Court Issues Another Fair Use Ruling Backing AI Developers

It’s been a week for AI-related court cases, with two US federal judges handing down separate (and somewhat contradictory) rulings on whether using copyrighted materials without permission to train AI can be considered “fair use.”

The first ruling:

• As reported here, on Monday June 23 Judge William Alsup sided partially with AI company Anthropic, declaring that training its AI model on the copyrighted materials of the three authors who brought the lawsuit counted as fair use.

The second ruling:

• On Wednesday June 25 in the same court – the US District Court for the Northern District of California – Judge Vince Chhabria ruled on a class action against Meta (developer of the Llama large language model) brought by 13 writers including Sarah Silverman, Richard Kadrey, Junot Diaz and others.

• As per Music Business Worldwide, they claimed Llama had been trained on their works without permission, and that it “would even reproduce parts of those works when prompted.”

• While determining that training AI on copyrighted material without permission is not fair use in most cases, the judge granted Meta’s request for a partial summary judgment.

• He found Llama is “not capable of generating enough text from the plaintiff’s books to matter.”

• In regards to the authors’ claim that by Meta using their works without permission it diminished their ability to license their works for AI training he said, “the plaintiffs are not entitled to the market for licensing their works as AI training data.”

• His full summary can be found here.

Good news for rightsholders?

• As per MBW, Chhabria offered a glimmer of hope for rightsholders when highlighting an argument that could work in their favor, namely that “allowing tech companies to train AI on copyrighted works would severely harm the market for human-created works.”

• Chhabria also criticized Judge Alsup’s argument regarding the transformative nature of AI, and the parallel he drew between training AI models and training schoolchildren to write well.

• He stated: “Using books to teach children to write is not remotely like using books to create a product that a single individual could employ to generate countless competing works with a miniscule fraction of the time and creativity it would otherwise take. This inapt analogy is not a basis for blowing off the most important factor in the fair use analysis.”

• He also dismissed an oft-heard argument by AI companies that forcing them to license materials for training would slow down or even stop the development of the technology.

Common ground:

• Both judges agreed on one thing: using pirated material to train AI is not acceptable.

• Anthropic will front court in December to defend using material from online libraries known to offer pirated books to train its AI model.

• In the case against Meta, the judge has allowed one part of the authors’ case to proceed, in which they allege Meta used a torrent file-sharing network to download illegal copies of books, and stripped right management information from them, violating the Digital Millennium Copyright Act.