Oct 20, 2023 - Technology

Music lyrics lawsuit could set AI copyright precedent

Illustration of a gavel icon with the handle made from a music staff covered in music notes made from zeroes and ones.

Illustration: Annelise Capossela/Axios

This week's music industry lawsuit against Anthropic adds yet another challenge to how AI firms train their large language models and offers a fresh reminder that generative AI remains a legal minefield.

Why it matters: The decisions courts reach in cases like this will lay the groundwork for decades of law governing AI.

Driving the news: Universal Music Group and other major record labels sued Anthropic on Wednesday for using its AI tool to distribute copyrighted lyrics without a licensing deal. Anthropic did not respond immediately to our request for comment.

  • The complaint focuses on OpenAI rival Anthropic's Claude 2, a chatbot that was released in beta in July.
  • Last month Amazon announced a $4 billion investment in Anthropic, joining Google's $300 million stake.

Conflict over LLM inputs and outputs has been brewing for a while and this lawsuit mirrors similar efforts from writers and actors suing AI companies over use of their work.

  • There are issues with both the way the AI engines are trained as well as the fact that, in some cases, they output content that reproduces part or all of a copyrighted work, Jason Peterson told Axios.
  • Peterson is the CEO of GoDigital Media Group, a media and technology company focused on intellectual property rights management and claims to be one of the largest independent music copyright holders in the world.
  • "This copying to storage for use in AI training is a likely clear-cut case of infringement, and the courts will put a stop to it," Peterson added in an email.

Zoom in: The record label claims against Anthropic, while similar to those made by writers and actors, could be easier to prove.

  • "These claims are stronger," Katie Gardner, partner at the law firm Gunderson Dettmer, told Axios. "Plaintiffs have identified output that is substantially similar (and in some cases identical) to the copyrighted input, and there is already a robust market for licensing music lyric data," Gardner said in an email.
  • "The brazen duplication of exact lyrics down to the letter is what sets it apart," says SoundExchange CEO Michael Huppe, who argues that artists should be entitled to the "three Cs" when it comes to AI use of their work: consent, credit and compensation.

The other side: AI companies are likely to argue that the training of their systems is fair use, with the AI systems simply scaling up the human knowledge process where people read up on existing information and use it to generate new ideas.

  • "Advocates for generative AI would say it's the same thing," says Charlene Liu, a partner at the law firm Haynes Boone.
  • However, there are key differences, especially when it comes to scale. "Generative AI is basically limitless," Liu said.
  • Liu said it's tough to know how courts will rule. "There is not much existing law because everything is so new," she said.

Between the lines: There's a long record of music publishers having greater success defeating "fair use" claims for lyrics than for other kinds of intellectual property.

  • Because songs are so short, any quotation from them risks being viewed by a judge as reuse of a "substantial" part of the work, which is one of the key standards for determining whether the fair use exemption applies.

Flashback: The cases courts decide during the early period of a new technology's deployment in society often set new rules — or provoke legislators to write new laws if they don't like the rulings.

  • 1991's Cubby Inc. v. CompuServe Inc. and 1995's Stratton Oakmont Inc. v. Prodigy Services Co. revolved around liability for online postings at a moment when online communication was a novelty.
  • Those cases inspired Congress to create Section 230 of the 1996 Communications Decency Act, which protected online publishers and platforms for years for liability over user content.

Our thought bubble: While the music business has become an early legal testbed for generative AI, the industry could also be one of the first to find a commercial path forward.

  • The industry has a track record of finding ways to credit and compensate dozens of people who contribute to an album in different ways. And, in the post-Napster era, the industry has also been keen on embracing new technologies, such as music streaming.

"AI is a fascinating technology and there are a lot of things it is going to do for the music industry," SoundExchange's Huppe told Axios. "Yes, there are these lawsuits out there. But the industry is also embracing AI."

  • Some tech companies and music rights holders are exploring ways that AI can create content that wouldn't otherwise exist, such as an artist getting to collaborate with a musician that is no longer living.
  • "I would say there is reason for optimism," Huppe said.
Go deeper