Generative AI Learning Community: Generative-AI and Copyright — Who Gets to be a Creator?
Generative-AI is here to stay, impacting everything from the markets for human- and AI-generated works to who gets to be a creator under copyright law. A Midjourney image recently won first prize at the Colorado State Fair, beating out dozens of submissions by human-creators. Getty has launched a “commercially safe” image generator trained on the large corpus of images made by Getty’s artists and photographers. Adobe has built AI generators for Photoshop, promising “accountability, responsibility and transparency” in their announcement for a range of new AI-driven tools. But are AI-generated works protected by copyright? Should they be? And what about the human-creators whose works are used to train AI? Do AI companies violate these copyrights when developing generative-AI tools?
Courts are grappling with these questions across the globe. Lawsuits against AI companies will soon reveal whether exceptions in copyright law, like fair use or text and data mining, permit the use of copyright-protected works by AI systems. Should the human-creators win, the effect may be that AI companies are required to license creative works en masse for AI training. What might be the follow-on impacts for creative industries and the market for creative works? What new barriers to entry might arise for smaller tech companies with limited resources to spend on licensing for new AI-technology development? What will be the impact for consumers, both in terms of competition and diversity in AI technologies, as well as the outputs they produce?
Fortunately, in-copyright creative works are not the only materials available for training. A growing corpus of public domain works unrestricted by copyright is being published as part of an open GLAM movement (Galleries, Libraries, Archives and Museums). Globally, at least 1,616 cultural institutions have published more than 95 million digital surrogates of public domain works for unfettered reuse. These range from digital surrogates of books, sculptures, paintings, drawings, scientific illustrations, music, letters, and other creative works, to transcriptions, descriptions, collections data, provenance data, metadata. Given copyright’s long term of protection, the demographic of this public domain reveals many of the biases that advocates for critical approaches to AI have fought diligently to expose in large image and language models. How might biases long-embedded in our cultural collections be operationalised for AI, leading to similar harms and outcomes that we know to be present?
This session will begin with a talk to problematize generative-AI through the lens of copyright. We’ll explore power imbalances in the data, technology and capital required to develop and shape AI, the (un)fairness of using human-creators’ works to train AI systems, and ask who shares (or not) in the profits—and how? We’ll also ask: could the underlying justifications for copyright shape new policies that result in more equitable outcomes? We’ll think about policy making in practice using copyright’s (alleged) goals to decide: what to do with generative-AI?
- Nuria Rodrïguez-Ortega. (2018). “Canon, Value and Cultural Heritage: New Processes of Assigning Value in the Postdigital Realm,” Multimodal Technologies and Interaction 2(2)
- Authors Alliance. (2018). “Amanda Levandowski on Fair Use for Fairer AI”
- Pamela Samuelson. (2023). “Generative AI meets copyright,” Science 981(6654), (or podcast option)
- Creative Commons. (2023). “Making AI Work for Creators and the Commons”
Andrea Wallace is an Associate Professor of Law & Technology at the University of Exeter in the UK. Her work examines the role of law and technology in transforming how we understand, consume and disseminate art and cultural heritage in the digital realm.
This talk is planned to take place in person (Lynn Chu Classroom LL002) and online. A link to join via Zoom will be sent to registrants shortly before the event.