The Problem of Licensing Works for the Training Process of Artificial Intelligence Algorithms

Abstract

One of the significant challenges in modern copyright law is determining the legal status of using copyrighted materials and related rights for training generative artificial intelligence (AI). This article examines how copyrighted works are utilized during AI training. It contrasts text and data mining (TDM) with AI training, concluding that TDM is just one of the techniques involved. Consequently, the legal frameworks in Directive 2019/790 on copyright and related rights in the Digital Single Market addressing TDM do not resolve the issue of copyright violations by generative AI developers. From a technical point of view, TDM and AI operations should be clearly distinguished from each other. Using works for AI training should be viewed as a new, unregulated field of exploitation of works and related rights beyond TDM. Therefore, in most cases, AI training will lead to the infringement of the economic rights of the creators and their successors. The authors explore the possibility of licensing works for AI training; however, the complexity in identifying which works an AI algorithm uses makes traditional copyright and related rights licensing models impractical. The solution is to introduce a statutory compensation model based on the assumption that AI producers use copyrighted works in the learning process. It is therefore postulated that a legal regulation should be introduced which would allow AI algorithm developers to use the works legally in the absence of licensing options. Therefore, the authors suggest implementing a system similar to the private copying levy found in Article 2 and Article 5(2b) of Directive 2001/29/EC. This approach is necessary to ensure fair compensation for copyright holders whose works and related rights are exploited.