Category Archives: Uncategorized

Upsetting Conventional Wisdom of Copyright Scholarship in the Age of AI

Mar 28, 2024 Christopher J. Sprigman6 comments

In our modern communication environment, conventional wisdom very swiftly captures and narrows our channels of thought. This is due in no small part to the unceasing production of commentary, which means that every perspective on any important new issue is made available and explored (to use the digital age’s most lamentable neologism) in “real time.”

That is true already of the copyright debates around AI. In particular, it has already become conventional wisdom, and the starting point for discussion, that the use of datasets containing copyrighted works for purposes of training AI models involves reproduction of unauthorized copies of those works, and so is a prima facie infringement. The lawfulness of AI training, it is then said, can be established only by determining whether that activity constitutes fair use.

Oren Bracha’s new article, The Work of Copyright in the Age of Machine Reproduction,¹ focused on copyright implications of AI, seeks to upset this conventional view. The process of using datasets of in-copyright works for AI training purposes, Bracha argues, does not implicate copyright’s reproduction right at all, at least if the concept of a “copy” is properly understood and informed by copyright’s essential commitments. The reason it does not involve the production of a copy, flows, for Bracha’s purposes, from copyright’s fundamental rule against propertizing ideas and other unprotectable elements in copyrighted works.

Bracha explains why many copyright sophisticates have so readily concluded that the mass copying of copyrighted works to train AI is prima facie infringement. His shorthand characterization of this argument is “a copy is a copy is a copy.” It rests on an assumption that the unauthorized copying of an entire copyrighted work must support at least a prima facie case of copyright infringement, regardless of the purpose for which the copy was made. In this view, infringement lies based on the fact that a copy was made, simpliciter. The “why” becomes relevant only later when making a fair use analysis. Courts routinely treat fair use as a defense, the burden on which is with the defendant, even if the Copyright Act explicitly directs otherwise.²

Bracha debunks this “a copy is a copy is a copy” argument as a “physicalism” fallacy and shows its inconsistency with “the basic purpose of the [copyright] field.” He argues that this purpose “is grounded in the production and use dynamics of expression and expression alone.” And hence, “[p]hysical facts—whether the making of physical objects, their display, or transfer of possession in them—are never relevant in themselves.” They are relevant “only to the extent they involve in some way the enjoyment by relevant actors of the use value of expression.”

This is why Bracha maintains that the conventional wisdom about physical copying is wrong. A “copy” made for the purpose of training AI is not the kind of copy that can infringe copyright’s reproduction right: “Making a new physical copy when the expression embodied in it will be experienced by no one is not any more relevant for copyright than using an existing copy as a doorstop.”

Moreover, the copyright irrelevance of training copies, Bracha insists, does not depend on circumstances: “Mere physical reproduction, delinked from enjoyment of the expressive value of a work and completely incidental to accessing the meta-knowledge of acquiring skill, is categorically placed outside of copyright’s domain.”

One virtue of Bracha’s argument is that it does not rest solely on theory. It attempts to ground itself instead on core copyright doctrine. Bracha argues, plausibly, that “the meta-knowledge of acquiring skill” (i.e., the thing that we want AI to “learn” when we copy copyrighted works for the purpose of AI training) belongs in the category of ideas, not expression.

Put differently, studying a work to understand its “style,” or the conventions it deploys, is, at bottom, nothing more than the discernment and consumption of ideas about that work and others like it. And if access to the work’s expression is necessary to access the work’s meta-knowledge “ideas”, then the expression is subject to merger; that is, all expression in the copyrighted work would then merge into the meta-knowledge idea. Copyright’s merger doctrine concerns copyrightability, and is not just a defense to infringement. As a consequence, the plaintiff’s infringement claim, Bracha argues, simply fails to launch. For that reason, the fate of AI training need not, Bracha argues, rest on the “slender shoulders” of the fair use doctrine.

There is much more in the article that, in similar style, attempts to upend the conventional wisdom about copyright and AI. I’ll leave those parts to this jot’s readers. I for one admired Bracha’s bold article even though I suspect his arguments are unlikely to convince courts mired in copyright formalism.

