The idea of "fair use" played a central role in the outcome of this case. Let's find out what it might mean for the future of gen AI.
Prince. Andy Warhol. Generative AI.
You may not immediately think that these three powerhouses of creativity overlap very much. However, they were linked by the U.S. Supreme Court’s recent decision in the Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith case, which may have implications for a booming AI industry.
Before we connect the dots, we need context. In 1981, professional photographer Lynn Goldsmith took a photo of Prince. Three years later, Vanity Fair licensed the image from Goldsmith, who owns the copyright. Unbeknownst to Goldsmith, Vanity Fair then commissioned Warhol to create a single silkscreen illustration of Prince for publication. However, Warhol created 15 additional images with some aesthetic variations, all without Goldsmith’s knowledge, permission or attribution.
Following Prince’s death in 2016, Vanity Fair's publisher Condé Nast licensed one of the 15 unauthorized Warhol works (titled “Orange Prince”) from the Andy Warhol Foundation for the Visual Arts (which owns the copyrights to Warhol’s works) for the cover of a tribute magazine to the late musician. When Goldsmith found out, she sued the Foundation for copyright infringement of her original 1981 image.
The dispute centered around whether Warhol's reimagining of the image — and its use on a Condé Nast magazine cover — constitutes fair use. The U.S. Copyright Office defines fair use as “a legal doctrine that promotes freedom of expression by permitting the unlicensed use of copyright-protected works in certain circumstances.” The case has ricocheted through the court system, with an initial ruling favoring the Foundation that was eventually overturned on appeal, then affirmed by the U.S. Supreme Court.
Now, does this have anything to do with generative AI?
To find out, I spoke to Sam Eichner, a seasoned intellectual property attorney at Pillsbury Winthrop Shaw Pittman LLP who has litigated copyright cases in federal court. He said that the exact answer remains to be seen “because generative AI is so new and a lot of threshold questions haven't even been answered yet.” So right now, the answer is “maybe.”
According to Eichner, in the near term, the Warhol fair use decision may “impact the litigation against AI system developers that is already pending nationwide,” but only if “those courts find that AI system training using unlicensed content actually constitutes copyright infringement.”
Over a longer term, the Warhol decision may discourage unlicensed training sets because, as training set licensing grows more prevalent, it may be harder for platforms to convince courts that unlicensed training sets constitute fair use; particularly that use of unlicensed training sets qualify as use with a new “purpose” or “character” that’s fundamentally different from the original uses of those works.
Read on to hear more of Eichner’s insights on this case and how it could impact the generative AI industry. The expert also shed light on how infringement and fair use could impact generative AI platforms users and what they (we) should think about.
The legal doctrine at issue in the Warhol case is fair use. It’s a defense to copyright infringement. If I go into court as a copyright owner to stop someone from using my work, or to recover damages, I need to prove infringement. If there's no copyright infringement to begin with, fair use doesn't matter; it would never come up. But if I can prove infringement, the defendant can then say, “Hold on. Yes, I used your work without your permission, but it was fair use.” What does that mean? It means that normal copyright protections must yield to a higher purpose (i.e., the First Amendment, which protects free speech), but only in the limited circumstances set forth by Congress.
In fact, there are four fair use factors that Congress set forth in Title 17. The Warhol decision was all about — and only about — the first of these factors: the purpose and character of the use. Why was this work used, how was it used, and how different is that use from the original author’s use? Since the 1990s, courts have articulated this important first factor as whether the challenged use was sufficiently “transformative” so as to give the original work a fundamentally different meaning or message.
This case addressed whether or not the first factor favored the Warhol Foundation. A majority of the Court said it did not. The majority basically said, “For purposes of this case, we don't much care about Warhol's creativity, or how high value the work may or may not be, or how creative it was, or how transformative it was. What we care about is the challenged use of that work, i.e., the placement of Orange Prince on the cover of a magazine.”
The Court held that this use was the same kind of use that photographers routinely make of their photographs, so the purpose and character of that use was not different from the original.
In determining whether the use made of a work in any particular case is a fair use the factors to be considered shall include—
The case arguably changed the fair use test and how courts apply it. That may affect generative AI platforms that use content without the authorization of content owners. Generative AI platform developers need billions of photos, songs and other content to train AI models to a point where they can function. The idea of getting billions of licenses from authors might seem impractical and fair use has the potential to allow developers to avoid this licensing burden; if it applies, unlicensed use is lawful, and there is no need to license billions of works.
