生成式人工智能时代文本与数据挖掘的著作权合理使用规则构建

Construction of Fair Use Rules for Copyright in Text and Data Mining in the Era of Generative Artificial Intelligence

摘要: 生成式人工智能模型在数据输入及训练阶段需依赖对海量文本、图像等作品的复制与预处理，然而我国现行著作权法却并未为此类行为提供合法基础，导致相关作品的复制权、演绎权等著作权侵权风险不断累积。文本与数据挖掘版权例外规则作为一项平衡技术发展与著作权人利益的制度创新，逐渐在国际社会得到认可与广泛应用。本文以欧盟已经相对成熟的文本与数据挖掘版权例外规则为蓝本研究，探寻这一制度的具体内容和司法实践效果并进行中国镜鉴，最后提出把文本与数据挖掘行为纳入著作权法合理使用情形的合法路径，同时建议配套必要的技术措施、设置开发者披露义务，并增加著作权人权利平衡和补偿机制。

Abstract: Generative artificial intelligence models, in the data input and training stages, rely on copying and pre-processing vast amounts of works such as texts and images. However, China’s current Copyright Law does not provide a lawful basis for such activities, leading to a continual accumulation of copyright infringement risks relating to reproduction and adaptation rights of the relevant works. The copyright exception rules for text and data mining (TDM), as an institutional innovation balancing technological development with the interests of copyright owners, have gradually gained international recognition and widespread application. This paper takes the EU’s comparatively mature TDM copyright exception framework as its blueprint, investigates the specific content of this regime and its judicial practice effects, and offers a Chinese mirror assessment. Finally, it proposes a lawful path to incorporate TDM activities within the legitimate uses under copyright law, while also recommending the accompanying technical measures, the introduction of developers’ disclosure obligations, and the strengthening of balance and compensation mechanisms for copyright owners.