AI Art Appeal’s Procedural Flaws Put Broader Ruling in Doubt

An appeals court panel’s focus on procedural issues in a case involving efforts to copyright AI-generated work left attorneys concerned the judges may sidestep larger questions about how copyright law regards the emerging technology.

Computer scientist Stephen Thaler made his case for reversing a US Copyright Office and lower court rejection of his registration of an AI-generated image during oral argument Sept. 19 before the US Court of Appeals for the D.C. Circuit. But the panel appeared primed to find he’d waived the right to make one of his key arguments at the lower court.

Thaler initially argued in his 2018 application to register “A Recent Entrance to Paradise,” that it was created autonomously by his “Creativity Machine” AI program. The Copyright Office rejected his application, then a district court judge threw out his lawsuit, in part, because he’d waived the argument that he authored the work while continuing to maintain the machine was the author.

The D.C. Circuit panel suggested he may now be barred from arguing he’s the author because his appellate brief didn’t claim the lower court erred in that judgment.

While the appeals court hinted at ruling that works created solely by AI can’t be registered, such a decision would leave wide-open questions about the degree of creativity required to register an AI-aided work and the scope of the resulting protection. Gaps between art and the words used to generate that work — or the prompt — could further complicate those outstanding questions.

“The appellate court seems to be reluctant to reach beyond the record in the case and address some really interesting and intriguing questions that weren’t really fleshed out” at the district court, IP attorney Paul Llewellyn of Arnold & Porter Kaye Scholer LLP said. “It certainly seems like the appellate panel feels like the appellant here, Dr. Thaler, is trying to kind of shift his argument on appeal.”

AI will inevitably create valuable works, so whether someone can own it becomes critical, said Mark Lemley, an intellectual property law professor at Stanford University who also who represents Meta Platforms Inc. and Stability AI in copyright lawsuits over training data. Even if the court takes the Copyright Office’s current stance that non-human, autonomous creativity doesn’t merit protection, he said, it still would have a major impact.

“If the court says, ‘We agree with the Copyright Office,’ then anything generated by the AI gets no copyright protection, and I think the world of copyright and entertainment and media changes pretty dramatically,” Lemley said.

There’s already “a bunch of things out there being generated by AI,” Lemley noted, from elements within creative works like movies and video games to short articles by news organizations. A ruling against Thaler could put those elements into the public domain, he said.

“The copyright industries have not yet fully internalized what it would mean to say, ‘You don’t get any protection in anything the AI generates,’” Lemley said, later adding, “That might be a good thing.”

‘Role of the Creator’

After the Copyright Office’s initial refusals to register Thaler’s work, he continued to maintain nothing barred non-human AI from obtaining a copyright. He also argued alternatively that he authored the work using his program.

The Copyright Office issued a final refusal and Thaler sued in the US District Court for the District of Columbia. But the court found a work couldn’t be registered absent human involvement, and said he’d waived his argument that he authored the work by continuing to maintain the machine was the author. The Copyright Office has subsequently said in a report that while it is well-established that copyright protection is only for works that are the product of human creativity, AI-generated works can be protected if there is sufficient human authorship—an argument its attorney repeated during oral argument.

Thaler’s attorney alternatively argued Thursday that he’s the author of the work in question in part because he built the machine that produced it. Circuit Judge Patricia Millett rejected the notion, saying Thaler having built the AI program “has nothing to do with” the rights to the work it produced and noting “Kodak makes cameras” but doesn’t get a copyright to pictures taken with them.

But some attorneys aren’t so sure.

“The point of copyright protection is it should reward creativity. It should be associated with a human being, not a machine.” IP attorney Agatha Liu of Duane Morris LLP said. “But there’s merit in claiming the creator of the machine being an author.”

Thaler designed the Creativity Machine, and that human effort—arguably more so than the prompt—led to the creation of an original work, Lui said. If building AI that makes creative works could lead to a copyright, she said rights could come down to a case-by-case analysis of the degree of creative contribution provided by the user of an AI program or its programmer—possibly with them being co-authors.

“It would be a little unfortunate if there’s no further discussion on the role of the creator of the AI,” Liu said.

‘Tipping Point’

Some have analogized AI to a tool like a camera, but a prompt or even the AI software generally bear little resemblance to the creative work. Cameras themselves now incorporate AI, though the distinction appears to be the degree of control exercised by the user, as someone taking an AI-aided picture still makes many of the choices common to analog pictures.

If Thaler loses his appeal, it won’t close the door to AI-assisted works, but would leave unanswered questions about “the tipping point between machine-generated and machine-assisted,” IP professor Jane Ginsburg of Columbia University said.

The Copyright Office’s 2023 guidance says works produced by machines that operate “randomly” or “automatically” without creative human input cannot be copyrighted. While an element of randomness in a creative work doesn’t preclude registration, Ginsburg said, it’s the degree of human involvement that matters.

“Nobody’s saying that Jackson Pollock’s paintings are not protectable because there was some element of randomness in his throwing paint on canvases,” Ginsburg said. “But it’s really trying to discern at what point the ability of a human being to control the process yields to the decisions taken by the AI program.”

Protection of work heavily impacted by AI would likely be thin given the distance between creative contribution and the output, multiple intellectual property attorneys said. But it’s unclear exactly how AI involvement would impact a court’s substantial similarity analysis comparing an AI-assisted work with an allegedly infringing work.

“We’re going to see more cases like this where there are more facts, that involve more human input, that involve iterative process,” Llewellyn said.

The case is Thaler v. Perlmutter, D.C. Cir., No. 23-5233.