Sep 21, 2023 2 min read

Meta says AI training is fair use in response to author lawsuit over LLaMA

Meta has responded to a copyright lawsuit filed by Sarah Silverman and two other writers insisting that its training of generative AI model LLaMA with existing copyright-protected works is ‘fair use’

Meta says AI training is fair use in response to author lawsuit over LLaMA

Meta has responded to the copyright lawsuit filed by three authors in relation to its generative AI model LLaMA.

Although not a music case, lawsuits like that filed by writers Sarah Silverman, Christopher Golden and Richard Kadrey are testing the copyright liabilities of AI companies that train their models using existing copyright protected works. In its response, Meta confirms that it believes that such usage of existing works constitutes 'fair use' under US copyright law.

The music industry - alongside all the other copyright industries - is adamant that if a technology company copies existing works onto a server in order to train an AI model, it needs to first get a licence from whoever controls the copyright in those works. However, some in the tech space reckon that such activity is allowed without licence, at least in some countries.

In the US, that relies on the concept of fair use, which allows the use of copyright-protected works without licence in certain scenarios. Exactly when fair use applies is always up for debate and, in the context of AI, that debate will likely happen through litigation like that filed by Silverman, Golden and Kadrey.

Meta states that its LLaMA model was developed by being exposed to "massive amounts of text from various sources, such as code, webpages and books ... this process involved dissecting text into trillions of word snippets or letter combinations (called 'tokens') and extracting a vast, complex set of statistical correlations as to how tokens are most likely to be arranged coherently".

"Copyright law does not protect facts or the syntactical, structural and linguistic information that may have been extracted from books like plaintiffs’ during training", it adds. "Use of texts to train LLaMA to statistically model language and generate original expression is transformative by nature and quintessential fair use - much like Google’s wholesale copying of books to create an internet search tool was found to be fair use in Authors Guild v Google Inc".

So there you go. Despite presenting a pretty forceful fair use defence in its legal filing, Meta actually reckons there are grounds for dismissing much of the authors' complaint for failure to properly state a claim. Either way, it seems likely that one of these AI copyright lawsuits will ultimately result in a big fair use debate in court.

Silverman, Golden and Kadrey have also sued Chat GPT maker Open AI, which also responded by citing fair use, before arguing that elements of the lawsuit should be dismissed for other technical reasons. But that hasn’t stopped another group of writers - including John Grisham, Jonathan Franzen, Mary Bly, Jodi Picoult, David Baldacci and George RR Martin - from filing a similar lawsuit against Open AI earlier this week.

Great! You’ve successfully signed up.
Welcome back! You've successfully signed in.
You've successfully subscribed to CMU.
Your link has expired.
Success! Check your email for magic link to sign-in.
Success! Your billing info has been updated.
Your billing was not updated.
Privacy Policy