- Meta trained its AI on pirated content.
- The practice is acceptable as ‘fair use’.
- Redistribution of copyright material by BitTorrent in question.
Since it became apparent that large language models were trained, in part, on copyrighted material, companies like Meta and OpenAI have invoked the ‘fair use’ argument to limit their liability. The law, drawn from 17.U.S.C. § 107, details limitations on exclusive rights by authors, artists, and other creators. The argument posited by big tech is that an AI ingesting an authors’ copyright materials does not pose a threat to the work in question as it’s not redistributed to the detriment of the rights-holder.
However, in the class action lawsuit filed by Sarah Silverman, Richard Kadrey and Christopher Golden against Meta, a third amended complaint has been allowed by presiding Judge Vince Chhabria. The plaintiffs allege that Meta used BitTorrent software to collect training data from the pirate shadow library, LibGen. And, as part of the BitTorrent process, it’s claimed that Meta seeded the material it collected for further illegal dissemination.
The default setting for most BitTorrent client software is to automatically ‘seed’ downloaded materials – that is, host the data and allow it to be further distributed on the BitTorrent network. By doing so, “Meta knew it was facilitating further copyright infringement by acting as a distribution point for other users of pirated books,” the amended complaint reads.
Meta has already admitted that it used unofficial sources, including pirated books, as training data for its AIs, citing the well-worn defence of ‘fair use’. Meta employees working on the company’s AI were uncomfortable with the practice, according to internal communications. These included a message that read, “Media coverage suggesting we have used a dataset we know to be pirated, such as LibGen, may undermine our negotiating position with regulators,” and employee Nikoly Bashlykov’s internal communication, “I think torrenting from a corporate laptop doesn’t feel right”.
Pirated content use OK by Zuckerberg
An internal memo from Meta presented to the court stated that the issue of the use of BitTorrent-ed materials had been escalated “to MZ” (Mark Zuckerberg), and the team “has been approved to use LibGen.”
The amended complaint reads, “By downloading through the bit torrent [sic] protocol, Meta knew it was facilitating further copyright infringement by acting as a distribution point for other users of pirated books. Put another way, by opting to use a bit torrent system to download LibGen’s voluminous collection of pirated books, Meta ‘seeded’ pirated books to other users worldwide”.
Meta has been asked to produce logfiles and seeding data from the BitTorrent software used, to show how much information was seeded to third-parties.
Previous cases brought on grounds of infringement of copyright against large AI providers have been at least partially dismissed in various proceedings over the last few years. In a case brought by the same class action representatives against OpenAI, Judge Araceli MartÃnez-OlguÃn dismissed the plaintiffs’ claims, citing fair use, as “nowhere in plaintiffs’ complaint do they allege that defendants reproduced and distributed copies of their books.”
If Meta does hold records of seeding data in the case brought against it by Silverman et al., the court may consider the act of redistribution not a ‘fair use’. However, it’s not normal for BitTorrent clients to log such information, so Meta may still escape censure, despite concerns raised by employees involved in gathering the training corpus.
