The Columbia Journal of Law & the Arts

Covers and Front Matter

Columbia Journal of Law & the Arts — 2025-05-30

Responsible AI Starts with Licensing

Roy S. Kaufman — 2025-05-30

Responsible AI starts with licensing. AI outcomes are strengthened by reliance on responsibly sourced, high-quality copyrighted works. Consent and human centricity are hallmarks of human advancement, and also will enable better AI. The default position of our copyright system is that the person who wishes to use copyright protected works must seek out and obtain a license before engaging in conduct that would implicate any rights protected by copyright law to avoid an infringement claim. This will only seem fair to most observers, given the intellectual and economic labor involved in creating original works, and also in light of the natural and economic justice of granting the creator the right to determine how her works will be used.

Of course, the copyright system contains exceptions in certain cases where authorization may not be needed at all (e.g., fair use in the U.S.) or where rights are limited to non-negotiated licenses (e.g., non-voluntary licenses and/or levies). But these exceptions or limitations to rights, in order to serve the public interest of promoting the creation and distribution of original creative works (as well as to comply with international law), must be carefully circumscribed to avoid unfairly prejudicing the legitimate interests of the author. To be acceptable under the Berne 3-Step test, they exist in areas of market failure.

AI training generally requires the express and voluntarily granted consent of the author. Any other approach threatens fundamental values underlying our copyright system. This observation is grounded in a number of core truths: that AI training requires the reproduction of protected works; that copies of such works are not (only) ephemeral or transitory and are stored in a manner that permits their retrieval for the purpose of producing expressive output that derives from the training data; and that such AI output do, and are likely to, directly compete in the marketplace for expressive works with the works on which the AI was trained, as well as to unfairly displace licensing opportunities that would otherwise exist for authors of the original works.

As we approach how to balance competing interests around AI technologies, we are faced with a litany of arguments that copyright is somehow not fit-for-purpose for this AI. These are the same arguments that were raised with respect to the sound recordings, cable, the photocopier, the internet, and every new technology where one party wanted to make money through uncompensated and unconsented to use of another’s creative works. AI is not as different as the advocates for unfettered, uncompensated reuse pretend.

Pro-Copyright, Pro-AI: The Power of Collective Licensing

Anita Huss-Ekerhult — 2025-05-30

The aim of this Article is to showcase the emerging AI licensing solutions pioneered by RROs and to explore how these frameworks address the challenges posed by AI’s reliance on copyrighted works. The discussion begins with a primer on the structure and operation of collective management frameworks, highlighting their effectiveness in managing rights for secondary uses. Following this, the Article examines the necessity of licensing in the AI context, emphasizing why exceptions and limitations under current copyright regimes are insufficient to adequately address the complexities of AI training. This section underscores the limitations of existing legal frameworks and the potential harm to rightsholders if AI systems continue to use copyrighted works without appropriate permissions or compensation. Finally, the Article presents an analysis of the emerging licensing solutions tailored to AI, illustrating how these initiatives by RROs are not only meeting the demands of the AI market but are also paving the way for sustainable and equitable practices at the intersection of copyright and technology.

Market-Based Licensing for Publishers’ Works Is Feasible. Big Tech Agrees.

Matthew Stratton — 2025-05-30

Generative AI (“GAI”) model developers prioritized speed to market over compliance with copyright law with respect to use of copyrighted works for training their models. Now facing over forty lawsuits, they have asserted fair use to evade responsibility, and they claim that licensing all necessary works is impossible.

This Article focuses on professionally created works only, with an emphasis on publishers’ works, and demonstrates that market-based licensing of professionally created works for training GAI models is feasible as measured by the number of licenses and the ability of GAI developers to afford them—both of which are points on which Big Tech agrees. The Article also provides insights on the licensing marketplace for publishers’ works as relevant to training GAI models. Finally, the Article underscores that the public interest is squarely on the side of marketbased licensing because all stakeholders benefit, and it will help ensure that publishers and authors may continue their vital contributions to America’s political, intellectual, and cultural systems.

