Category: Artificial Intelligence (AI)

On the Use of AI Likeness in Motion Pictures

In my book, published in 2020, I speculated about a biopic made with an AI-generated likeness of Carrie Fisher; and this week, Variety reports that a motion picture about Edith Piaf is now in development that will use AI-generated likenesses of the famed torch singer. So, now that the hypothetical is reality, what are the considerations beyond the obvious loss of job opportunities for performers to play these roles?

One possibility, of course, is that audiences won’t respond well to this approach to biopic. Part of the appeal of the genre is watching a skilled actor play an important, beloved, or even detested figure. I can only speculate about what would be lost if, for instance, the motion picture Ray did not feature Jamie Foxx becoming Mr. Charles, but my instinct is that the overall experience for the viewer would be diluted. This is not to say that new forms of expression will not emerge from experimenting with AI likenesses (e.g., CGI already features in reenactment sequences for documentary projects), but these are aesthetic and market considerations. What about the ethical or legal questions?

Mostly Not a Copyright Matter

To begin, the right to dramatize a real person’s story in a motion picture is not necessarily a copyright matter. Copyright applies if, for instance, biographical material has already been written (or otherwise recorded) because copyright only attaches to a work fixed in a tangible medium of expression. So, if a film is based substantially on a single biographical book about Edith Piaf, then the film is arguably a “derivative work” of the book, and the book’s author reserves the right to allow production of the film. That said, facts are not protected by copyright law (i.e., many authors can write biographies about the same figure), and a biopic can be produced based on an original screenplay encompassing the writers’ own research into the subject.

Rights of Publicity

I mention all that to distinguish copyright rights from what are often referred to as “life rights,” a broad, industry term that is often viewed as adjacent to state law rights of publicity (ROP). Importantly, though, ROP entails use of a likeness, which does not cover use of biographical information for the purpose of portraying a real person as a character. While it is often wise (and courteous) to obtain permission to use part or all of someone’s story in a motion picture, individuals do not have absolute control over such portrayals, especially if they are public figures and the depictions are based on an available record. If the law held otherwise, the Trump family could, for instance, exert control over all dramatizations of The Donald, and any future biopics would basically be remakes of The Greatest Story Ever Told.

Of course, there are certain limits. Although the bar for defamation or libel can be high, a motion picture portrayal of a real person can trigger a valid claim of this nature, and it is interesting to ask whether these considerations may be different when the producer uses an AI-generated likeness to perform a potentially damaging scene. Biopics often interpolate the narrative by producing scenes or dialogue that are plausible but not necessarily factual. In this context, then, might the additional step toward “reality” with the use of an AI-generated likeness of the subject alter the burden on the producer to adhere more faithfully to the record?

ROP laws vary state-by-state, and are statutory in half the states, but nowhere do they apply to an expressive use of a likeness in contrast to a commercial advertising use. You might have read that Scarlett Johansson is suing app developer Lisa AI for use of her generated likeness in a social media ad posted on X. Variety describes the facts thus:

The ad, reviewed by Variety, begins with an old clip of Johansson behind the scenes of Marvel’s “Black Widow.” Johansson says, “What’s up guys? It’s Scarlett and I want you to come with me…” before a graphic covers her mouth and the screen transitions into AI-generated photos that resemble the actor. A fake voice imitating Johansson then continues speaking, promoting the AI app. “It’s not limited to avatars only. You can also create images with texts and even your AI videos. I think you shouldn’t miss it,” says a voice that sounds like Johansson.

Based on that description, this sounds like a textbook violation of Johansson’s ROP, using her likeness for the purpose of advertising, and there could also be a violation of Disney’s copyright for use of the Black Widow clip. This is not the first story of its kind to emerge concurrent with recent advances in AI, and I believe that companies using celebrity likenesses in this way should face damage awards and injunctions sufficient to achieve deterrence. If Johansson et al. do not seek punitive outcomes in these cases, then the violation and litigation can become promotional opportunities for the infringing companies—a strategy right out of the Silicon Valley playbook:  infringe now, apologize(ish) later, and reap the publicity benefits of the violation.