A similar issue arose in the Google Books case. Google didn’t license the rights to every book on Google Books, and the Second Circuit Court of Appeals found that Google was protected by fair use in part because Google Books did something new with those works; it added value to society by giving those works a new purpose. Generative AI is arguably similar in that respect; it is a revolutionary technology, and many would say it does something different and new that creates value for society. Before Warhol, one might think that a similar analysis could apply to generative AI, and that fair use could protect generative AI platforms trained on unlicensed datasets. While that may still be true, after Warhol, the first factor analysis has arguably changed.
After Warhol, if the purpose and character of the original use is the same (or similar to) AI training set use, the first factor doesn't favor fair use. Artists are already licensing their content to generative AI platforms. Adobe Firefly advertises itself as a platform with 100% licensed and authorized content. Shutterstock recently inked a licensing deal with OpenAI. So there’s clearly a licensing market there. Under Warhol, courts may reject AI developer fair use defenses because training set use — the challenged use — is arguably no different than the training set licensing we are already seeing in the market.
Over a longer term, the case may affect generative AI platform decision making around whether to license training sets and may even encourage those developers to license training sets to avoid legal uncertainty.
There's an obvious distinction between how platforms use content in training sets and how users of AI platforms use content. Right now, Warhol seems more likely to impact the AI platforms, the companies that develop them and perhaps the content owners whose work is fed into those platforms. How Warhol impacts generative AI users, if at all, is unclear and will depend on the specific use that a generative AI user is making.
Again, fair use only matters if there is infringement, i.e., unauthorized copying to create a substantially similar work. If you can prompt an AI to write something in the style of Bukowski, that is unlikely to raise fair use issues because copyright doesn't protect the style of an author's work. However, if you upload copyright-protected content to a generative AI platform to produce artwork, now you are implicating the content owner’s exclusive rights. To the extent your conduct constitutes infringement, fair use (and Warhol) becomes relevant.
While copyright law has limitations, and won’t protect ideas or an artist’s style, there are other proprietary rights that AI users should be aware of, like trademarks and right of publicity. If you're targeting someone else’s style to impersonate them, or in a manner that is likely to mislead people, you could be running afoul of trademark laws or unfair competition laws. Right of publicity can also come into play when using a person's style, voice or image. And false advertising can become relevant to the extent generative AI outputs are used in advertising or marketing in a manner that is likely to mislead consumers about products and services.
While the Warhol decision has nothing to do with those other areas of law, and is solely about fair use, those other areas of law are relevant to those that use generative AI.
If you're an artist waiting to see whether you have a claim, perhaps this decision is encouraging. Warhol arguably narrows the fair use defense that AI developers can raise as it arguably requires courts to look at the precise nature of the challenged use, not the societal value that generative AI arguably provides. So the decision may create some incentive to bring those claims.
The decision may also impact the pending copyright litigation against generative AI platform developers. For example, there is a case filed by three artists against Stability AI, Midjourney and DeviantArt. In that case, Midjourney has already said in court filings that it has an “obvious” fair use defense, so Warhol may play a role there.
There are also several other copyright infringement cases pending against generative AI platform developers. Warhol could tip the scales in favor of some of those plaintiffs, but only if there is infringement to begin with. Platforms function differently, they’re trained in different ways and they utilize different AI technologies, so the answer may be different in each case. But in those cases where unauthorized copying is found, fair use may well come into play.
Technically, fair use may apply whether or not the challenged use is commercial. But the commerciality of the use weighs against fair use. Of course, as a practical matter, if somebody posts a Star Wars meme on their Facebook or Instagram and it gets a few likes, Disney may not care. But whether a rights holder will object is separate from the legal question of whether something is actually fair use or not.
In Warhol, the Supreme Court emphasized the commerciality of the challenged use, saying it “looms larger” because the purpose and character was not different from the original. But the flip side is that where the purpose and character of the challenged use is different, the fact that a use is commercial matters a lot less. So, whether a use is commercial is ultimately not the most important consideration.
The Supreme Court’s decision in Oracle’s case against Google over Oracle’s Java API is a good example of this. Google used portions of that API for the Android platform, which is obviously commercially successful and has many users. Despite the commerciality of the use of portions of the Java API for the Android platform, it was still fair use.
Commerciality also comes into play in the fourth fair use factor, which asks “Does this impact the author's licensing market?” Warhol did not address this factor, but it still matters in the overall fair use analysis.
Shopping Jasper for a large business? Update your input to 100+ employees to meet with our team. Please use business email to meet with our team.