I have a new draft paper with Katherine Lee and A. Feder Cooper on copyright and generative AI: Talkin’ ‘Bout AI Generation: Copyright and the Generative AI Supply Chain. The essay maps out the different legal issues that bear on whether datasets, models, services, and generations infringe on the copyrights in training data. (Spoiler alert: there are a lot of issues.) Here is the abstract:
This essay is an attempt to work systematically through the copyright infringement analysis of the generative AI supply chain. Our goal is not to provide a definitive answer as to whether and when training or using a generative AI is infringing conduct. Rather, we aim to map the surprisingly large number of live copyright issues that generative AI raises, and to identify the key decision points at which the analysis forks in interesting ways.
I know that I say that every paper was especially fun to work on. This one too. Katherine and Cooper are computer scientists who work on machine learning, I am the law-talking guy, and every single paragraph in the paper reflects something that one or more of us learned while writing it.