Abstract
The ongoing litigation between generative AI companies and copyright holders has intensified in the past year. While much of the debate revolves around fair use in the legal domain, key technical questions about copyright risk quantification, mitigation, and compensation remain unresolved. In this talk, I will focus on two specific open technical questions in GenAI copyright:
-
What factors may trigger copyright risks in GenAI models and how to quantify the risk?
-
If copyright infringement is established for a GenAI model, what plausible technical solutions could model deployers implement to mitigate risk, and how effective are these measures?
Finally, I will touch on other pressing open technical questions in the field, including potential pathways for fair compensation of copyright owners in cases of confirmed infringement.