A recent Supreme Court ruling in the Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith et. al. case will have broad implications for the generative AI platform architectures pushing the boundaries of copyright infringement and fair-use law. 

Feature Image Source: Image processing of  Warhol’s Prince portrait – along with images of Prince and Warhol generated by OpenAI’s DALL-E2 using the prompt:  “3D renderings of Prince and Andy Warhol.”

Background

The images in question

Smithsonian Magazine sets the stage for the case: 

“In the 1980s, Andy Warhol created an illustration of the musician Prince, which drew heavily from an existing image by photographer Lynn Goldsmith.   Now, four decades later, the Supreme Court has ruled that the Pop artist infringed on Goldsmith’s copyright. Experts say that the case has complex implications for the ever-shifting boundaries of copyright infringement and fair-use law.   

Seven out of the nine Supreme Court justices ruled against Warhol’s estate, which argued that the portrait in question should be considered the work of Warhol alone.   “To hold otherwise would potentially authorize a range of commercial copying of photographs, to be used for purposes that are substantially the same as those of the originals,” wrote Justice Sonia Sotomayor in the majority opinion.  The two dissenting justices, Elena Kagan and John Roberts, argue that this ruling could stifle the creativity of artists who want to riff off of copyrighted material.  “It will impede new art and music and literature,” wrote Kagan. “It will thwart the expression of new ideas and the attainment of new knowledge. It will make our world poorer.” 

NBC News adds:  “In a win for photographer Lynn Goldsmith, the court ruled 7-2 that Warhol’s images did not constitute “fair use” under copyright law, a decision that will have considerable impact on various creative industries. The ruling is beneficial to people who own copyrighted content upon which other works are based, and it could have a negative impact upon artists who make new works based on existing material.” (4)

The legal battle revolves around a divisive question: After Warhol altered the image, did it belong only to him, or should Goldsmith retain some degree of control?

The court’s decision relies on how the image was used. In this case, wrote the justices, the commercial nature of the Prince portraits means that the images aren’t covered by fair-use law. (1) 

The Andy Warhol Copyright Case That Could Transform Generative AI

“…the decision could greatly impact how copyright law is applied to what AI tools do with human-made works.”

Wired Magazine complements the Smithsonian’s coverage with the details of pending legal cases grappling with the potential copyright infringement and fair-use issues of artificial intelligence “training” on vast amounts of previously existent images:  “Currently a trio of artists is suing Midjourney, Stable Diffusion maker Stability AI, and DeviantArt, claiming that the tools are scraping artists’ work to train their models without permission…all three companies filed motions to dismiss, claiming that AI-generated images bear little resemblance to the works they’re trained on and that the artists didn’t specify which works were infringed.

The artists are being represented by Matthew Butterick and the Joseph Saveri Law Firm, which also filed a class action against OpenAI, GitHub, and GitHub’s parent company Microsoft for allegedly violating the copyrights of coders whose work was used to train the Copilot programming AI, part of the ‘no-code ecosystem.’ Getty Images filed a suit in January against Stability AI claiming ‘brazen infringement’ of its image licensing catalog.

At the heart of many of these debates about AI’s impact on creative fields are questions of fair use. Namely, whether AI models trained on copyrighted works are covered, at least in the US, by that doctrine. Which is why we’re talking about Warhol.

“There’s a version of this case where it’s so obviously a derivative work,” says Ryan Merkley, managing director at Aspen Digital and chair of the Flickr Foundation. Goldsmith’s photo was provided for a single use but was used multiple times. “Why didn’t Goldsmith get paid for the thing she got paid for the first time?”

The case has confounded observers, attorneys, and artists. It’s difficult to know whether Warhol appreciated Goldsmith’s contribution to the Prince series or how Prince felt about Warhol’s use of his likeness. Ultimately, those questions may never be answered. But what the Court must decide is whether Warhol’s piece is a significant transformation of Goldsmith’s photograph, and thus protected by fair use, or if it’s copyright infringement. Either way, the decision could greatly impact how copyright law is applied to what AI tools do with human-made works.

In the age of generative AI, it has a whole new relevance. 

“Copyright is meant to be an incentive for creation, and AIs don’t need that incentive,” says Merkley. “I think if you let AIs make copyright, it will be the end of copyright, because they will immediately make everything and copyright it.” To illustrate this, Merkley describes a world where AI systems make every potential melody and chord change and then immediately copyright them, effectively barring any future musician from writing a song without fear of being sued. This is why, he adds, “copyright was meant for humans to make.”