Licensing of Text for Generative AI: Learnings from Non-AI Licensing Practices

Regan Smith — 2025-05-30

Models (“LLMs”) are moving fast, fueled by developer demand for access to, and the ability to use, increasing amounts of high-quality textual data. With this constant demand for quality literary works come questions around how licensing practices can enable technological developments while preserving the contours of copyright law and sufficiently incentivizing human authorship of books, journalism, and other literary content for human readers and AI uses alike.

While some venture that licensing for generative AI purposes is “impossible,” many companies have negotiated partnerships with media publishers or publishing houses for generative AI uses. Meanwhile, others query whether there is a need to build or adjust licensing systems to better facilitate licensing of textual content, whether through regulatory updates, increased use of collecting societies, or augmenting data management infrastructure.

Before declaring the status quo of marketplace licensing insufficient, it makes sense to take stock of where we have been, where we are, and where we might be going. This short piece hypothesizes that some current bumps in generative AI licensing stem from uncertainty in an emerging market, not inherent difficulties in licensing at scale for professionally published content. Given that generative AI is still in its nascency, content licensing is not close to a one-size-fits-all standard. The time is ripe for marketplace developments, and experimentation in private arrangements between rightsholders and users. The Article also provides a brief primer in copyright principles of licensing regulation and overviews guideposts for collective management of content, based on experiences outside of AI. While voluntary collective licensing can play a valuable role in the AI licensing market, these guideposts may assist authors and other licensees as they consider whether, with whom, and on what terms to affiliate with a licensing intermediary.

Comments on How AI May Affect the Motion Picture Industry

Ron Wheeler — 2025-05-30

Annotated transcript of dicussion with Ron Wheeler on how AI may affect the motion picture industry.

The Stock Photo Industry and Generative AI

Nancy E. Wolff — 2025-05-30

This Paper presents the evolution of and the effect of generative AI on the image licensing industry. The Paper discusses the history of the image licensing industry, then analyzes the value of the contemporary image licensing market, with a focus on the metadata associated with the images. It will provide analysis into the effects of the emergence of generative AI technology on the image licensing industry. The Paper synthesizes ideas presented at the 2024 Symposium of the Kernochan Center for Law, Media and the Arts.

Past and Present Copyright Tribunals for Setting Royalties in the United States

Steve Ruwe — 2025-05-30

I was asked to participate in The Kernochan Center’s Symposium addressing “Past, Present and Future of Copyright Licensing.” I noted that in light of my current role on the United States Copyright Royalty Board, my presentation and discussion participation would focus on the past and the present of statutory Copyright Licensing in the United States. I chose to exclude any personal outlook on the future of Copyright Licensing, leaving that to other participants. The same holds true for this Article, which adheres to the topics addressed in my presentation. Thus, as the Symposium and the public look to potential licensing solutions that may emerge amidst the development of Artificial Intelligence products, my hope is to offer a brief, and high-level, background on how the United States has approached statutory licensing in the copyright realm. In doing so, I often look to the Register of Copyright’s 2015/2016 study, Copyright and the Music Marketplace, and recommend that study as a far more comprehensive portrait of the music licensing landscape at it existed at the time—prior to the enactment of the Music Modernization Act in 2018. Additional Copyright Office publications are available with more comprehensive information regarding the statutory licenses addressed herein.

The Past is Prologue—How Prior Challenges with New Technology May Guide the Music Industry in Dealing with AI

Elliott Peters — 2025-05-30

In a peer-to-peer file-sharing case1 involving tens of millions of users sharing millions of sound recordings, Judge Sidney R. Thomas of the Ninth Circuit Court of Appeals wrote:

The introduction of new technology is always disruptive to old markets, and particularly to those copyright owners whose works are sold through well-established distribution mechanisms. Yet, history has shown that time and market forces often provide equilibrium in balancing interests, whether the new technology be a player piano, a copier, a tape recorder, a video recorder, a personal computer, a karaoke machine, or an MP3 player.

For the music industry, artificial intelligence (“AI”) is a technology full of virtue and promise that has already proven valuable in numerous ways. In the creation of music by human artists, AI has been useful in assisting in the production of songs and sounds, as well as in the automation of related tedious technical tasks, thereby freeing creators to focus more deeply on their artistry. With respect to consumers, AI has been useful in helping to organize, categorize, and index their music, as well as in supporting their discovery of music through song curation, playlist creation, and recommendation. At the same time, AI also poses a number of disruptive threats. By training on unauthorized uses of copyrighted works, AI can create music that has the potential to oversaturate the market, thereby undermining the artistic integrity of music created by human beings and threatening the economic welfare of creators. How might these various threats be minimized so that AI neither inflicts serious harm to the careers of artists and songwriters nor cripples an industry that is based on and supports human creativity? History may offer a guide.