In the end, the conventional wisdom is likely to be durable. That doesn’t mean it’s right.

The (White) Person of Skill in the Art?

Mar 12, 2024 Sarah FackrellAdd a Comment

Persons having ordinary skill in the art (PHOSITAs) play a number of important roles in patent law. The validity of a patent may, for instance, depend on whether PHOSITAs would likely have developed the claimed invention. A PHOSITA is a hypothetical person, like the “reasonable person” in torts. But are they also (effectively) White? If so, what does that mean for our patent system? These aren’t entirely new questions, as past articles (including these) have at least raised the issue. Goodman and Patterson enrich this literature with two case studies involving patent applications for useful inventions related to Black hair care. They show how cultural gaps between Black inventors and (mostly White) patent attorneys and patent examiners can create significant hurdles to effective patent protection.

This is mainly due to how patent protection is obtained. Unlike a copyright, which arises automatically when the conditions of the Copyright Act are satisfied, a patent must be granted by the U.S. Patent & Trademark Office (USPTO). To obtain a patent, an inventor or their assignee must submit an application which is then substantively examined by a person with technical skill in the relevant area. The process is complex enough that applicants generally seek the services of a patent attorney or agent to help them draft and prosecute the application.

Most patent attorneys and agents are White, as are most patent examiners. Thus, in practice, the PHOSITA on whose judgment patentability depends effectively ends up being a White person of skill. This might not matter (or matter too much) when drafting and evaluating certain types of patent claims. But, as Goodman and Patterson show here, it can create critical problems for certain inventions, such as those relating to Black hair care.

One case study examines two Black inventors’ experience in obtaining a patent for a new and innovative hair sculpting tool. As Goodman and Patterson explain:

The hair sculpting device is colloquially referred to as a hair sponge, and has a handle portion attached to a pliable portion (sponge) with bores spaced and sized to shape and style hair. Their hair sponge created a twisted hair style within minutes when the hand-twisting method took hours. This could be used for children and adults who did not have the time, money, or patience for a hand-twisted style, and was versatile enough to work for those with short hair and long hair.

Based on this description, the innovation and value of the invention seem clear. But after reviewing the history of the examination, Goodman and Patterson suggest neither the attorney who drafted the application nor the examiner who evaluated it understood the innovation or appreciated its value. The authors show how these understandings (or misunderstandings) could have materially affected the prosecution process, resulting in a patent that may have been narrower (and thus less valuable) than what the inventors may have been entitled to.

In a second case study, Goodman and Patterson examine a reverse situation: A patent prosecuted by a pro se inventor. This inventor understood the culture and the value of the invention, but lacked formal education in the law and experience in patent administrative processes to obtain the optimal patent. Even though the inventor obtained a patent, it was likely narrower than it might have been had the patent examiner had a broader cultural perspective on this type of innovation.

So what to do? Goodman and Patterson “propose amending elements of the patent system to reduce the cultural capital gaps between practitioners, examiners, and clients. This includes amending intake processes, advancing USPTO search tools, and improving access to competent pro bono services and pro se support.” They further suggest the USPTO “explore employing a set of anthropologists or sociologists” to work on these issues. They also suggest various ways that law schools could help prepare culturally competent attorneys.

In particular, Goodman and Patterson suggest that the (mostly White and mostly male) people who are currently allowed to prosecute patent prosecutions—i.e., those with scientific or technical educations—need more cultural education. Another idea is to open the doors to the patent bar to more people who already have the relevant cultural knowledge.

The USPTO has long required that all patent practitioners have scientific or technical backgrounds. But is that really the most relevant type of expertise for all inventions? Goodman and Patterson note that “[t]he cultural capital necessary for [understanding and evaluating] Black hair care inventions” includes “including knowledge of Black hair texture and style, perceived monetary value of the Black hair care industry, and existing product expertise.” The inventions discussed in this article—while certainly innovative and valuable—do not seem so technologically complex that they would require a degree in engineering to prosecute.

There is now precedent for opening up the patent bar to people with specifically relevant knowledge. In November, the USPTO announced that it was creating a system of limited admission to allow people with art & design backgrounds to prosecute design patents.

