ARTIFICIAL INTELLIGENCE AS A CHALLENGE TO COPYRIGHT LAW

Abstract

The authors of this paper examine whether creations generated by artificial intelligence based on user prompts and instructions can be considered copyrightable works, and whether existing copyright limitations provide a sufficient legal basis for using copyrighted works to train AI systems. It is well known that generative AI, based on brief user inputs and instructions, can produce various forms of content such as text, sound, photographs, drawings, videos, and more. The result can resemble a literary, musical, visual, or cinematic work. However, AI operates through a form of statistical prediction and, as such, cannot imprint its own specific and individual mark on its creations, simply because it is not a human being and lacks an individual spirit. Considering this, such creations are not regarded as copyrighted works. Furthermore, legislation and legal literature generally agree that an author is a natural person who has created a work. The authors assess that this fundamental principle of copyright law is unlikely to be challenged in the foreseeable future. As for copyright limitations—at least within our legal system—they alone do not constitute a sufficient legal basis for training AI systems. Regarding the use of copyrighted works for AI training purposes, the authors conclude that model developers must obtain permission from rights holders or strictly adhere to one of the existing statutory copyright exceptions.