Regulating IP exclusion/inclusion on a global scale: the example of copyright vs. AI training

Abstract

This article builds upon the literature on inclusion/inclusivity in IP law by applying these concepts to the example of the scraping and mining of copyright-protected content for the purpose of training an artificial intelligence (AI) system or model. Which mode of operation dominates in this technological area: exclusion, inclusion or even inclusivity? The features of AI training appear to call for universal and sustainable "inclusivity" instead of a mere voluntary "inclusion" of AI provider bots by copyright holders. As the overview on the copyright status of AI training activities in different jurisdictions and emerging laws on AI safety (such as the EU AI Act) demonstrates, the global regulatory landscape is, however, much too fragmented and dynamic to immediately jump to an inclusive global AI regime. For the time being, legally secure global AI training requires the voluntary cooperation between AI providers and copyright holders, and innovative technolegal reasoning is needed on how to effectuate this inclusion.

Keywords: artificial intelligence, copyright, text and data mining, inclusivity, inclusion, AI Act