So what if, in addition to trying to make sure people who currently qualify to be registered patent practitioners (i.e., those with certain technical educations or experiences) get more cultural education, we allowed more people with the relevant cultural education and experiences to become registered patent practitioners? Perhaps inventors like the ones featured in this article would be better served if the patent bar were opened up to allow people with this type of cultural knowledge to prosecute (and thus develop expertise in prosecuting) patent applications. In this way, Goodman and Patterson’s article also adds to the important work done by others (such as Will Hubbard and Mary Hannon) in challenging the patent-bar eligibility status quo.

Taxing Collusion

Feb 28, 2024 Kristelia García1 comment

Commentators have long raised the alarm about over-consolidation in the entertainment industries and the resulting barriers to entry seen in the downstream market. One line of scholarship identifies copyright law as a lever that policymakers might use to promote new entry into the copyright industries. Another looks to antitrust law to alternately break up, or prevent, over-consolidation. Some scholars have suggested utilizing both copyright and antitrust. Still others, myself among them, have expressed skepticism in the ability of either copyright or antitrust to effectively remedy the problem, and instead hope to borrow regulatory ideas from other contexts.

In an engaging new article, Rachel Landy takes a fresh look at the challenge of over-consolidation and power in the music industry, and proposes two novel approaches­­—tax and mandated transparency—for restoring competition and encouraging market entry in the streaming music space.

Using the book publishing industry as a foil, Landy describes the music industry as an uber-concentrated market controlled by a mere three record labels—the perfect setting for “braiding” formal contractual arrangements and informal industry norms in a way that breeds trust and information-sharing among the participants, to the detriment of the platforms and consumers. Braiding, she explains, has allowed the music industry to retain control over the distribution of its content through both formal mechanisms like most-favored nations (MFN) clauses, and informal ones like tacit collusion and parallel pricing, all while evading any violation of antitrust law. Landy argues that braiding affords the labels an undeserved surplus while thwarting innovation in the space, all to the detriment of consumers.

After summarizing some of the literature on restoring competition in the music industry through copyright, antitrust, or both, Landy suggests two alternate approaches aimed at decreasing the labels’ incentives to coordinate, or increasing their incentive to defect, or both. First, she makes a call for mandatory disclosure of contract terms as a means of discouraging so-called tacit collusion. While this would inarguably increase transparency, it would also run counter to a core (and requisite) principle of private ordering; namely, that it is private.

Her other (and, in my opinion, most promising) proposal is to utilize tax law—a mechanism frequently used to influence behavior, but that has not (to my knowledge) yet been applied in the entertainment context—to discourage parallel behavior amongst the record labels. She essentially argues for a tax on any revenues owing to the parallel practices engaged in by the labels; i.e., revenues that they would not have earned if recorded music was a competitive marketplace. Specifically, Landy proposes imposing a higher tax rate on any revenues earned by virtue of an MFN on price. She also proposes a new tax be imposed on any breakage—i.e., the amount that the labels earn from a minimum guarantee unrelated to the underlying per-stream value of the licensed works. If the minimum guarantees come down and the labels are forced to compete on price, she reasons, the market will open up to new entrants. While Landy’s proposal is made in the music context, it might be applicable to other content industries as well.

The idea of a sort of “collusion tax” is intriguing. One thing I like about Landy’s tax proposal is that it is arguably consistent with taxes that we see in other contexts­—e.g., luxury taxes and excess-profit taxes—such that it doesn’t necessarily come off as punitive. We’d expect the industry to balk at any measure or regulation that may be perceived as penalizing the labels for employing sound business measures—like MFNs—to increase revenue, but because tax law treats revenue from different sources differently as a matter of course, the proposal belies at least the most obvious prospective critique.

Of course, the numbers required for the proposed tax calculations will not be easy to come by, and are ripe for manipulation. A tax on the labels’ MFN and breakage overage in addition to the tax they already pay on earned revenue may also be viewed as double-dipping on the part of the state. Given the complexity of the recorded music and downstream markets, higher tax liability alone is unlikely to definitively unravel consolidation in the music industry, but it’s a step in the direction of more creative thinking when it comes to checking industry consolidation, and I think that’s a good thing.