Likeness as Individual vs. Likeness as Performer

Because Johansson is a real person and an actress, it’s important to distinguish between her AI-generated likeness as herself and her AI-generated likeness as a character in a motion picture. As mentioned, ROP is neither federal law nor (as it stands) applicable to the use of Johansson’s likeness 30 years from now in a biopic about her. Regarding her likeness as a performer, the SAG-AFTRA agreement with the producers appears to strike a balance for now—namely that filmmakers retain some latitude for the use of generative AI replicas, but they cannot willfully “cast” performers in motion pictures by means of AI replication without permission or compensation. We shall see how things play out for lesser-known and background performers going forward, but for the time being, we won’t see Johansson’s AI avatar playing any roles without her permission.

And that brings us back to projects like the Piaf biopic. While there is no legal barrier, at least in U.S. law, to prohibit the use of AI likenesses of real people for expressive purposes, a film like the Piaf project implies a challenge for the future negotiating power of performers, if audiences demonstrate that they enjoy motion pictures without flesh-and-blood actors. This is, of course, the anxious question on the mind of every performing and creative artist since the explosion of AI over the last couple of years. Will the machines replace us all? Will the motion picture industry, for instance, be reduced to a few hundred computer programmers, ten executives, and a couple of social media gurus?

I tend to doubt that such bleak outcomes are likely. Even without new legal frameworks, I expect some of the frenzy over certain creative uses of AI will settle down as producers discover ways to enhance, rather than overwrite, the fundamental magic that makes an expressive work like a motion picture engaging for viewers. That may be wishful thinking, of course. I suppose it’s possible that the future will comprise a world of AI performers with artificial biographies, which will then be dramatized by subsequent generations of AI performers. At that point, though, I imagine the “audience” will also be AIs that “woke up” and wiped us all out. So, it won’t be our problem.

Photo by: Artyme83

AIs Don’t Learn Jack Shit About Art

While people may continue to debate whether robots dream of electric sheep, let us please stop entertaining the notion that AIs “learn from artistic works the same way human artists learn” to make art. In a recent article solidly arguing that Big Tech is going to win again for exploiting creators to develop AI, Peter Csathy concludes:

For those of you who push back and argue that humans “train” on pre-existing copyrighted works all the time when they create works inspired by (or even “in the style of”) of others, let’s be clear. They typically aren’t plagiarizing or making actual copies. But generative AI is when it “scrapes” each and every word.

Csathy is right, of course, but even his counterargument still accepts the premise of the analogy. And that’s part of the problem. Because the analogy is dumb and should be rejected as dumb, or at least useless in the broader discussion about machine learning and generative AI. The comparison of AI “training” to human artistic “training” fosters a legal, moral, and cultural equivalency that should be dismissed with prejudice, if only because whatever we call the product of generative AI, it ain’t art.

A child finds a shell on the beach she thinks is pretty. She takes the shell home, cleans it off, and places it on a nightstand or other surface to decorate her room. The shell is fun to look at, and its texture, shape, and color inspire the child to hold it in her hand, study it for long periods of time, and perhaps even make new discoveries about it. The shell shares many qualities with art, but it is not art for the simple fact that no human made the object. Likewise, autonomously, AI-generated works are just pretty sea shells on the beach.

The essential anthropic contribution to artistic expression is not merely a doctrinal principle of copyright law (i.e., one cannot own rights in the “works” of nature), but it is axiomatic to the nature of art as both practice and experience. Whether good or bad, high or low, decorative or provocative, commercial or non-commercial, art, by definition, is made by humans. In fact, it is the only enterprise I can think of—other than religion—that entails an instinct or acceptance that something ineffable and profound is inherent.

Art is talismanic much like an autograph, rare book, or historic artifact. The value of an original Van Gogh is not merely underwritten by its uniqueness but by a metaphysical—perhaps even spiritual—sense that the canvas, paint, and expression are all imbued with eidolons of the artist and his place in the human continuum. The instinct to perceive meaning in objects or to form personal relationships with works of expression may be ineffable, but the phenomenon cannot be denied any more than the element of faith can rationally be stripped from religious ritual. With a little practice, I could correctly perform a religious rite, but because I’m an atheist, it would be a meaningless act. An observer might not know, but I would, and so (according to the faithful) would God. Likewise, “art” without the undefinable ingredient (call it what you will) is as empty as a prayer without faith.

