关键词: 人工智能； 数据训练； 著作权侵权； 著作权合理使用

Abstract: Using a large number of works to training large AI model will cause many copyright infringement risks. However, it does not damage the value of the works and the legitimate rights and interests of the copyright owner. Including the behavior of AI training works to the scope of fair use of copyright is a necessary mechanism to resolve such risks, and also a necessary choice to promote digital technology innovation and the development of AI industry. In this regard, the explanatory theory has some applicable space, but it also has some defects. It is more desirable to construct the fair use rules of AI training works based on the legislative theory. On the way of legislation, the route of restricting copyright is better than the route of lawful use of data. In the legislative arrangement, China does not need to distinguish the training works behavior between the scientific research type and the commercial type. Legislation should ensure the legitimacy of the acquisition behavior of the works, and should not expand the exception of “prohibit avoidance technology measures” at this stage. The use of works shall be limited to the copy and translation of works during the training phase, and should not cover the communication over information networks. In addition, it is necessary to amend the obligation to sign the author to the AIGC.

Key words: artificial intelligence; data training; copyright infringement; fair use of copyright