Is it Possible to License Works Used in the Learning Process of Artificial Intelligence Algorithms?

Abstract

One of the more pressing problems of contemporary copyright law is the current status concerning the legality of using copyrighted works and objects of related rights to train generative AI. This article describes the process of training AI focusing on how copyrighted works are used. A comparison is made between the phenomenon known as Text and Data Mining (TDM) and the AI training process, which leads to the conclusion that TDM is only one of the techniques used in the AI training process. Therefore, the legal regulations contained, for example, in the Directive (EU) 2019/790  on copyright and related rights in the Digital Single Market (CDSM)  concerning TDM do not solve the problem of copyright infringement by generative AI producers. The authors consider evaluating the possibility of licensing works in the AI training process; however, towing to the difficulty in identifying the list of works that an AI algorithm uses to acquire certain information, traditional copyright and related rights licensing models cannot be applied. Therefore, the authors postulate that a solution similar to, the so-called, Private copying levy in Article 2 and Article 5(2b) of Directive 2001/29/EC InfoSoc should be introduced into the copyright system. In their opinion, only this solution will provide copyright holders with fair compensation for the exploitation of their works and objects of related rights. The proposed solution, according to the authors, could operate parallely with other licensing models as long as such solutions are formed in certain areas related to the AI training process.

Keywords: Artificial Intelligence, Copyright Law, Data Mining, InfoSoc