Whether readers agree with any of this, perhaps it is enough to simply understand that artists do not merely “learn” to make art by studying the mechanics of prior art. Yes, that is often part of the artist’s education but not necessarily the most important part. And many artists are autodidacts without any kind of formal training. But whatever training, methods, or media may be cited to describe the journey toward art-making, what the artist fundamentally does is synthesize experience into expressive works that both comment upon and alter human experience. And since AI’s can’t have human experience, they really can’t learn shit about art.

Image source by: ipopba

Before Generative AI, Big Tech Taught Artists to Abdicate Copyright Rights

One of the more challenging aspects of copyright advocacy is the fact that many artists and creators are conflicted about enforcing their own rights, and from observation, the disconnect is ideological. For the last 30 years, copyright skepticism has been woven into political narratives rooted in criticism of corporations and the excesses of capitalism—popular themes among the political left, which encompasses most artists. Now that generative AI developers are turning creative works into “pink slime,” and artists are suddenly more interested in their rights, it might help to recognize that the industry deploying AI is the same one that taught creators to advocate against copyright in the first place.

The year 2011 was an extraordinary time to jump into the fray. It was immediately apparent that allegations of “copyright maximalism” were deeply intertwined with a sincere and animated belief that the internet would foster a new and potent form of direct democracy to confront a litany of injustices. Copyright enforcement was characterized as a barrier to that promise, and so, the Stop SOPA campaign (to kill anti-piracy legislation) became part of a larger, frenetic collage that included OWS protests, European pirate parties, Anonymous, Wikileaks, etc., all feeding an atmosphere of revolution that corresponded with headlines and memes claiming that “Hollywood” wanted to use copyright to break the internet and stifle speech.

But Big Tech’s promise to democratize everything was a Trojan Horse from which the AI bots have now emerged to ransack the village. Not only did promoters of the “free flow of information” elide the fact that their platforms were as likely to produce the January 6th insurrection as the “Pussy Hat” March, but the allegation that copyright was a barrier to information flow had nothing to do with liberating our speech and everything to do with limiting their liability.

Every time members of the creative community echoed the anti-copyright messages pumped out by Fight for the Future, the EFF, Public Knowledge, or the platforms themselves, what was really being advocated was a lack of accountability for online service providers. I never fully understood how one of the most exploitative industries in history managed to turn anti-corporatist sentiment to its advantage, but I assumed it was the gestalt of the internet. The illusion that social platforms belong to the people was a charade that enabled Google, Facebook, et al. to camouflage their interests as our rights.

That theme has aged about as well as the tobacco industry’s efforts to sell freedom to get smokers to ignore cancer, but it’s been almost two years since Big Tech’s “Big Tobacco moment,” and little has changed. Neither in Congress nor the courts have online service providers been held accountable for much of anything—and that’s with laws on the books. When we consider that, for almost three decades, the major platforms have acted in bad faith with their end of the DMCA bargain, and the courts have interpreted Section 230 as an unlimited liability shield, it is hard to feel hopeful about a legal framework for accountability for harms resulting from AI.

In fact, certain AI tools (e.g., LLMs) may imply a wider “neutral” buffer between potentially harmed parties and potentially liable parties. “Knowledge” and “intent” are key factors in establishing liability, and we have watched Big Tech play shell game with the concept of what they can “know” or “intentionally” control about activity on their platforms. AI tools could take these shenanigans to the next level, enabling new forms of harm with an even weaker nexus linking the machines to the people who design and operate them.

In the copyright world, platform operators have consistently circumvented their obligations under the DMCA with shrugging statements like We can’t police the internet, alluding to staggering volume while conjuring an association with authoritarianism. Now, the circumstances are different. It is a near certainty that every creative work made has been, or will be, ingested into one or more AI training models, and unless the courts find this to be an act of mass piracy and order disgorgement of the datasets, creators may have to accept that their work is being turned into pink slime.

While it is encouraging to see artists take a more active interest in copyright rights as a response to AI, it is also a bittersweet transition in light of all that has happened so far. Whatever comes next, I hope the creative community will recognize that copyright rights are the closest thing to labor rights the independent artist has. And these rights should not be weakened or abandoned for the sake of more billionaires making false promises about democracy and free speech.