Streaming is not the first new-fangled technology to disrupt traditional business models in the music industry, and it won’t be the last. Indeed, the multitude of class action infringement suits against generative AI developers suggests we may be just getting started. Landy’s article is an insightful contribution to a growing body of scholarship that looks at the way law interacts with the copyright industries, and is recommended reading for scholars who aspire to strike the ever-elusive balance between incentivization and access.

Auratic IP in a Culture of Copies

Feb 8, 2024 Felicia CaponigriAdd a Comment

People often want to experience aura, especially from the products we purchase. Professors Bechtold and Sprigman explore this phenomenon in their interdisciplinary and comparative article Intellectual Property and the Manufacture of Aura. It explains how and why companies manufacture aura to make their products more attractive in the market to consumers.

They offer an “eclectic array of examples” (P. 357) from Italian Bergamasco salami to Birkenstock sandals to NFTs, and more. Why do you choose a Bergamasco salami for your next aperitivo instead of any random salami? That’s likely because you are not just wanting to consume a product made of excellent ingredients from Bergamo, Italy. You are also seeking to consume a specific narrative and participate in a cultural experience implicitly communicated by the narrative aura infused into the Bergamasco salami.

One key takeaway from Bechtold’s and Sprigman’s work is that manufacturers in certain places seem to be infusing their products with narratives hoping to make a difference in the market. These “auratic narratives” work in a complex of law, marketing, social norms, communities, and businesses despite, or in part because of, the presence of multiple copies of the product and the lack of one original.

Divided into three kinds – authorial, place, and originality – the Bechtold and Sprigman auratic narrative is a descendant of Walter Benjamin’s celebrated and much discussed conception of aura. Identifying the essence of Benjamin’s aura as a “ ‘halo of preciousness’ that marks [an art object] as authentic” giving it “aesthetic authority” (P. 294), the article also builds on previous scholarship on authenticity, especially by Amy Adler and Laura Heymann.

The Bechtold-Sprigman article also exists squarely within a tradition exploring formal and informal rules of IP exemplified by the previous scholarship of Sprigman and other co-authors. But while Benjamin’s aura was tied to one, often tangible, original work of art in his celebrated 1936 article, the central move of Bechtold’s and Sprigman’s work is to see that auras now inform so many of the mass-produced copies that we buy. These copies benefit from successful storytelling, which may be backed by an intellectual property (IP) right. Key stars of auratic narratives are an author, a place, or provenance.

A second takeaway is that at times auratic narratives are buttressed by and even subsumed into IP law, including copyright and geographical indicators (GIs) and, at times, trademark laws. The authors deem this an “auratic use of IP” which deepens or invents “our understanding of the social significance of an article of consumption” (P. 299). In this sense, the authors implicitly point to a use of IP which has been perhaps more explicitly present in the control of now public domain works.

Consider, for instance, the confusion created by Italian museums’ continued enforcement of the right to an image of a cultural property under Italian cultural heritage law. While a type of mutant copyright, it is, however, intended to be something closer to the control of the auratic authorial, place and even originality narrative which Bechtold and Sprigman expertly outline.

In this sense the article opens an important door to the role of storytelling and culture in our consumption and preferences for products. There is also a role for “invented heritage” (P. 348) in the market and under IP legal frameworks. The way GIs and copyright law apply to salami and Giacometti sculptures (Pp. 320-21) give some preference to place and authorial narratives that feed into GIs’ recognition of a community and place and copyright’s recognition of an author. Should we then be thinking more critically about whether those communities and places and authors matter to the producer and marketer more than to consumers and the public at large? That is, should consumers more critically engage with the auratic narratives that are marketed to them? Should IP law engage in a more stringent or skeptical review of attempts to emphasize the author of a multiple so a producer benefits from IP protection (as Birkenstock has done to secure copyright in its Madrid sandal in Germany) (P. 317)?

The authors outline how often, in Europe, a “museum test” has at times proved decisive when evaluating whether a design is eligible for a copyright. What role do museums and other cultural institutions have in supporting auratic narratives? How should museums and cultural institutions’ role affect a new justification of IP as providing incentives “for the production and consumption of auratic experience” (P. 300)? In a world where there is a Vitra Factory Store and Design Museum (P. 313), the question seems timely and one for which this insightful article lays a foundation.

