The market dilution argument is deceptively simple. It proposes that even when AI systems do not directly copy or harm any individual copyrighted work, the sheer volume of machine-generated output could crowd human authors out of the marketplace, weakening the economic incentives that copyright law was designed to protect. If a publisher can generate endless novels, illustrations, or articles at negligible cost, the theory goes, human creators may find their works displaced and their livelihoods eroded. This reasoning has begun to appear in courtrooms as a way to extend copyright liability beyond traditional infringement, reaching the training process itself and the outputs of foundation models trained on vast collections of copyrighted material.

Samuelson and her colleagues question whether this theory holds up under scrutiny. In their view, the argument oversimplifies the relationship between human and machine creativity, treating AI as a uniform threat to all creative markets rather than a technology whose effects vary enormously depending on context. The authors contend that any serious assessment of market dilution must account for how AI is actually being used, the extent of human contribution to the works in question, and the specific creators and markets affected. A graphic designer using AI to accelerate routine tasks faces a very different situation from a novelist competing against fully automated fiction, and a stock photography market operates under dynamics distinct from those of fine art or scientific publishing.

This insistence on specificity carries a direct challenge for the courts. The two federal rulings so far have found that training AI models on copyrighted material can qualify as fair use, the legal doctrine that permits certain uses of copyrighted works without permission from rights holders. But those decisions have not settled the broader dispute, and the market dilution theory represents one of the most consequential arguments still in play. If courts accept the idea that generative AI inherently dilutes markets for human-authored works, the consequences could extend far beyond the parties in any single case. As Samuelson and her colleagues write, rulings against fair use on dilution grounds may substantially raise liability risks for all developers and users of generative AI in ways that are difficult to predict.

The authors are particularly concerned about the ripple effects on the scientific community. Generative AI holds considerable promise for advancing research, from literature synthesis to hypothesis generation to domain-specific modeling, but that promise depends in part on the legal permissibility of training models on large collections of existing material. Court rulings that restrict the use of copyrighted works for training could create widespread legal uncertainty, affecting scientist users and builders of generative AI systems, including open-source and domain-specific models that underpin much of contemporary computational research. The warning is not abstract: researchers who fine-tune models on scientific corpora, developers who release tools for public use, and institutions that deploy AI in laboratories all stand to inherit liability risks whose contours no one can currently map.

What the authors propose instead is a discipline of evidence. Claims of harm to creative markets, they argue, should be evaluated through rigorous empirical research focused on particular AI systems and particular creative markets, rather than through broad assumptions about the impact of AI on human authorship. This means longitudinal data tracking how creative industries actually evolve as AI tools spread, alongside surveys of creative practices that reveal how human authors are integrating, resisting, or adapting to machine assistance. The relationship between human and AI creativity is not a zero-sum contest in their account; it is a set of interactions that differ across genres, media, price points, and audiences, and only sustained observation can distinguish genuine displacement from speculative fear.

To make that observation possible, Samuelson and her colleagues suggest collaboration between AI developers and researchers. Access to data about how models are trained, what outputs they produce, and how those outputs circulate in commercial markets would allow scholars to test dilution claims against reality rather than intuition. Such partnerships are not without complications, since proprietary interests and privacy concerns complicate data sharing, but the authors frame them as essential if policy is to rest on evidence rather than anecdote. The alternative, they imply, is a legal regime shaped by contested narratives in litigation, where the loudest claims of harm or benefit substitute for careful measurement.

Even if evidence eventually establishes that AI harms particular creative markets, the authors do not believe courts are the right institution to respond. In their view, legislatures are better equipped than judges resolving individual copyright disputes to develop targeted policies such as licensing schemes or public funding for creators. A legislature can weigh competing interests across an entire sector, design compensation mechanisms, and adjust rules as technology evolves, while a court ruling in a single case must apply legal doctrines to facts presented by the parties before it. The distinction matters because copyright doctrine was built around copying and distribution, not around the diffuse economic effects of a new production technology, and stretching it to cover market dilution risks distorting both the doctrine and the technology.

The Policy Forum lands at a hinge point in the relationship between copyright law and artificial intelligence. The fair use question remains unsettled despite two favorable federal rulings for AI developers, the volume of litigation continues to grow, and the market dilution theory offers plaintiffs a route to liability that does not require proving copying at all. Samuelson and her colleagues have staked out a position that resists both extremes: they do not dismiss the possibility that AI could harm creative markets, but they insist that such harm must be demonstrated, measured, and addressed by institutions capable of crafting proportionate responses. For scientists, developers, and creators alike, the message is that the next few years of litigation and legislation will determine whether generative AI develops within a predictable legal framework or under a cloud of unpredictable liability.

What happens next will shape more than courtroom outcomes. If empirical research confirms that specific markets face genuine displacement, targeted legislative remedies could protect creators without freezing AI development. If research instead shows that human and machine creativity coexist and complement each other in most domains, the dilution theory may fade as a legal strategy. Either way, the authors have reframed the debate around a question that can actually be answered: not whether AI is good or bad for creativity in the abstract, but which systems, which markets, and which creators are affected, and by how much. That reframing may prove to be the most consequential contribution of all.

Subject of Research: The legal and economic debate over whether generative AI dilutes markets for human-created works and the role of copyright law, empirical research, and legislation in resolving it.

Article Title: Does generative AI dilute markets for human-created work?

Article References: Does generative AI dilute markets for human-created work?. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: generative AI, market dilution, copyright law, fair use, Pamela Samuelson, Science journal, AI litigation, creative industries, human authorship, AI policy, licensing schemes, scientific research