Now imagine that same tactic applied to prescription drug formulations or computer chip architecture. And that’s where steering the massive ship that is copyright runs into choppy waters. Copyright is a keystone in global trade agreements: The North American Free Trade Agreement, the Trans-Pacific Partnership, and others rely on a shared recognition of copyright between nations. Granting AI copyright would fundamentally alter trade policy. It could further erode or destabilize international relations.

What Next?

 “The Warhol ruling…has added a lot of complexity back to the issue of fair use, and no one is feeling that more…than AI companies.”

“The US Copyright Office determined recently that art created solely by AI isn’t eligible for copyright protection. Artists can attempt to register works made with assistance from AI, but they must show significant “human authorship.”

The office is also in the midst of an initiative to ‘examine the copyright law and policy issues raised by artificial intelligence (AI) technology.'” (2)

Jonathan Bailey at Plagiarism Today itemized What the Warhol Ruling May Mean for AI:

• The case represents a rare ruling by the Supreme Court on the issue of fair use.
• Though the exact amount and direction of the change is yet to be determined, it is inevitable that this case will shift the way future fair use decisions are reached in lower courts.
• The biggest fair use question in front of the lower courts right now is the issue of generative AI.
• Nearly every major player in the AI space is facing a copyright infringement lawsuit from one or more parties. This includes Stability AI, which is being sued by Getty Images and a pair of class action lawsuits that target GitHub, Midjourney and, once again, Stability AI. 
• Those lawsuits have just begun and, as many have noted, we’re very much in the “wild west” days when it comes to AI and copyright.
• Generative AIs work by ingesting, or processing, large amounts of human-generated work and then training the model to create “new” work based upon that entered data.
• Nearly all data that AIs have ingested, including text and image AIs, has been without permission from the original creators. This means that AIs are built on large volumes of copyright-protected material that they are using without permission. 
• AI companies have long argued that their use of that source material is allowed because it’s a fair use. Their argument for that, primarily, has been how transformative AI-generated works are. 
• By shifting the focus away from how transformative the use is, the Supreme Court devalued the best argument in favor of AI companies. Now, transformativeness must be contrasted with other elements, most notably how the new work competes with and/or replaces the original in the marketplace.
• That, however, is not likely a discussion AI companies are eager to have:
  • Stock photographers, for example, should have little trouble proving that the new works are used to compete with stock photos.
  • Likewise, journalists should have no trouble showing how AI-generated text is used to replace news articles.
• Things get even worse when you realize AIs often are tasked with producing works that are “in the style of” a particular creator, making works that are designed to directly compete with that artist’s work.
• To be clear, this doesn’t mean that those suing AI companies are a lock to win. The issue is still complicated, and the Supreme Court made it clear that tranformativeness is still very much a factor and AI companies still have arguments in their favor. 
• However, the argument that AI companies have largely based their businesses around has been severely weakened, and that should give them pause.

“There’s a very good chance that the first major fair use cases after the Warhol ruling will involve AI.”

• Right now, there’s a wide range of thoughts on that.   Some feel that this is a game-changing moment for fair use and that AI companies need to license their use as their fair use arguments are “discredited.”
• Others think that the Warhol decision will be interpreted narrowly as, like most fair use cases, the outcome was heavily fact-specific.
• More recently, the Supreme Court ruled in the Google v. Oracle case, with a deliberately narrow ruling that focused only on programming code and the facts of that specific dispute.  However, we’ve already seen lower courts cite this ruling, including when it’s not directly applicable.
• Simply put, lower courts have a bad track record of limiting the scope of a Supreme Court fair use decision, even when the Supreme Court expressly says they should. 
• Could AI’s transformativeness be enough to save it? That’s what the courts have to decide.
• Two weeks ago, many, if not most fair use experts, felt AI had a clear upper hand. Now, the arguments in favor of AI are much more complicated and nuanced.
• The Warhol ruling, by most accounts, has added a lot of complexity back to the issue of fair use, and no one is feeling that more strongly and more immediately than AI companies.

Bailey concludes:  “We won’t really know what the impact of the Warhol ruling is until after the first courts begin to cite it. That, in turn, will likely take a few months.  One of the adages of the legal world is that the legal field is constantly lagging behind technology. While largely true, this may be one time when the law successfully cut off a technology by getting ahead of it.” (3)

https://oodaloop.com/archive/2023/05/28/when-artificial-intelligence-goes-wrong-2/