Artificial Intelligence: Copyright & Consequences

Abstract

This thesis explores the complex interplay between generative artificial intelligence (AI) models and their usage of copyrighted materials from the perspective of the legal and economic implications that originate from this practice. The normative focus is on ensuring that the copyright holders are protected, while balancing their rights with those of the AI developers who generate such models. This thesis begins by explaining the legal and economic rationale for copyright protection, and the economic value from both copyrighted materials as well as AI-generated ones. The impact of using copyrighted material in AI models on the rights and interests of creators, users, and owners is then critically examined, unveiling a myriad of ambiguities and uncertainties.

The thesis then delves into the worldwide, EU, and Member State level copyright legislation, noting that there is a need for more specific regulation on all levels. The three-step test of copyrightability unraveled through the legal analysis is deemed to be too vague for legal use, whereas the highly codified copyright exceptions previously known in the EU do not provide necessary guidance either. Given this context, the EU legal framework is analyzed and compared against the more flexible US framework, which is based on fair use principles. The thesis proceeds to argue that although a US-inspired framework would require grave changes to the EU legal system, it may benefit all stakeholders involved by providing a clearer, and quicker system to work with.

In chapter five, in light of the conclusions of Chapter four, recommendations are made to combat the challenges discussed throughout this thesis. They are ranked on their ability to be successfully implemented, as well as their enforceability from an EU perspective.

The assessment was conducted within the context of a regulatory flexibility framework, as it was observed that this approach would offer the optimal means of determining efficacy through the implementation of a regulatory sandbox. Well-scoring recommendations here relate to the opt-out provision, new licensing schemes, and worldwide regulations. Whereas DRM (digital rights management), and the fair use principles were more difficult to square with the said perspective. Finally, chapter six concludes by drawing the threads together of all points discussed in this thesis.