Meta Halts Book Licensing for AI Training, Court Documents Reveal

# meta # Lawsuit # court # books

In a recent development in the ongoing AI copyright lawsuit against Meta, new court documents have surfaced that lend weight to earlier rumors about the company hitting pause on talks with book publishers. These discussions were meant to secure licenses for training data to fuel some of Meta's generative AI models. The case in question, Kadrey v. Meta Platforms, is just one of several similar legal battles navigating the U.S. judicial system, where AI companies find themselves at odds with authors and other copyright holders. Typically, the AI companies, as defendants, argue that using copyrighted material for training falls under "fair use," a stance fiercely contested by the plaintiffs, who are the copyright owners. The latest filings, submitted to the court on Friday, contain excerpts from depositions of Meta employees, taken by the plaintiffs' legal team. These documents hint that some at Meta believed that negotiating licenses for book data to train AI might not be a viable long-term strategy. During his deposition, Sy Choudhury, who heads Meta's AI partnership initiatives, mentioned that their attempts to connect with various publishers were met with lukewarm reception. "I don’t recall the entire list, but I remember we had made a long list from initially scouring the Internet of top publishers, et cetera," Choudhury stated, according to the transcript. "And we didn’t get contact and feedback from — from a lot of our cold call outreaches to try to establish contact." He added that only a handful of publishers actually engaged with them. The court transcripts reveal that Meta put a hold on certain book licensing efforts related to AI in early April 2023, citing issues with timing and other logistical hurdles. Choudhury explained that some publishers, particularly those dealing in fiction, didn't actually possess the rights to the content Meta was interested in licensing. "I’d like to point out that the — in the fiction category, we quickly learned from the business development team that most of the publishers we were talking to, they themselves were representing that they did not have, actually, the rights to license the data to us," he said. "And so it would take a long time to engage with all their authors." Choudhury also revealed during his deposition that Meta has previously paused other AI-related licensing efforts. He cited an instance where they tried to license 3D worlds from game engine and game manufacturers for their AI research team but received little interest. "And in the same way that I’m describing here for fiction and textbook data, we got very little engagement to even have a conversation [...] We decided to — in that case, we decided to build our own solution," he explained. The plaintiffs' legal team, representing notable authors like Sarah Silverman and Ta-Nehisi Coates, has amended their complaint multiple times since the lawsuit was initiated in the U.S. District Court for the Northern District of California, San Francisco Division, in 2023. The most recent amendment accuses Meta of comparing pirated books with copyrighted ones available for licensing to evaluate the feasibility of entering into a licensing agreement with a publisher. Furthermore, the complaint alleges that Meta utilized "shadow libraries" filled with pirated e-books to train several of its AI models, including the well-known Llama series of "open" models. It suggests that Meta might have accessed these libraries through torrenting, a file-sharing method that involves uploading, or "seeding," the files one is downloading, which the plaintiffs claim constitutes copyright infringement.

Related article

Comments (33)