Bechtold and Sprigman give us at least one methodological tool to use when answering these questions: public goods theory. Providing IP rights based on the production of auratic narratives may be appropriate if such a right would prevent the undersupply of auratic narratives, allowing producers of multiples to prevent free-riders and allowing auratic narratives to remain uncongested.

Bechtold’s and Sprigman’s article opens a door to deeper considerations of heritage and culture in IP law scholarship that consider products designed for reproducibility in a variety of industries, from fashion to food to furniture, and art.

The application of cultural heritage law to prevent mechanical reproductions for commercial purposes of the very works of art to which Benjamin referred is meant in part to support the official supply of an auratic narrative around copies of the work of art, whether it is Michelangelo’s David or Botticelli’s Birth of Venus. This control is deployed to prevent unauthorized free-riders from referencing the auratic narrative around these same works.

The application of cultural heritage law to certain copies is meant to protect an auratic narrative. This provides a problematic example of how using the law to privilege certain auratic narratives over others can undermine cultural dialogue. This may affect the very acceptance and celebration of aura needed to support consumers’ desire to consume a specific narrative and participate in a cultural experience, as they would with a Bergamasco salami.

As a call to “debate…the optimal design of auratic IP” (P. 358), Bechtold and Sprigman’s article opens an important door for academic and practical dialogue about the power and recognition we should extend to the producers of auratic narratives that increasingly make their designs, copies, and other products cultural artifacts.

The article helps to create a bridge between two areas of law that are perceived to be different and, especially in market nations, unrelated: cultural heritage law and IP law.

Bechtold and Sprigman’s article offers an analytical framework that might overcome the silos of each law by deploying a new term, auratic narrative. Auratic narrative can encompass both heritage values and values that are more commonly associated with intellectual property, like distinctiveness and originality. Thanks to Professors Bechtold and Sprigman comparative dialogue between cultural heritage law and intellectual property law, especially for junior scholars, may be more possible, and fruitful.

Free Riding as a Pro-Defendant Impulse

Jan 25, 2024 Graeme DinwoodieAdd a Comment

American trademark scholars have almost uniformly decried the role of free riding in calibrating the scope of trademark rights. They have argued that the language of “reaping where you have not sown”—to use the famous, but doctrinally discredited, agricultural metaphor of INS v. Associated Press in 1918—distorts trademark law and expands protection beyond that necessary to ensure consumers receive accurate information about the source of goods in the marketplace. Yet, there is a (relatively well-grounded) suspicion that, despite this almost universal scholarly condemnation, the impulse to protect mark owners against free riding remains a resilient force when courts decide trademark cases. If this suspicion is indeed sound, arguments for less expansive trademark rights that rest on the rejection of free riding as a relevant variable are destined to fail.

In Trademark Free Riders, Mike Grynberg brilliantly and persuasively outlines (with real panache) an alternative approach by which to advocate for less robust trademark rights. He assumes arguendo the reality of judicial attention to (and distaste for) free riding and articulates an approach to trademark law that uses the free riding of trademark owners to justify a less expansive scope of protection. That is to say, what happens when trademark owners reap where they have not sown? If courts are in fact sensitive to free-riding as a relevant consideration in allocating rights, might they not confine the scope of right when evidence of trademark owner free-riding is presented to them?

For example, should trademark law (or courts in trademark cases) not take account of the fact that mark owners may opportunistically exploit meanings developed in popular culture generally or, more narrowly, by their customers? The use of HOGS to describe Harley-Davison motorbikes was initially opposed by the company. Its meaning is attributable to efforts of its customers. Should the tardy and begrudging acceptance by the company of a name developed by its customers figure into the ownership and scope of any trademark rights?

Grynberg thus operates in a reality that too many scholars are not even willing to countenance. This approach is very much consistent with Grynberg’s broader approach to trademark law, having written in an important article 15 years ago about the way that consumer interests can be reconfigured (and differently advanced) to secure a more balanced trademark regime than one-sided consumer paternalism might suggest.