In this Article, I will:

set out the issues surrounding four moments in music industry history in which a new technology (often, a new format) posed challenges to copyright law and/or business norms of the time;
describe how those challenges were overcome and their disruptive effects muted; and
highlight how the lessons of these past challenges may be useful as the industry confronts the challenges posed by AI.

Using Past Legislation as a Template for Future AI Licensing Legislation

Joe Keeley — 2025-05-30

Artificial intelligence (“AI”) has become a major public policy issue in Washington across a range of industries. The copyright community has also been focused on AI policy, most notably over the issues of training AI models on copyrighted works and the copyrightability of generative AI. This Article focuses on AI training.

Copyright issues surrounding training are currently the subject of significant litigation in U.S. district courts, predominantly the Southern District of New York in which the fair use defense has been raised by AI companies. Copyright issues surrounding AI training may very well reach the U.S. Supreme Court. There is certainly enough money at risk on both sides of the issue to make it likely that at least once of the many current cases will eventually be heard by the Court. If the Supreme Court determines that a license is not generally required, the licensing question (and the basis for this Article) ends the day such a decision is announced. If, however, the Supreme Court determines that a license is often required, Congressional action to enable either collective or compulsory license may be needed in certain circumstances even though direct licensing has already occurred and will no doubt continue.

From 2004 to 2007, I served as copyright counsel to then-House Judiciary Charmain James Sensenbrenner of Wisconsin during which time I led weekly Congressional negotiations over two copyright bills. The first bill was an update to the Section 115 compulsory music licensing system. The legislation, titled the Section 115 Reform Act, was designed to modernize a paper-based licensing system for the digital music services era. The second bill, titled the Orphan Works Act, was a copyright industry-wide bill to address the longstanding orphan works licensing problem on a work-by-work basis. Neither bill was signed into law by Congress, although an updated version of the Section 115 Reform Act later became the now-enacted Music Modernization Act of 2018. One could also say that the Music Modernization Act proved that Congressional staff never leave since I was the lead negotiator on that bill as well. Although none of these three bills had anything directly to do with AI, Congress often builds upon what it has previously debated or enacted as a basis for future legislation. Thus, it is possible that the prior music licensing and orphan works bills could provide some basis for AI licensing legislation.

Operational Considerations for Collective Licensing Frameworks in the Music Publishing Industry

Lidia Kim — 2025-05-30

A large part of my work is to negotiate and issue blanket music publishing licenses that cover the availability for use of full catalogs of music on behalf of the songwriters and rightsholders my company represents. It should not come as a surprise if I were to say that licensing negotiations often center around what is the appropriate value for the use of music (i.e., ample consideration for the rights granted under contract), and that the economics of a deal take the spotlight. After all, the goal is to get music professionals paid for their work.

However, equally important to the success of a deal, and therefore a key component of negotiating a blanket license, is operational—how is the use of music managed, and what are the responsibilities of each of the contracting parties to administer the license and pay the underlying rightsholders. Administering an individual synch license is straightforward—the licensee knows what music will be used and how and relays that to licensor, licensor issues the license, collects payment, and administers the royalties. The thought exercise becomes more complicated with digital service providers (DSPs) whose services and platforms host and make available seemingly limitless quantities of music, where the volume of usage is high or the extent of usage unknowable (or both).

Extended Collective Licensing for Use of Copyrighted Works for Machine Learning

Johan Axhamn — 2025-05-30

The fast development of generative artificial intelligence (“AI”) services—such as ChatGPT, Midjourney, Dall-E—have within a short period of time gained immense uptake and popularity. At the same time, such services have given rise to fundamental challenges from a copyright perspective. Court proceedings have been initiated in many jurisdictions on the compatibility of such services with copyright legislation.

