Copyright and the Dynamics of Innovation in Artificial Intelligence

Abstract

Copyright can determine access to data, which may, in turn, affect innovation in Artificial Intelligence (AI). In this paper, we conduct an empirical analysis to investigate the relationship between copyright and the dynamics of AI innovation. Some countries provide exceptions in copyright law that are relevant for AI, particularly concerning text and data mining and the doctrine of fair use. The study takes a global perspective, comparing different countries based on the breadth of exceptions in copyright law. We find that countries with broader copyright exceptions tend to exhibit higher levels of AI innovation, as measured in a larger number of AI research publications, more active participation in open-source AI projects, increased AI patent filings, and a higher rate of AI venture formation. These findings suggest that statutory provisions enabling data access can be pivotal in fostering AI innovation. However, while broad copyright exceptions may support current AI innovation by allowing the use of existing works, they might also inadvertently discourage the creation of new works, leading to negative long-term consequences for the availability and quality of training data. We discuss how mandatory licensing may provide a setting that aligns the dynamic incentives of rightsholders and AI developers.