This piece thus highlights a fascinating conundrum for scholars and litigants in trademark cases which parallels that faced by advocates before the U.S. Supreme Court. If, as Justice Elena Kagan suggested in her 2015 Scalia Lecture, that “we are all textualists now,” would not a wise advocate or scholar seeking to persuade the court advance textualist arguments? Many litigants, including in intellectual property cases before the Supreme Court, have taken the lesson offered by Justice Kagan on board (in my view, to the detriment of intellectual property law), even as Justice Kagan has voiced doubts about her 2015 pronouncement). So should trademark advocates and scholars grasp the nettle of free riding and make arguments for why trademark owners should have narrower (or no) rights because of their free riding (even if they believe, normatively, that free-riding should have no role to play in trademark law)?

Grynberg does carefully note some reasons why free riding might actually have a resonance for judges in trademark cases. For purposes of this Article, Grynberg is non-committal about the wisdom of that choice. There is surely a case for its relevance at the margins. The legislative history of the Lanham Act explicitly talks of protecting against the exploitative conduct of “pirates and cheats”. It takes some methodological insularity to read that reference and conclude that the statute simply targets “those who increase search costs”, as advocates of law and economics would have us believe. That’s a bigger and more contentious debate, however, and one with which Grynberg does not need to engage to make his point. Grynberg simply takes the relevance of free riding to judges as a given. As a result, he offers an approach to trademark law grounded in the reality of judges who decide trademark cases. And those scholars who wish to influence the development of trademark law would be well-advised to heed his counsel.

Without rehashing all of the numerous examples that are teased out with meticulous rigor, Grynberg develops a taxonomy that would helpfully inform the arguments of litigants and scholars in trademark cases seeking to confine the scope of trademark rights. For example, Grynberg notes that trademark owners free ride on culture, using and potentially appropriating memes and other forms of communal communication (something very much at issue in current debates over the use of the “failure to function” ground for rejecting applications for trademark registration). Likewise, he recognizes that what Mark Janis and I have called “surrogate uses” (where the public adopts a source-identifying term not used, and perhaps opposed, by the trademark owner, such as the HOGS example noted above) represent annexation by the trademark owner of understanding developed without an iota of effort or intent by the mark owner.

Moreover, Grynberg suggests that certain trademark owners may even obtain free rider benefits from competitor activity. Where protection is sought for a term that straddles “source” and “category”, or a shape that is a both a configuration mark and an attractive product design, trademark law seeks to ensure that any trademark rights protect the former and not the latter. But Grynberg points out that when we protect a term or shape that possesses such duality (as we do unless the competitive effects are substantial enough to trigger genericness or functionality doctrine), the mark owner may benefit from rival producers seeking to promote a competing product in the same category or of a similar design. That is, the efforts of competitors may consolidate public understanding of the first producer’s mark. Grynberg offers a vocabulary that the courts appear to recognize instinctually – “they didn’t earn it” – which might persuade courts to discount any public understanding not closely tying the product to the first producer (as opposed to the product type).

Whether one agrees with every example (and there are good debates to be had about some), this is a conversation-changing tour de force of an article. Professor Grynberg opts pragmatically to operate in the world we observe – perhaps the most basic commitment of a social scientist – and offers an approach to trademark law that litigants and scholars seeking (thus far without much success in recent years) to limit the scope of trademark rights, would do well to bear in mind.

To Pay or Not to Pay (for Training Generative AI), That is the Question

Dec 18, 2023 Christophe GeigerAdd a Comment

As the first empirical evidence is published on the consequences of Generative AI systems on labor markets³, broad anxiety is felt from creator communities on the effects of this technology on their income streams. Consequently, the question of how to deal with Generative AI from a copyright law perspective is gaining a lot of attention globally. Several lawsuits have been filed in the US by creators against AI operators and the first attempts to legislate that matter have already been introduced at the national level. The EU is currently finalizing an ambitious regulation package called the “AI Act” with important implications for its copyright regime, in particular the implementation of transparency obligations concerning copyright-protected works used to train the AI algorithms. In this context, Martin Senftleben’s new article Generative AI and Author Remuneration is particularly timely and proposes a very inspiring reflection on what could be the way forward regarding copyright reforms in this field.

