The decision was based on the conclusions reached by another federal judge while overseeing a lawsuit against an AI art generator.
A federal judge has rejected a majority of Sarah Silverman's claim for Meta's unauthorized use of the author's copyrighted book trainee intelligence model, the second court ruling in favor of an AI company in this legal battle over novel intellectual property issues.
U.S. District Judge Vince Chhabria on Monday rejected one of the authors' core theories, that Meta's AI system is itself an infringing derivative work that is only possible with information extracted from copyrighted material. "He wrote in the order:"This makes no sense. "We cannot understand the llama model itself as a recasting or adaptation of any of the plaintiff's books"。
Silverman's other argument was that every result produced by Meta's AI tools constituted copyright infringement, but this argument was dismissed because she did not provide evidence of any of the outputs"It can be understood as a recast, conversion or adaptation of the plaintiff's book"。Judge Chabria gave her lawyers a chance to reopen the claim, as well as five other claims that were not allowed to be filed.
Notably, Meta has not filed a dismissal allegation that copying books to train its AI models has reached the level of copyright infringement.
The decision builds on another federal judge's oversight of a lawsuit by an artist suing an AI art generator for using billions of pieces from the internet as training data. In that case, U.S. District Judge William Orrick similarly questioned the underlying argument in the lawsuit that artists could prove infringement when the AI tool did not create the same material. He called the allegations"Flawed in many ways"。
Some of the issues raised in the lawsuit may determine whether creators will be compensated for using their footage to train a chatbot that mimics humans because of the potential for such robots to cripple their labor. AI companies insist that they don't have to get permission because they're protected by the fair use defense for copyright infringement.
According to the complaint filed in July of this year, Meta's AI model"Every piece of text in the training dataset is copied", and then"Gradually adjust its output to bring it closer"Expressions extracted from the training dataset. The lawsuit revolves around the assertion that the entire purpose of LLAMA is to imitate copyrighted expressions and that the entire model should be considered infringing derivative works.
But Judge Chabria called that statement"Not true", because there is no allegation or evidence that LLAMA (short for Large Language Model Meta AI) is based on an existing copyrighted work"Recast, transform, or adapt"Target.
Silverman's other main theory – and other creators suing AI companies – is that every output produced by AI models is an infringing derivative and that companies benefit from every response initiated by a third-party user, allegedly constituting vicarious infringement. The judge concluded that her lawyer – who was also the ** lawyer suing the artists of StabilityAI, Deviantart and Midjourney – concluded that her lawyer"Mistakenly thought"--Since their books were copied in their entirety during the training of Llama, there is no need for evidence of substantially similar outputs.
Judge Chabria wrote:"To succeed on the theory that LLAMA's results constitute derivative infringement, the plaintiff does need to allege and conclusively prove those achievements'In some form contained'Part of the plaintiff's books. His reasoning is identical to that of Orrick, who argues in his lawsuit against StabilityAI"The derivative work of the alleged infringer must bear some resemblance to the original work or contain protected elements of the original work"。
This means that in most cases, the plaintiff must provide evidence that the infringing work produced by the AI tool is identical to its copyrighted material. This can pose a significant problem, as they have in some cases already admitted that none of the outputs may exactly match 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 an infringement has occurred.
Other claims rejected by Chabria included unjust enrichment and violations of competition law. He considers these claims excluded to the extent that they are based on surviving claims of copyright infringement.
Meta did not 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 authors' lawsuits and filed in federal court.