No one knows how US courts will rule on the growing list of AI copyright lawsuits. It will be a long road. Copyright law often takes decades to adapt to new technologies; for example, more than a decade passed from the launch of the pioneering file-sharing music software Napster until courts shut down the last file-sharing service. Fewer than three years have passed since OpenAI launched ChatGPT.
In the absence of legal clarity, licensing is racing ahead. OpenAI has signed with AP, News Corp, Time, The Atlantic, Financial Times, Axel Springer, Le Monde, and PRISA, as well as scientific and academic publishers Taylor & Francis, Wiley, and Oxford University Press and trade book publisher HarperCollins. Dozens of additional such deals reportedly exist – some confidential.
Licensing will be as crucial as litigation in developing the rules of the road for using copyrighted material in AI training. Many copyright owners consider licensing to AI platforms an urgent necessity. Some are offering simple and reasonable licenses, in the hopes that AI platform companies will find them attractive as an alternative to the “sue now, license later” strategy that the music industry followed in the file-sharing era of the late 1990s. Major record labels prevailed over Napster on appeal in 2001, and then (apart from a few small-scale experiments) did not make any significant licensing deals with third-party digital music services until Apple’s iTunes appeared in 2003.
Some AI licensing deals include provisions that extend beyond content rights, including online traffic referrals and joint product development initiatives. These can also be valuable. As Nick Thompson, CEO of The Atlantic, has explained, the company’s deal with OpenAI “provides a way for us to help shape the future of AI.”
AI developers have strategic reasons to enter into licensing deals that may protect them from future litigation. It is a lesson learned from history. Starting in the 1970s, the photocopier disrupted publishing by making content dramatically easy and cheap to reproduce without permission. The US was preparing a major revision to its copyright law at the time, which resulted in the Copyright Act of 1976. While Congress declined to include provisions covering large-scale photocopying in the Act, it did recommend that publishers “work out means by which permissions for uses beyond fair use can be obtained easily, quickly, and at reasonable fees.”
Publishers – primarily scientific journals — responded by creating a collective licensing organization, the non-profit Copyright Clearance Center (CCC). From its launch in 1978, the CCC organized and offered a voluntary blanket photocopying license. This “one-stop shopping” license to a large catalog of content led many companies to take CCC’s license, even before the legal necessity was tested in the courts.
One company that declined was petroleum giant Texaco. It argued that photocopying was fair use. A group of publishers sued Texaco in 1985 and, after 10 years, prevailed in a federal appeals court. The court cited CCC’s blanket license as a factor as to why Texaco’s photocopying should not be considered fair use. AI companies may find themselves in analogous situations if courts should rule similarly on fair use in AI systems.
Get the Latest
Sign up to receive regular Bandwidth emails and stay informed about CEPA's work.
The CCC operates to this day and added AI rights for internal use to its Annual Copyright License last year. In addition, content aggregators – most of which are startups — are emerging to offer blanket licenses for AI developers. Most focus on specific types of content, such as Created by Humans (books), Protege Media (video), GCX (Global Content Exchange; music), and vAIsual (images). These content aggregators have already organized a trade association: the Dataset Providers Alliance. The large stock image agency Shutterstock, which aggregates image creators’ assets for bulk licensing, has also made licensing deals with AI platforms.
The sheer volume of licensing deals from individual rightsholders as well as aggregators has led to concerns about fragmentation. AI companies may be willing to take licenses if they are reasonable in cost and easy to administer. What they do not want is to have to take licenses separately from every copyright owner or pay royalties according to thousands of different schemes – as services like Spotify and Everand do today for music and e-books respectively. In that scenario, some AI developers may find it preferable to keep obtaining training data without permission.
Many stakeholders believe that the market should be allowed to decide. A precedent exists in the music industry: although digital music services have taken licenses with record labels and recording artists individually, independent artist distributors (TuneCore, CD Baby), independent label licensing entities (Merlin), and various collective management organizations have emerged to streamline the process.
Both US courts and Congress may consider the ease of AI licensing in determining whether to require it. There is ample time. It took 73 years – from the invention of sound-recording technology in 1888 to the Rome Convention in 1961 — for international law to confer copyright protection on sound recordings, and 10 years longer in the US. Even as technology advances at warp speed, the market and the law will both take time to sort out the AI challenge.
Bill Rosenblatt is an adjunct professor in the Music and Performing Arts Professions Department at NYU. He is the co-author of several books, white papers, and peer-reviewed articles on digital media and copyright. Bill has served as an expert witness in dozens of litigations related to copyright, digital media, security, and music business issues in the US, Canada, and Europe. He also chairs the annual Copyright and Technology Conferences in New York. Bill holds a B.S.E. in Electrical Engineering and Computer Science from Princeton University and an M.S. in Computer Science from the University of Massachusetts.
Bandwidth is CEPA’s online journal dedicated to advancing transatlantic cooperation on tech policy. All opinions are those of the author and do not necessarily represent the position or views of the institutions they represent or the Center for European Policy Analysis.
How the US and Europe Can Come Together
Read More
Read More From Bandwidth
CEPA’s online journal dedicated to advancing transatlantic cooperation on tech policy.
Read More