Byte the Beat: Copying or Creating?

Sammanfattning

Artificial intelligence (AI) and text and content-generating tools have been around for a while, bringing significant changes, challenges, and opportuni-ties, especially regarding copyrighted materials such as music. As AI sys-tems have become more sophisticated in generating new creative works, questions about these systems' potential copyright infringement have gained prominence. This thesis delves into the legal landscape where creativity and technology intersect and explores the complexities surrounding copyrighted music's use in training AI models and the subsequent creation of new musical out-puts. The central focus is whether it constitutes a copyright infringement under EU law when generative AI mimics an artist's sound and style. To address this, the thesis will examine how EU law handles the use of copyrighted musical works in training AI models and how EU law address-es the copyright status of AI-generated musical outputs that imitate existing artists. These questions are explored through a legal dogmatic method. A compara-tive perspective is also incorporated to a lesser extent to examine how dif-ferent legal frameworks address these issues. Musical works and their related right, such as phonograms, are generally protected by law, granting rightsholders certain exclusive rights, such as the right to reproduce the work. However, when AI generates music, the musi-cal pieces are often reproduced multiple times in the training phase of the AI system. The thesis finds that these reproductions are not covered by the exemption for temporary acts of reproduction in Article 5(1) InfoSoc Di-rective. The exemption for text and data mining in Article 3 and Article 4 of the Directive (2019/790) on copyright and related rights in the Digital Single Market (DSM Directive) can potentially cover some reproductions of copyrighted musical works. Still, these exemptions are narrow in scope. Article 3 applies research organisations and cultural heritage institutions that conduct text and data mining activities on a non-commercial basis. Moreover, the general exemption in Article 4 faces practical hindrances, as rightsholders can reserve their right to prevent their works from being mined. Therefore, the mentioned provisions are not sufficient for AI music generators to train their generative models on copyright-protected music in most cases. AI-generated output may constitute a derivative work. However, each case must be assessed individually, as some AI-generated musical outputs might include samples of existing works, whilst others may only be similar in style, which generally is not protected by copyright. The AI-generated voice is also typically not protected by copyright, as the voice is not con-sidered fixed in a tangible medium or reflective of the author's own intel-lectual creation in most circumstances. Thus, while the output may stay within the legal framework, the input often does not comply with EU copy-right laws, rendering the overall process of generative AI mimicking an artist's sound and style unlawful.