。在输出阶段,在生成物构成侵权内容的情况下,若AI运营商采取了事前预防措施和事后纠正措施,则其可以被认定为无过错而免于承担赔偿责任。在司法实践中,我国法院应灵活运用四要素标准认定数据输入行为是否构成合理使用,并对于尽到合理注意义务的AI运营商减免其因输出侵权内容所承担的赔偿责任,从而推动人工智能产业的科技向善。

Utilizing works for data training in generative artificial intelligence can maximize public interest,aligning with the value objectives of copyright law.In the input stage,copyright-compliant AI typically generates non-infringing content,and the data input behavior is transformative,which,according to the four-factor test,should constitute fair use.In contrast,copyright-violating AI often produces infringing content,in which case the data input behavior constitutes infringement.In the output stage,if the generated content constitutes infringement,AI operators who have taken preventive measures beforehand and corrective actions afterward may be deemed faultless and thus exempt from liability.In judicial practice,Chinese courts should flexibly apply the four-factor test to determine whether data input behavior qualifes as fair use,and reduce or exempt the liability of AI operators who have fulfilled reasonable duty of care for infringing content outputs.This approach would foster the responsible development of the AI industry.