A new court filing in an AI copyright lawsuit against Meta adds credit to previous reports. The company reports that it has “suspended” with book publishers on licensing transactions to provide training data to some of its generative AI models.
This filing was done by Kadreyv. It relates to the Meta platform case. This is one of many such cases involving the US court system that has pitted AI companies against authors and other intellectual property owners. In most cases, the accused (AI companies) in these cases argue that training of copyrighted content is “fair use.” The plaintiff – the copyright holder – opposed loudly.
The new application filed in court on Friday would negotiate an AI training data license for the book, including a partial transcription of a meta employee deposit taken by an attorney for the plaintiff in the case. suggests that they felt it was not scalable.
According to one transcript, Sy Choudhury, who leads Meta’s AI Partnership Initiative, said Meta’s outreach to various publishers had encountered “a very slow intake of engagement and interest.”
“I don’t remember the entire list, but I remember first creating a long list from looking for the internet of top publishers” – from many of our cold calls, trying to establish contact Outreach to do so. ”
Choudhury added, “As you know, we got engaged, but not many, but there were a lot.”
According to the court’s transcript, Meta suspended certain AI-related licensing efforts after encountering “timing” and other logistical setbacks in early April 2023. Choudhury said some publishers, particularly fiction book publishers, have actually been found to have no rights to content that Meta is considering licenses in accordance with the transcript.
“In the fiction category, we would like to quickly learn from the business development team that most of the publishers we are talking about have expressed themselves that they actually don’t have the right to a license. Data for us,” Choudhury said. “And it’s going to take a long time to interact with all of their authors.”
Choudhury said during the deposition, Meta suspended licensing efforts related to AI development on at least one other occasion.
“We are aware of the licensing efforts, for example, we tried to license the 3D World of the various game engines and game makers of the AI research team,” says Choudhury. “And just like I’m talking about fiction and textbook data here, there was little engagement to even have a conversation (…) and in that case I decided to build my own solution. ”
Lawyers for plaintiffs, including bestselling authors Sarah Silverman and Tanehishi Coates, have revised their complaints several times since the case was filed in the US District Court of the San Francisco Division, the Northern District of California, in 2023. . The plaintiff’s lawyer filed it to determine whether it makes sense for Meta to cross-reference certain pirated books, among other crimes, to pursue a license agreement with the publisher. claims to have cross-referenced copyrighted books available for licenses.
The complaint also uses the “Shadow Library” to train several company AI models, including “open” models of the popular Lama series, using the “Shadow Library” that includes pirated e-books. accusing them of using. According to the complaint, Meta may have secured some libraries via torrents. Torrenting, a way of distributing files throughout the web, requires that the torrenter simultaneously “seed” or upload files that they are trying to “seed” or upload at the same time.