Judge Says Authors Made the Wrong Arguments in Ruling for Meta on Fair Use

On Wednesday, the U.S. District Court for the Northern of District of California ruled that a number of well-known authors, including Sarah Silverman and Ta-Nehisi Coates, failed to successfully argue that the market for their works was significantly harmed by Meta’s use of the works to train its generative AI tool, Llama.

The original lawsuit was filed in July 2023 but was dismissed. The current suit was filed in December 2023 was trimmed down to focus on the social media giant’s alleged admission of using the copyright-protected books. In the lawsuit, the authors accuse Meta AI researchers of communicating about potential legal issues that might arise due to the use of copyrighted works in generative AI training. The authors wrote, “a key reason Meta chose not to share the training dataset for Llama 2 was to avoid litigation from using copyrighted materials for training that Meta had previously determined to be legally problematic.”

In a 40-page summary judgment ruling issued just two days after another landmark decision on whether AI inputs infringe copyright, Judge Vince Chhabria said that the copying of works for training large language models (LLMs) will usually be found to be infringing, but that in this case the plaintiffs’ arguments missed the mark. The authors contended that Llama’s ability to reproduce snippets of their works and the diminishment of their ability to license their works due to Meta’s use of the works without permission weighed against fair use. But Chhabria said they instead should have argued that “Meta has copied their works to create a product that will likely flood the market with similar works, causing market dilution.”

At the same time, Chhabria was clear that AI companies will not always win these cases. Addressing the general argument that ruling against AI companies  in these cases will “stop the development of this groundbreaking technology,” Chhabria called the suggestion “ridiculous.” He explained:

“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.”

Generally, said the judge, companies like Meta are going to have to figure out a way to pay creators for their works or face infringement liability, but based on the evidence presented here, while “this ruling does not stand for the proposition that Meta’s use of copyrighted materials to train its language models is lawful…these plaintiffs made the wrong arguments and failed to develop a record in support of the right one,” said Chhabria.

Ultimately, Chhabria found that fair use factors one, three and four favored Meta while only the second factor favored the authors. However, Chhabria was highly skeptical of Meta’s arguments on the fourth factor and outlined clearly how the plaintiffs could have defeated them. But “where a defendant introduces evidence of a lack of market harm, ‘and the plaintiff fails to introduce empirical evidence countering such a showing, the fourth factor should be weighed in the defendant’s favor,’” he concluded.

Chhabria again dismissed Meta’s arguments that the “public interest” would be “badly disserved” by a finding that copyrighted text could not be used by AI companies, calling it “nonsense”:

“As mentioned earlier, a ruling that certain copying isn’t fair use doesn’t necessarily mean the copier has to stop their copying—it means that they have to get permission for it. So where copying for LLM training isn’t fair use, LLM developers (including Meta) won’t need to stop using copyrighted works to train their models. They will need only to pay rightsholders for licenses for that training.”

But because the plaintiffs failed to win decisively on the fourth fair use factor here due to their lack of evidence of market dilution, “Meta is entitled to summary judgment on its fair use defense to the claim that copying these plaintiffs’ books for use as LLM training data was infringement.”

Image Source: Deposit Photos
Author: Skorzewiak
Image ID: 656716908