重塑复制权:生成式人工智能数据训练的合法化路径

Restructuring Reproduction Rights:Legal Path for Data Training in Generative Artificial Intelligence

脱离了适宜复制权生存的原始环境,而对复制权进行宽泛的权利扩张,是导致生成式人工智能数据训练遭遇法律障碍的根本原因。以增进公众福祉和产业发展为考量,合法化数据训练中复制作品的行为,解决之道在于重塑复制权,而不是重塑合理使用。在权利法的视角下,重塑复制权应对复制权的功能进行重新定位,以“固定性+传播性”的构成要件定义侵犯复制权的行为。在技术发展的早期阶段,重塑复制权后的合法化路径还应配合著作权注意义务的施加,促进平台不断改进用户指令端和内容输出端的技术控制和内容生成能力。在指令端及输出端符合著作权注意义务的阶段性场景要求,以及获取作品的行为没有违背权利人的意愿时,数据训练中对作品的复制可视为不具备“传播性”而不构成对复制权的侵犯。

The broad expansion of reproduction rights,detached from the original environment conducive to their existence,is the fundamental reason why data training for generative artificial intelligence faces legal obstacles.To enhance public welfare and industry development,legalizing the reproduction of copyrighted works in data training necessitates restructuring reproduction rights rather than rethinking fair use.From the perspective of rights law,restructuring reproduction rights should reposition the functions of these rights,and define the infringement of reproduction rights in terms of the constitutive elements of"fixity and transmissibility".In the early stages of technological development,the restructuring of reproduction rights should also refer to the duty of care in terms of copyright to encourage platforms to continuously improve their technical controls and content generation capabilities,both at the users'instruction end and at the content output end.When the practices at the instruction and output stages meet the requirements of duty of care in copyright,and the act of obtaining works does not contravene the rights holder's intent,reproducing works for data training can be considered devoid of"transmissibility",thus not an infringement of reproduction rights.