UNAUTHORISED USE OF COPYRIGHT WORKS FOR TRAINING AI SYSTEMS IN THE EU AND THE UK: COPYRIGHT INFRINGEMENT AND LIABILITY ISSUES

Abstract

perform various tasks and an instrumental part of the realisation of this objective are the techniques of machine learning and deep learning. The successful operation of these programming techniques relies on the availability of training data and that training data often includes copyright works. When such works are used for the purpose of training AI systems, this gives rise to the questions whether this use of copyright works constitutes a copyright infringement and if so, who is liable for the infringement. The present dissertation discusses these questions from the perspective of EU and UK copyright law. It traces the ways in which copyright works may be used at the different stages in the process of training AI systems and the circumstances in which such uses should be regulated by EU and UK copyright law. It then discusses whether under EU and UK copyright law the unauthorised use of copyright works interferes with the reproduction right and/or other exclusive rights and whether it can be legitimised by the relevant existing copyright exceptions – the temporary copy and the text and data mining (TDM) exceptions. Finally, the dissertation suggests as to how the liability for the infringement of copyright through the unauthorised use of copyright works should be allocated.