This research critically analyses the problem of allocating authorship in the creative output from the perspective of EU copyright law. Various sources including academic literature and EU case law have established that authorship is a strictly human trait. Up until recently the question of authorship, while still being highly relevant, did not however raise major problems for computer-generated works. In this respect, two candidates for the final authorship claim emerge – the software programmer and the user of the software. Major developments in the field of machine learning have however challenged this established status quo. For that reason, the question of copyright authorship has surfaced again in a slightly different context.
That said the hype around artificial intelligence often confuses the audience. In order to fully appreciate and tackle the emerging legal issues in this discourse, one should have a good grasp of how the technology at hand actually works. However, taking into account the particularities of the machine learning process this research concludes that the causal link between the creative choices of either of these two human candidates is far detached from the final creative elements in the product. Consequently, in the case of ML-generated works the final creative output remains “authorless”, as Professor Ginsburg puts it.
This work strives to analyse the issues from the perspective of the communications theory as advanced by Professor Drassinower and Dr. Craig (Abraham Drassinower, What’s Wrong with Copying? (Harvard University Press 2015) and Carys J Craig, Copyright, Communication and Culture: Towards a Relational Theory of Copyright Law (Edward Elgar Publishing 2011)). Considering that there is no clear EU theory of copyright law, which drives the legislative wheel, this work seeks to devise one. Based on the premise that there is nobody to converse with, this theory purports that copyright law should not arise in the context of such authorless works. In order to do that, my work looks at the technical side of machine learning, which is then matched to the communications theory.
My approach examines topics of authorship, ownership and originality, which are certainly logically connected when one is posed with the question of whether machine learning-generated works should be protected with copyright law. The underlying policy considerations of my thesis are economic and pose the vital question of whether any market failure will indeed materialise if such works remain in the realm of the public domain.
The scope of this work is on EU law and it only studies the issues from a copyright perspective. Various other tangent topics are indeed of interest, such as the patentability of such works and the potential protection with a database right. The aim of my work is to study the status quo from a pure copyright perspective, i.e. focusing on the authorial claims. In the discussion, however, the work touches upon the potential protection of such machine learning-generated works with a neighbouring right. However, before moving into that direction, clear evidence of the above-mentioned market failure needs to be demonstrated.