In my last post, I discussed some of the allegations that “machine learning” (ML) with the use of copyrighted works constitutes mass infringement. Citing the class action lawsuits Andersen and Tremblay, I predicted that if the courts do not find that ML unavoidably violates the reproduction right (§106(1)), copyright law may not offer much relief to the creators of the works used for AI development. As of last week, it remains to be seen whether we’ll get to that question after Judge Orrick of the Northern District of California stated that he is tentatively prepared to dismiss the suit with leave to amend the complaint. The judge did indicate that a claim of direct infringement could survive, but we’ll have to see what comes of an amended complaint.
As mentioned in the last post, if the court does not find a valid claim of copyright infringement, the other allegations will likely fail as a result. Nevertheless, though the state allegations may be moot in the class cases filed thus far, I had intended in this post to look at whether any non-copyright remedies present much hope for creators. For instance, the Andersen complaint alleges violations of statutory and common law rights of publicity and violations of statutory unfair practice prohibitions in the State of California.
Right of Publicity and Works “in the style of…”
One of the most concerning aspects of generative AI is that it allows a user to prompt the system to make a work “in the style of [named artist].” Karla Ortiz, in her testimony to the Senate Judiciary Committee on July 12, stated, “[Artist] Greg Rutkowski, had his name used as a prompt between Midjourney, Stability AI and the porn generator Unstable Diffusion, about 400,000 times as of December 2022. (And these are on the lower side of estimates).” This is a deeply personal assault on the identity, work, and potential livelihood of an artist who has spent years mastering his craft and developing that distinctive style which can now be mimicked by a computer. But as a matter of doctrine, copyright does not protect style. So, can state laws like the right of publicity (ROP) offer any relief?
Only half the states in the U.S. have statutory rights of publicity, though most states recognize a common law ROP, which may prove more expansive in a litigation. The California statute, considered one of the strongest, prohibits the use, without consent, of a person’s likeness, name, voice, or signature for commercial purposes. In particular, advertising a product, service, or viewpoint so that a representation of the individual implies that person’s endorsement, is a paradigmatic violation of ROP and may infringe the speech right of the individual. But does prompting a generative AI to create an image “in the style of [artist]” implicate the artist’s ROP? Maybe and sometimes.
Because a user can prompt the production of a visual work “in the style of Greg Rutkowski,” one obvious implication is that there will be hundreds or thousands of “Rutkowskis” in the world which the artist did not create. If any of those AI-generated works are substantially similar to an existing work he did create, then he may have claims of copyright infringement—and potentially too many to contemplate addressing. But what about the works that do not look substantially similar to anything in the artist’s portfolio but to the observer, do look like “new Rutkowskis”?
In the fine-art trade, the duty to validate provenance and authenticity (i.e., not commit forgery) should offer some protection for those artists who sell their work in galleries etc. But in the commercial market, if a new vodka brand wants images in the style of, say, Molly Crabapple for its ad campaign but doesn’t want to hire Molly for the job, they could use generative AI to make something in her style. No question this implies that artists will lose gigs, but whether ROP offers a remedy to Crabapple herself in this hypothetical is questionable.
To begin, artwork in the style of the artist is not a “likeness.” Thus, Crabapple would have to prove that observers seeing the ads would perceive the images as her work and that the images, therefore, result in an unlicensed endorsement in violation of ROP. That can be a high bar to reach, let alone repeat in what could amount to multiple complaints by just one artist—and potentially in multiple states! Although a violation of ROP can stand alone without an underlying violation of some other law, it is a fact-intensive, case-by-case consideration and, therefore, hard to imagine how it can support allegations of harm to a plaintiff class as alleged in Andersen.
Moreover, some states only recognize celebrity ROP and not average citizen ROP. So, in the case of the visual artist, what is the threshold where he or she is famous enough to be considered a celebrity? Crabapple, Rutkowski, et al. are very well known in the art world, but they’re not movie-star well known to the general public. So, what constitutes “celebrity” in this context? We don’t know. The specific problems caused by generative AI are brand new.
