类型化视角下人工智能数据训练著作权规则研究
Research on Copyright Rules for Artificial Intelligence Data Training under the Perspective of Typology
王静1
作者信息
摘要
人工智能数据训练加速技术迭代,驱动产业升级.然而,可版权性作品的获取与利用使其难免著作权侵权争议.数据训练特性、产业发展需求以及域外规制动向充分说明,人工智能数据训练合理使用化是大势所趋.决策式人工智能与生成式人工智能在运行原理与输出形式上存在显著区别,著作权法应当实施类型化规制.前者利用数据的功能性价值,构成"转换性使用",后者利用数据的表达性价值,应当将其纳入可"选择退出"的合理使用情形.在制度配套设计上,需明确人工智能开发者的数据训练披露义务以保证著作权人"退出"可能性,同时针对强化学习中使用者的数据训练行为,构建人工智能开发者侵权提示责任减轻机制,实现使用者与开发者的责任分担.
Abstract
Artificial intelligence data training accelerates technical iteration and drives industrial upgrading.However,the acquisition and utilization of copyrightable works often lead to copyright infringement disputes.Data training characteristics,industrial development needs and extraterritorial regulatory trends fully demonstrate that the fair use of artificial intelligence data training is a general trend.There is a significant difference between decision-making artificial intelligence and generative artificial intelligence in terms of operation principle and output form,so the copyright law should implement a typological regulation.The former utilizes the functional value of the data,which constitutes"transformative use",while the latter utilizes the expressive value of the data,which should be included in the fair use situation of"opt-out".In terms of system design,it is necessary to clarify the obligation of artificial intelligence developers to disclose data training to ensure the possibility of copyright holders to"opt out",and at the same time,for the user's data training behavior in reinforcement learning,it is necessary to reduce the liability of artificial intelligence developers when they make infringement cues,so as to realize the liability sharing between users and developers.
关键词
数据训练/决策式人工智能/生成式人工智能/合理使用/选择退出Key words
Data Training/Decision-Making Artificial Intelligence/Generative Artificial Intelligence/Fair Use/Opt-Out引用本文复制引用
基金项目
2021年度国家社科基金重大项目(21&ZD142)
出版年
2024
NETL
NSTL
万方数据