Author Bio: Ding Xiaodong is a professor and doctoral supervisor at the School of Law of Renmin University of China, and deputy director of the Future Rule of Law Institute.

Abstract: The copyrightability of works by AI like ChatGPT and Wenxinyiyan not only requires positive law and normative doctrinal analysis，but also requires deep jurisprudential reflection. If the use of copyrighted works as training data could be considered as fair use，then AI works should be considered as within the public domain，while at the same time law could provide unfair competition and sui generis rights protection for AI works. If copyrighted works as training data could not be fairly used，however，AI enterprises should be granted economic rights under copyright law. The traditional copyright theory is based on subjective creativity and human－computer dichotomy，which is not uncontroversial even in the pre－AI era，and is even more untenable in the AI era. Copyright law should break away from the “romantic author”. It should protect AI works based on specific institutional functions，rather than the all or nothing application of the entire copyright rules. The law should endow artificial intelligence enterprises with initial signature rights and provide reputation incentives for enterprises．For artificial intelligence works with high public risks，the obligation of signature in public law should be imposed. The deconstruction and reconstruction of the modular copyright institution not only provides more reasonable protection for artificial intelligence works，but also deepens the understanding of copyright itself.

Key Words: Artificial Intelligence Works; Incentive Theory; Ｒomantic Author; Modularity Theory; Signature