论通用人工智能训练数据版权侵权之归责原则

The Principle of Liability for Copyright Infringement of AGI Training Data

生成式人工智能作为通用人工智能的起点,在为数字经济带来极大潜力的同时也带来了版权侵权风险,推进技术发展和保护文艺创作的难题凸显。转变以往的末端视角,训练数据作为通用人工智能发展壮大的基础前提,解决其“合法、海量、高质”问题十分重要。对训练数据的高要求和现实开发中的难获得,导致通用人工智能训练数据处理中不可避免地存在侵犯作者复制权、翻译权、改编权等风险。当前的规制思路缺乏对通用人工智能自身特性和商业模式的关注,侵权归责原则不够清晰。以鼓励通用人工智能基础大模型发展为价值导向,采取过错责任归责原则并区分风险层级规定不同程度的注意义务,在制度与技术交互中推进良法善治。

As the starting point of general artificial intelligence,generative artificial intelligence brings great potential to the digital economy,but also brings the risk of copyright infringement,and the problem of promoting technological development and protecting literary and artistic creation becomes prominent.Change the previous end perspective,training data as a basic premise for the development and growth of general artificial intelligence,to solve its“legal,massive,high quality”problem is very important.The high demand for training data and the difficulty of obtaining it in practical development make it inevitable that the training data processing of general artificial intelligence will infringe the author’s right of reproduction,translation and adaptation.However,the current regulatory ideas still lack attention from the characteristics and business model of general artificial intelligence,and the principle of tort liability is confused.With the value orientation of promoting and encouraging the development of general artificial intelligence basic large model,the principle of fault liability is adopted,and different degrees of duty of care are stipulated by distinguishing risk levels,so as to promote good law and good governance in the interaction between system and technology.