New study explores role of generative AI in using copyrighted material

Generative artificial intelligence (AI) facilitates the compilation of huge volumes of data, which often include copyrighted materials. While debates about the legality of the process abound, in a recent report issued by the US Copyright Office, researchers examine to what degree allowing or restricting this practice serves the economic objectives of copyright. In short, does AI shift the balance between incentives and access, and what policies could recalibrate that balance?

A chapter, written by researchers at Carnegie Mellon University, appears in a book on AI and copyright policy edited by the chief economist at the U.S. Copyright Office. The report features discussions by members of an ad hoc committee of economic scholars convened to address economic issues at the intersection of AI and copyright policy.

“One of the goals of copyright is to facilitate cultural and scientific innovation, which requires balancing the economic rewards that can be captured by producers of creative works with their ability to access existing works as part of the creative process,” explains Michael D. Smith, Professor of Information Technology and Public Policy at Carnegie Mellon’s Heinz College, who contributed to the report.

“That cumulative creative process is the foundation of innovation, and some consider generative AI as similar because it ingests existing works and produces something ostensibly new,” Smith continues. “In that sense, the algorithms are engaging in the sort of innovation process that copyright policy aims to encourage.”

However, two questions are key, say the researchers: What social benefits come from developers having access to training materials? And what are the implications for the incentives of human creators to produce works? In the report, they consider the second question by examining the impact of using generative AI to compile data on commercial incentives to create and on intrinsic incentives to create, and by suggesting licensing as a potential solution.

“There are few available policy instruments to combat incentives for holders of copyrights to further limit public access to their works in response to ingestion,” suggests Rahul Telang, Professor of Information Systems and Management at Carnegie Mellon’s Heinz College and at its Tepper School of Business, who contributed to the report.

“Indeed, the only potentially viable solution may be a licensing requirement for ingestion, although it can come with challenges and limitations, including issues related to transparency and enforcement.”

The report includes a discussion of the issues associated with requiring copyright holders to opt out of having their copyrighted data used to train models, which the European Union mandates. This would change the nature of current copyright protections by shifting the burden for action from the user of the copyrighted material to the copyright owners, which represents a meaningful burden on rightsholders, the authors say.