New Federal ROP Law?
In testimony before the SJC along with Karla Ortiz, General Counsel for Universal Music Group Jeffrey Harleston broached the subject of adopting a federal right of publicity, and the idea was at least entertained by some of the senators. A federal ROP could theoretically address some of the new harms to artists caused by generative AI. Not only would a federal statute provide a uniform, national framework, but the new law would be written with an understanding of AI and its potential harms as a foundation of legislative intent. Further, it was raised in committee that ROP should apply to everybody and not just celebrities.
The Motion Picture Association (MPA) filed comments in response to this discussion, and these were focused largely on the fact that, historically, ROP has applied to commercial/promotional uses, but not to expressive ones. The MPA is right to point to some tricky considerations that would need to shape a new federal ROP in order to strike a balance between disenfranchising creators or performers through AI-generated replicas while allowing use of the technology to create expressions that are protected by the First Amendment.
In fact, in my book, I allude to a hypothetical future biopic about Carrie Fisher that (with the family’s permission, of course) might dramatize scenes using AI replicas. Whether this use of the technology would be an engaging choice in lieu of casting an actress play young Carrie is a question of aesthetics and culture, but not a question that can or should be addressed as a matter of law. Suffice to say, the contours of a prospective new federal ROP are complex enough to be subject of future posts.
Unlike an ROP complaint, unfair competition does not stand alone as an allegation. In general, these laws bar businesses from gaining unfair advantage by engaging in some form of prohibited conduct. In the Andersen et al. class action suits, the underlying conduct is alleged to be copyright infringement, which allegedly makes the AI developers unfair competitors with the plaintiff class of artists. This state allegation would seem to have merit if the court finds the developers liable for violation of §106(1) of the Copyright Act, and unlike ROP, I can see it surviving as a complaint for a whole class—i.e., as unfair competition against all artists. That would be encouraging. But if, for instance, the court finds that not all the named plaintiffs have standing (i.e., do not have registered works in suit), it’s hard to say what this does to the unfair competition complaint as argued.
What About Trademark?
It is tempting to wonder whether certain creators can find protection in trademark law. In addition to the cost of registering and maintaining a trademark, only certain artists would be able to make effective use of this form of intellectual property. In this instance, trademark only protects use of the artist’s name in commerce, and since it is already illegal to trade in forgeries, registering one’s name as a trademark may be redundant and little protection against the use of AI to produce “in the style of” works.
Relatedly, the Federal Trade Commission (FTC) may have a role to play to protect consumers against fraud stemming from the uses of generative AI. As to the FTC stepping in, presumably they could respond to or seek prophylactically to protect consumers from forgery at scale. Not unlike the rampant proliferation of counterfeit products sold via eCommerce sites, generative AI certainly presents the opportunity for some party to start generating mass forgeries of popular artists and selling those in the millions. As such, measures to restrict the use of artists’ names in generative AI may belong in the FTC’s wheelhouse.
In both this post and the last, my intent is not to advocate on behalf of the AI developers. Far from it. Instead, I am trying to kick the tires of existing law to ask whether the law is sufficient to the task of protecting authors of creative works. Because, overall, I’m not sure it is, though it is also essential to note that every type of work has different implications (i.e., voice actors’ rights are more likely to sound in ROP than visual artists’ rights).
One thing is certain. Generative AI is not comparable to the printing press, camera, phonograph, or any more recent changes to production and distribution enabled by digital technology. Debate about AI must be sequestered from discussions about technologies of the past because few, if any, of those revolutions are instructive to the moment. There is no doubt that AI implies new regulation in medicine, finance, IT, security, and just about everywhere else it will invade; and there is no reason why Congress cannot adopt the same posture in order to protect America’s creative culture and economy.
Image source by: idaakerblom