Break It 'Til You Make It: An Exploration of the Ramifications of Copyright Liability Under a Pre-training Paradigm of AI Development

Abstract

This paper considers the potential impacts of a pre-training regime on the application of copyright law for AI systems. Proposed evaluations of the use of copyrighted works for AI training have assumed a tight integration between model training and model deployment: the model's application plays a central role in determining if a training procedure's use of copyrighted data infringes on the author's rights. In practice, however, large, modern AI systems are increasingly built and deployed under a pre-training paradigm: large models may be trained for a multitude of applications and then subsequently specialized toward specific ones. Thus, I consider copyright's indirect liability doctrine to consider the effect of copyright on the current market structures involved in the development and deployment of AI systems. The main contribution of this paper lies in its analysis of what indirect copyright liability litigation for technologies in the past have to say for how AI companies may manage or attempt to limit their copyright liability in practice. Based on this analysis, I conclude with a discussion of strategies to close these loopholes and of the role that copyright law has to play within the AI policy landscape.