生成式人工智能数据训练中的“非作品性使用”及其合法性证成

"Non-Work Use"Nature of Generative Artificial Intelligence Data Training and Its Legitimization

生成式人工智能数据训练中的作品使用行为是否应当纳入著作权权利范围,关涉到产业发展的重要利益关系。数据训练中的作品使用,具有“非特定性”,不指向具体而特定的单个作品,从功能阶段上看,属于生产过程性的中间使用;从后续效果上看,对于生成式人工智能这一基础工具具有增强效应。因此,应当将其界定为“非作品使用行为”而排除在著作权权利范围之外。生成式人工智能应用对著作权人并未造成整体激励的削弱,只是导致了激励行为的结构性调整,因而不需要将著作权保护延伸至数据训练来予以补偿。著作权法下排除保护训练数据“非作品性使用”的具体方案选择中,从权利范围“直接排除”保护的方案总体上优于“先进后出”的合理使用方案。

Whether or not the use of works in generative Al data training should fall within the scope of copyright rights is an important stake in industrial development.The use of works in data training is"non-specific"and does not point to a specific individual work;from the perspective of functional stage,it belongs to the intermediate use in the production process,and from the perspective of subsequent effect,it has an enhancement effect on the basic tool of generative artificial intelligence.Therefore,it should be defined as"non-work use behavior"and excluded from the scope of copyright rights.Generative AI applications do not weaken the overall incentives for copyright holders,but only lead to a structural adjustment of incentives,and thus do not need to be compensated by extending copyright protection to data training.Of the specific options available under copyright law to exclude protection for"non-work uses"of training data,the"direct exclusion"from the scope of rights is generally superior to the fair use option.