A Scanner Darkly: Copyright Liability and Exceptions in Artificial Intelligence Inputs and Outputs

Abstract

Abstract This article delves into the legal issues surrounding the use of copyrighted works in training artificial intelligence (AI). It examines two critical questions: firstly, whether accessing and analysing copyrighted works for AI training constitutes copyright infringement; and secondly, whether outputs generated by AI from these inputs infringe on copyright. The primary focus is on the United Kingdom’s jurisdiction, with comparative analyses from EU law and a few US cases. The article aims to bring clarity to these multifaceted legal issues by thoroughly exploring the technicalities of machine learning and its implications for ongoing and future litigation. The hypothesis suggests that most inputs may fall under existing exceptions and limitations, but the outputs’ legality may depend on individual case specifics. The conclusion reflects on the inevitable legal challenges that accompany technological advances, as happened in the past with file sharing and the early World Wide Web. The article highlights the current stage of AI technology and law, suggesting that legal precedents or legislative actions might eventually settle disputes. It also emphasizes the importance of finding technological solutions – like metadata standards or proactive initiatives – to balance copyright holders’ rights with AI innovation. The article concludes by acknowledging the irreversible emergence of AI in our lives and the legal system’s need to adapt, offering equitable solutions to copyright holders while fostering technological advancement.