ABSTRACT
This paper challenges the view among some copyright scholars that the use of unlicensed copyrighted material to train AI systems is a form of non-expressive fair use. Scholars and technologists who argue for broad non-expressive fair use in AI training data are reflect an outdated vision of Internet exceptionalism that fails to account for developing ethical and AI policy norms. In contrast, a licensing regime for crawling and scraping copyrighted works to use in AI training would foster the growth of training data markets, improve data quality, and serve the values of AI fairness, transparency, accountability, and privacy. In addition to these traditional copyright arguments, many advocates suggest that training an AI is like educating a human being. In response to this intuition, this paper explore various views in machine ethics on the status of AIs and argue from a eudemonistic perspective that AI technology should serve human interests. This human-centered approach, this paper argues, belies the claim that AI proprietors should have free, unfettered access to copyrighted works.
Opderbeck, David W, Copyright in AI Training Data: A Human-Centered Approach (December 29, 2023), Oklahoma Law Review, volume 76, 2024 (Forthcoming); Seton Hall Law School Legal Studies Research Forthcoming.
Leave a Reply