Some scholars see the development of AI as a gradual process, to be dealt with, like earlier technologies, through incremental adaptation of the copyright framework. For others, AI represents so fundamental an innovation—a disruptive technology, a game changer, an apocalypse—that it threatens to shake copyright law to its very foundations. The Economist has described the challenges as a “battle royal.”

These technological and legal developments—and related economic consequences—have, in turn, raised political and scholarly interest in the issues at stake. For example, the World Intellectual Property Organization (“WIPO”) has dedicated studies and seminars to the topic, the Association Littéraire et Artistique Internationale (“ALAI”) 2023 Congress in Paris focused on AI and copyright, and several jurisdictions have or are considering specific provisions in copyright law of relevance to this emerging technology. Entire symposia, including this one—the Kernochan Center’s 2024 annual symposium The Past, Present and Future of Copyright Licensing—are dedicated to related copyright issues.

A copyright-related question that has gained much attention is whether the output generated by generative AI services can obtain copyright protection, and if so, who the author is. Another question, which is the focus of this contribution, is whether the use of copyright protected content as part of the “training” of the AI—i.e., machine learning—constitutes copyright-relevant use, i.e., falls within the rights protected by copyright. And if so, whether the so-called extended collective licensing model could be a relevant vehicle (or mechanism) for clearing rights for such use. Related to aspects of extended collective licensing, issues have been raised around whether there are challenges associated with competition law that need to be taken into account.

Against this backdrop, this Article is structured as follows. Section I, deals with machine learning and copyright, i.e., whether and to what extent the use of copyrightprotected content as part of the “training” of the AI (machine learning) constitutes copyright-relevant use. Section II describes and discusses whether the extended collective licensing model could be a relevant mechanism for such use. Section III focuses on some challenges from a competition law perspective, and also relates to some relevant provisions in the EU directive on collective rights management. Section IV sets out some concluding remarks.

Identifying an “Effectively Competitive” Market: The Work of the Copyright Royalty Board

David R. Strickler — 2025-05-30

Because the statutes supplant the actual market, the CRB Judges must establish a “hypothetical market” that satisfies the statutory standard. A critical element in that regard is the testimony of the parties’ economists, which consists of various forms of economicThe United States Copyright Royalty Board (“CRB”) establishes royalty rates for compulsory statutory licenses of sound recordings paid by Webcaster licensees to sound recording companies. These rates are set by the government, rather than the market, because the licensed sound recordings are not simply individual copies of discrete sound recordings in competition with each other, but rather are collections of repertoires offered under a blanket license by the major record companies and one independent consortium—who control the vast majority of recordings.

Accordingly, coexisting with the efficiencies of collective ownership is the market power of these collectives. Absent a regulatory rate, the collectives, “complementary oligopolies,” can exploit the “Must Have” (essential) nature of their blanket licenses by setting non-competitive royalty rates. When rates are not “effectively competitive,” the market is beset by inefficient and exploitive pricing.

It is for this reason that many collective licensors are subject to royalty rate regulation. The fact that unregulated copyright collectives may not achieve an economic optimum establishes the strong theoretical foundation for the regulation of such collectives.

The CRB’s three-judge panel is required by statute to “establish rates . . . that would have been negotiated in the marketplace between a willing buyer and a willing seller.” To accomplish this economic task, the Judges preside over adversarial trial proceedings between licensors and licensees.

Because the statutes supplant the actual market, the CRB Judges must establish a “hypothetical market” that satisfies the statutory standard. A critical element in that regard is the testimony of the parties’ economists, which consists of various forms of economic modeling. The experts who proffer such testimony are typically well-credentialed economists who have been, inter alia, the highest-ranking economists at the Antitrust Division of the U.S. Department of Justice and distinguished professors of microeconomics and industrial organization. Their direct oral and written testimonies, akin to expert reports, are subject to adverse expert rebuttals, examination by skilled counsel, and inquiries from the Judges, one of whom (the author of this Article), by statute, “shall have significant knowledge of economics.”

This Article focuses on a seminal opinion by the CRB Judges, affirmed by the D.C. Circuit: the Web IV Determination holding that the statutory “willing buyer-willing seller marketplace” standard shall reflect the workings of an “effectively competitive” market. In all subsequent CRB royalty rate setting determinations, the Judges have applied this “effective competition” standard to the particular facts of the case.