In most circumstances, it’s illegal for companies to use copyright-protected material to train their AI systems without permission or payment, a court said, while finding that a lawsuit from Sarah Silverman against Meta doesn’t present one of those cases.
U.S. District Judge Vince Chhabria on Wednesday sided with Meta on the novel legal question of whether AI companies are covered by fair use, the legal doctrine in intellectual property law that allows creators to build upon copyrighted works absent licenses. It’s the second decision this week finding in favor of an AI firm on the issue, with another federal judge on Monday ruling against authors in a separate lawsuit by concluding that Anthropic is on solid legal ground over the legality of training.
Still, Judge Chhabria cautioned not to extend his ruling as a defense of the practice since he was constrained by lawyers for the authors choosing not to advance certain arguments that he viewed as favorable to their side. Those arguments, in his view, relate to AI tools generating works that are so similar to creators’ works that they’ll compete with the originals and indirectly substitute for them.
“No matter how transformative LLM training may be, it’s hard to imagine that it can be fair use to use copyrighted books to develop a tool to make billions or trillions of dollars while enabling the creation of a potentially endless stream of competing works that could significantly harm the market for those books,” Chhabria wrote.
In Wednesday’s ruling, the court called Meta’s utilization of the books for training “highly transformative.” By learning from the authors’ works, the company’s large language model Llama can edit emails, write skits or provide translation services, according to the order.
The logic mirrors that of U.S. District Judge William Alsup, who concluded that authors in the lawsuit he’s overseeing don’t have the right to exclude Anthropic from using their works to train its technology as long as they purchased the books. Like any reader who wants to be a writer, the judge said, the Amazon-backed company’s AI tool draws upon works not to replicate or supplant them but to create something entirely different, according to the order.
Still, a finding that a work is “transformative” doesn’t automatically provide protection from infringement under copyright law. Other factors are considered in the analysis. This includes potential harm in the market for the copyrighted material.
Legal maneuvering played a critical part in Meta’s win. Lawyers representing the authors opted to pay barely any attention to the issue of Meta copying books to create a product that will likely flood the market with similar reproductions — a theory the court called a “potentially winning argument.” Instead, they focused on how Meta’s theft of authors’ books for training harms the market for licensing their works for that purpose.
Consider people using AI tools to generate massive amounts of text in significantly less time than it would take to write it themselves. In that scenario, they use services like ChatGPT or Claude to create and sell books, competing with the works used for training by OpenAI or Anthropic.
“It’s easy to imagine that AI-generated books could successfully crowd out lesser-known works or works by up-and-coming authors,” Chhabria wrote. “While AI-generated books probably wouldn’t have much of an effect on the market for the works of Agatha Christie, they could very well prevent the next Agatha Christie from getting noticed or selling enough books to keep writing.”
The court said that lawyers for the authors should’ve offered evidence that Meta allows users to create works that directly compete against Silverman’s memoir or Rachel Louise Snyder’s nonfiction works on domestic violence.
Also a focus of the ruling: the notion that using books to teach people is not remotely similar to using them to create a product that a single person could employ to generate countless competing books using a fraction of the time and creativity it would otherwise take.
Across dozens of lawsuits over the past two years, AI companies defending themselves from accusations of illegally hoovering up any and every creative work on the internet to train their systems have offered arguments that can be read as more rhetorical than legal: Don’t rule against us or you’ll stop the development of a groundbreaking technology.
To this, the court responded, “These products are expected to generate billions, even trillions, of dollars for the companies that are developing them. If using copyrighted works to train the models is as necessary as the companies say, they will figure out a way to compensate copyright holders for it.”
Meta still faces a separate claim related to allegations that it illegally distributed the authors’ books during the process to pirate their works.
The authors, who include Silverman, Ta-Nehisi Coates and Richard Kadrey, are represented by David Boies, Joseph Saveri and Matthew Butterick, among others.
