US Judge Rules in Favor of Meta in AI Training Copyright Dispute

A U.S. judge ruled in favor of Meta on Wednesday, dismissing claims from authors who alleged that the tech company breached copyright laws by using their works without consent to train its Llama artificial intelligence.

San Francisco District Court Judge Vince Chhabria determined that Meta’s use of these works was “transformative” enough to qualify as “fair use” under copyright law, marking the second victory for AI companies in court this week.

Nonetheless, the ruling came with a warning that authors might have made a compelling case that utilizing copyrighted materials to train such advanced generative AI could allow tech companies to create tools that enable countless users to compete against them in the literary market.

“Regardless of how transformative (generative AI) training may be, it’s difficult to see how it could be considered fair use to employ copyrighted books in order to develop a tool that could generate billions or even trillions of dollars while potentially creating an unending array of competing works that might hurt the market for those books,” Chhabria noted in his ruling.

Training large language models driving generative AI requires substantial amounts of data.

Musicians, authors, visual artists, and news organizations have filed lawsuits against various AI companies for using their content without authorization or compensation.

Typically, AI firms defend their practices by claiming fair use, arguing that training AI models on extensive datasets fundamentally alters the original material and fuels innovation.

“We appreciate today’s court decision,” a Meta representative said in response to an inquiry from AFP.

“Open-source AI models are driving transformative advancements, productivity, and creativity for both individuals and businesses. The fair use of copyrighted materials is an essential legal framework for developing this innovative technology.”

In this case, a group of authors accused Meta of downloading pirated copies of their works to train the open-source Llama generative AI, according to court filings.

Books referenced in the lawsuit include Sarah Silverman’s comedic memoir “The Bedwetter” and Junot Diaz’s Pulitzer Prize-winning novel “The Brief Wondrous Life of Oscar Wao.”

“This ruling does not imply that Meta’s use of copyrighted materials to train its language models is legal,” the judge clarified.

“It solely indicates that these plaintiffs made the incorrect arguments and did not substantiate the appropriate claim.”

Concern for the Market?

Earlier this week, another federal judge in San Francisco supported AI company Anthropic in its effort to train its models on copyrighted books without authors’ permission.

District Court Judge William Alsup ruled that the training of the Claude AI models using both purchased and pirated books was permissible under the “fair use” doctrine in the U.S. Copyright Act.

“The use of the contested books to train Claude and its predecessors was exceedingly transformative and fell under fair use,” Alsup wrote in his decision.

“The technology in question is among the most transformative that many of us will see in our lifetimes,” Alsup commented, likening AI training to the way humans learn from reading books.

This ruling arose from a class-action lawsuit brought forth by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who accused Anthropic of illegally reproducing their books to train its chatbot, Claude, a competitor to ChatGPT.

Alsup denied Anthropic’s request for a blanket exemption, ruling that the company’s practice of downloading millions of pirated books to create a permanent digital library was not justified under fair use protections.