A federal judge has dismissed most of Sarah Silverman’s lawsuit against Meta for the unauthorised use of the author’s copyrighted books to train its generative artificial intelligence model, marking the second ruling by a court siding with AI companies on novel intellectual property issues raised in the legal battle.
US District Judge Vince Chhabria on Monday flatly rejected one of the authors’ core theories: that Meta’s AI system is itself an infringing derivative work, made possible only by information extracted from copyrighted material. “This is nonsense,” he wrote in the order. “There is no way to understand the LLaMA models themselves as a recasting or adaptation of any of plaintiffs’ books.”
Another of Silverman’s arguments, that any result produced by Meta’s AI tools constitutes copyright infringement, was rejected because she didn’t provide any evidence that any of the results “could be understood as a recasting, transformation, or adaptation of Plaintiffs’ books”. Chhabria gave her lawyers a chance to refile the claim, along with five others that weren’t allowed to proceed.
Notably, Meta didn’t move to dismiss the claim that copying books to train its AI model rises to the level of copyright infringement.
The ruling builds on the findings of another federal judge overseeing a lawsuit brought by artists against AI art generators for using billions of images downloaded from the internet as training data. In that case, U.S. District Judge William Orrick also struck a blow at the core of the lawsuit, questioning whether artists could prove copyright infringement in the absence of identical material created by the AI tools. He called the allegations “deficient in numerous respects”.
Some of the issues raised in the litigation could determine whether creators are compensated for the use of their material to train human-mimicking chatbots, which have the potential to undercut their work. AI companies claim they don’t need to secure licences because they are protected by the fair use defence to copyright infringement.
According to the lawsuit, which was filed in July, Meta’s AI model “copies each piece of text in the training dataset” and then “progressively adjusts its output to more closely resemble” expressions extracted from the training dataset. The lawsuit revolved around the claim that the entire purpose of LLaMA was to mimic copyrighted expression, and that the entire model should be considered an infringing derivative work.
But Chhabria called the argument “not viable” in the absence of any allegations or evidence that LLaMA, which stands for Large Language Model Meta AI, was “recast, transformed or adapted” based on a pre-existing copyrighted work.
Another of Silverman’s main theories – along with other creators suing AI companies – was that every output produced by AI models was an infringing derivative, with the companies profiting from every response initiated by third-party users, which allegedly constituted an act of vicarious infringement. The judge concluded that their lawyers, who are also representing the artists suing StabilityAI, DeviantArt and Midjourney, were “wrong to say” that – because their books had been copied in full as part of the LLaMA training process – evidence of substantially similar outputs wasn’t necessary.
“To prevail on a theory that LLaMA’s outputs constitute derivative infringement, Plaintiffs would indeed have to allege and ultimately prove that the outputs ‘incorporate in some form’ a portion of Plaintiffs’ books,” Chhabria wrote. His reasoning echoes that of Orrick, which found in the StabilityAI case that “the alleged infringer’s derivative work must still bear some similarity to the original work or contain the protected elements of the original work”.
This means that in most cases, plaintiffs will have to provide evidence of infringing works produced by AI tools that are identical to their copyrighted material. This is potentially a big problem, as they have conceded in some cases that none of the output is likely to be a close match to the material used in the training data. Under copyright law, a substantial similarity test is used to assess the degree of similarity to determine whether infringement has occurred.
Other claims dismissed in Chhabria’s order include those for unjust enrichment and breach of competition law. To the extent that they’re based on the surviving copyright infringement claim, he found that they are preempted.
Meta didn’t immediately respond to a request for comment.
In July, Silverman also joined a class-action lawsuit against OpenAI, accusing the company of copyright infringement. The case has been consolidated with other copyright claims in federal court.
