Abstract Generative AI systems require vast amounts of training data, much of it scraped from the internet without creators’ consent. Critics often characterize this practice as “theft,” but such claims require showing that AI training violates creators’ property rights. However, this position risks implying that human learning and works of inspiration would count as theft, too. This paper develops two arguments for moral restrictions that apply uniquely to AI training, not to human learning. The first contends that even if creators possess extensive moral property rights over the use of their content, in the human case, these property rights are typically overridden by a countervailing right to freedom of thought. Since AI systems lack such a right, creators’ property rights remain intact, requiring AI companies to obtain permission to use their work. The second argument contends that even without such broad property rights, creators have the moral right to restrict which copying technologies may be used to capture their work, allowing them to block web crawlers for AI training while permitting ordinary human browsing. The paper concludes by exploring the policy implications of these moral arguments.
James McIntyre (Tue,) studied this question.