One of Senftleben’s main concerns is to find a workable approach not to disincentivize AI innovation while at the same time creating new revenue streams for “flesh and blood authors” to secure remunerations that will improve their working and living conditions. Indeed, the starting point of the author is that:

the increasing sophistication of AI systems will inevitably disrupt the market for human literary and artistic works. Generative AI systems provide literary and artistic outputs much faster and cheaper. It is therefore foreseeable that human authors will be exposed to substitution effects. They may lose income as they are replaced by machines in sectors ranging from journalism and writing to music and visual arts.

In the article, much attention is devoted to analyzing whether the training of the machine learning algorithm with copyright-protected works can be permitted under copyright law. This is a complex question, and it is fair to say that no jurisdiction in the world has a straightforward answer to it, as no copyright law has yet passed having generative AI technology in mind. As machine learning is based on Text and Data mining (TDM), from a copyright point of view, the question often concentrates on whether exceptions and limitations that allow TDM can cover these uses as well.

While the Big Tech AI industry claims this situation falls under the US fair use exception, the content industry considers on the contrary that these uses are covered by the exclusive right and should be licensed. In Europe, a recently introduced TDM exception offers the possibility (under certain circumstances) for right holders to “opt-out” of the exception for text and data mining and to retain full control of their work (article 4 of the Directive for Copyright in the Digital Single Market). Some large collective management organizations have already announced that they will opt out their entire repertoire from TDM activities for machine learning, which will significantly reduce the available training material for AI systems.

As Martin Senftleben rightly underlines, applying this “opt-out”-mechanism to generative AI is not a satisfying solution because it would inhibit the development of this technology and thus make the European Union unattractive for AI developers. For the same reason, he rejects the idea of submitting these uses to the exclusive right: “The need to obtain individual authorizations and manage remuneration payments for AI training constitutes an additional cost factor in the form of transaction costs and licensing fees. If the costs involved are too high, it will negatively impact the ability of the EU’s AI sector to compete on the world market”.

The core of Senftleben’s proposal lies in the argument that copyright law, in order to compensate human authors for the reduction in their market share and income through Generative AI, should introduce an AI levy system in the form of a statutory remuneration and ensure the payment of equitable remuneration to creators. In his model, the statutory remuneration would however not be related to the TDM use of protected works for AI machine learning purposes, but it is “the literary and artistic output of generative AI systems” that serves “as a reference point for a legal obligation to pay remuneration”.

According to the author, focusing on an “output-oriented AI levy system can be applied uniformly to all providers of generative AI systems in the EU. In contrast to a remuneration obligation focusing on the input dimension and AI training activities, the output-oriented levy approach avoids the risk of disadvantages for EU high-tech industries. All providers of generative AI systems are equally exposed to the levy payment obligation the moment they offer their products and services in the EU.” This lump-sum remuneration would have to be paid by AI developers when their systems produce AI-generated output that have the potential to serve as substitute for works made by human authors. To counter legal/doctrinal concerns and to give theoretical support to the proposal, Senftleben refers to the theory of the “domain public payant” (“paying public domain”), according to which the exploitation of the public domain should at least partly serve the living generation of authors.

Admittedly, the proposal put forward by Martin Senftleben is very European in its spirit as its income redistribution rationale might not be an easy fit for all copyright traditions, in particular the US one. From a European point of view, however, the proposal is certainly compatible with a tradition of remunerated exceptions, as there is an established practice and case law about the distribution rules in favor of creators of this kind of remuneration via collective management organizations. In this respect, maybe submitting TDM for machine learning in the context of generative AI to a remunerated exception could be a workable alternative?⁴ Indeed, at the policy level, it will be more difficult to achieve consensus on a proposal based on a paid public domain, as advocates of a robust public domain might be favorable to ameliorate the remuneration situation of creators, but less sympathetic to the idea of a domaine public payant.

In any case, there is no doubt that this important article provides further arguments to consider the position of creators in forthcoming copyright reforms in the field of AI and more generally helps to reflect on how to finance creative ecosystems in a fast-moving technological environment.