EU Copyright Law and Machine Learning: A Net of Authorship Claims

This thesis analyses EU copyright authorship through the lenses of machine learning (ML). More specifically, it studies the problem of allocating authorship in the creative output of a ML process. Its primary focus are literary, musical and artistic works generated though a ML process. The research question entails two dimensions. The substantive dimension analyses whether EU copyright law protects ML-generated works, while the normative one studies whether EU copyright law should protect ML-generated works. The first dimension has been addressed by many scholars, who focus on the theoretical justifications for copyright law. The main and mostly shared view is that copyright does not subsist in ML-generated works due to the absence of an author whose free and creative choices are evident in the creative process and product. The added value of this thesis, which brings it beyond the state of the art, is that it analyses thoroughly the technology behind ML-generated art in the field of computational creativity. Thanks to this technical analysis, this thesis maps authorship claims in the ML process. In that sense, it provides new evidence to support the main argument that copyright law does not protect ML-generated works due to the absence of an originality causation. The normative dimension studies the EU legislative competences in the field of copyright law. To that end, cultural considerations, while present, have taken a backseat and let the flexible internal market goal take the lead. Article 114 TFEU is the main legal basis, but it lacks any normative content. It can thus serve as a convenient legal basis to push the EU legislative agenda towards protection of ML-generated works with copyright law. Following this direction is risky since at this stage, there are no authoritative and thorough impact assessments identifying the clear need to extend copyright law to cover such emergent works. This thesis advocates for a cautious evidence-based approach to be adopted by the EU legislator, which puts at the centre stage the balanced internal market, guided by procedural safeguards such as public consultations and impact assessments, but also the subsidiarity and proportionality principles. The methodology of this thesis is underlined by explanatory and evaluative techniques. The first research objective is to provide new evidence supporting the general argument that EU copyright law does not protect ML-generated works. In pursuing this explanatory goal, the thesis engages comprehensively in the technicalities of the ML process and explains why the ML process would not always result in copyright protected works. It does not strive to merely describe the protectability criteria under EU copyright law. Instead, it engages in a profound law and technology assessment by dissecting the ML computational creativity process. The second research objective engages in a normative assessment by arguing that the functional competence of Article 114 TFEU should not be manipulated to the extent that copyright protection expands to cover such emergent works where the human author is too detached from the creative process. In pursuing this evaluative research objective, this thesis aims to assess the limits of the EU’s internal market competence in view of the ML-copyright law debate.