A federal judge has denied a copyright infringement lawsuit from a group of authors who sided with Facebook's parent meta platform and accused the company of stealing works to train artificial intelligence technologies.
Wednesday's ruling from US District Judge Vince Chhabria was second in a week from a federal court in San Francisco to dismiss a major copyright claim from the book's author against the rapidly developing AI industry.
Chhabria discovers that the 13 authors who sued Meta “had a false argument” and throws the case. However, the judge also said the ruling was limited to the authors of the case and does not mean that it is legal for Meta to use copyrighted material.
The plaintiff's lawyers — a group of well-known writers including comedian Sarah Silverman and writers Jacqueline Woodson and Tanehishi Coates — did not immediately respond to a request for comment Wednesday. Meta also did not respond immediately to requests for comment.
“The ruling does not represent the proposition that Meta uses copyrighted material to train language models,” wrote Chhabria. “It only represents the proposition that these plaintiffs had made the wrong argument and were unable to develop a record in favor of the right thing.” Meta won the demand to dismiss the case, which could turn out to be a victory for Pyrrhaic. In his 40-page ruling, Chhabria repeatedly demonstrated why she believed that Meta and other AI companies had become serial copyright infringers in training their technology in books and other works created by humans, and appeared to invite other authors to present it to his court in a way that allowed them to proceed with the trial.
While raising questions about whether companies are engaged in illegal activities by supplying copyrighted materials to AI training models without permission, the judge wrote: Chhabria said, “In many circumstances, it is illegal to copy copyrighted work and train a generative AI model without permission. This means that you must pay copyright holders the right to use the material in order to avoid liability for copyright infringement.”
The judge also scoffed at the argument that advances would slow down during critical technology times by requiring AI companies to comply with decades-old copyright laws. “Technology is certainly groundbreaking,” writes Chhabria. “But the suggestion that an unfavourable copyright ruling would stop the technology on its truck is ridiculous. These products are expected to generate billions of dollars for the companies that are developing them. Using copyrighted work to train models, as companies say, they will find ways to compensate copyright holders.”
On Monday, US District Judge William Alsap ruled that the humanity of the AI company did not break the law by training chatbot Claude on millions of copyrighted books, but the company still has to go to trial to illegally obtain those books from the pirate website instead of buying them.
However, the actual process of AI systems has been distilled from thousands of written works and is “typically transformative” textual passages that have been certified as “fair use” under US copyright law, allowing them to create their own textual passages.
In his Meta ruling, Chhabria criticized Alsup's inference on artificial cases, claiming that “Alsup focused on the transformative nature of the generated AI, putting aside concerns about the harm that it causes to the market for its trained work.” Chhabria suggested that such harm could be filed.
In Meta's case, the authors argued in court filing that Meta was “liable for large-scale copyright infringement” by obtaining the book from a pirated online repository and supplying it to Meta's flagship generative AI system llama.
Long, clearly written textual text, as seen in books, is extremely useful in teaching generative AI chatbots to human language patterns. The author's lawyers argued that “meta should have paid” to purchase and license these literary works.
Meta rebutted in a court application that US copyright laws “allowing the illegal copy of the work to be transformed into new ones,” and that the new AI-generated representations coming out of chatbots are fundamentally different from trained books.
“After nearly two years of lawsuits, there is no evidence that anyone has used llamas as an alternative to reading plaintiffs' books, or even what they can do,” argued Meta's lawyer.
Meta says that the llama will not output the actual work that it copied, even if asked to do so.
“You can't use llamas to read Sara Silverman's childhood descriptions or the story of Junot Diaz, a Dominican boy who grew up in New Jersey,” the lawyer wrote.
Meta, accused of pulling these books from online “Shadow Libraries,” claims that the method used “has nothing to do with the nature and purpose of its use,” and that the results would be the same as if the company had instead signed a contract with the actual library.
Such a transaction is how Google built an online Google Books repository of over 20 million books, but also fought a decade of legal challenges before the US Supreme Court in 2016, before a lower court ruled that rejected a copyright infringement claim.
The author's lawsuit against Meta has abdicated CEO Mark Zuckerberg and revealed internal conversations at the company about the ethics of using a long-standing pirate database.
