Impacts of AI on US Copyright Law

This chapter provides an overview of the clash between copyright and generative AI, exploring the tensions and possible solutions. Although this is a global issue and implicates multiple legal jurisdictions and traditions, this chapter will primarily focus on US copyright law. We begin with an overview of copyright itself and, importantly, illustrate that copyright has not remained static since its inception. We move on to introduce fair use, which ostensibly exists to act as a safety valve that prevents copyright monopolies from stifling future creativity, and is the main provision that supports training AI models on copyrighted material. The discussion section follows to interrogate the two key copyright issues raised by generative AI. First, we look at the authorship requirement of copyright: whether generative AI outputs are themselves copyrightable, and how the current stance affects the market value of these works. Second, we discuss the tensions in training generative AI models on copyrighted works. In assessing the claims for training as infringement and training as fair use, we will illustrate the culturally deterministic role played by copyright. As Goldstein (Copyright’s highway: The law and Lore of copyright from Gutenberg to the celestial jukebox. Hill and Wang, 1994, p. 35) reminds us, “Other more familiar bodies of law—contracts, torts, and crimes—have an even longer history, but none occupies this special place in ordering a nation’s culture, high and low.” We finish this section with a discussion of possible solutions, some of which are beginning to emerge at the time of writing. The chapter concludes by affirming the crucial questions before judges and legislators as they decide whether copyright is a facilitator of or hurdle for generative AI. These are decisions that will carry significant ramifications for concepts of authorship and ownership, as well as creativity and culture more broadly.