This lawsuit highlights one of the big problems facing AI-generated music: Using lyrics for AI training. [UPDATED]

[This article, first published on October 21, was updated with new information on the lawsuit. – AC]

Artificial intelligence cannot be creative on its own. Programs need to be trained in order to return results on queries. How do you train an AI? By feeding it data, language, words, video, audio created by actual humans. In other words, AI cannot exist without learning from us.

This is one of the points made by the writers and actors in the Hollywood strike. It probably won’t be long before someone will be able to tell an AI program to generate a script with the push of a button. But that AI-generated script will be the result of training on not just the work of countless writers and actors but directors, lighting people, special effects experts, costume designers–every gig that goes into making a movie or a TV show. Shouldn’t they have a stake in any AI creation?

It’s the same with music. An AI program can’t come up with a song about, say, a bad breakup without first ingesting dozens, hundreds, thousands of songs written by real people who had real breakups. Shouldn’t those artists be involved in any AI discussions?

This is the heart of the matter of a new lawsuit by Universal Music, ABKCO, and Concord Music Publishing against Anthropic, an AI program named Claude (in which Amazon has a US$4 billion investment) that will generate music on demand. The plaintiffs allege that Anthropic has been trained using copyrighted material without permission. The publishers are complaining that the company too “massive amounts of text”–lyrics–without asking.

The lawsuit names over 500 songs with infringed copyright, including “Uptown Funk” by Mark Ronson and Bruno Mars as well as Beyonce’s “Halo.”

I quote from a lawyer for the publishers: “[It is] “well-established by copyright law that an entity cannot reproduce, distribute, and display someone else’s copyrighted works to build its own business unless it secures permission from rightsholders.”

The use of lyrics by digital entities be they websites, apps, or bots, have always been contentious and litigious. Feeding lyrics to train an AI program opens up a whole new legal frontier with copyright that may not be settled anytime soon.

This will be interesting. (Via Hypebeast)

UPDATED INFO: Music Business Worldwide looks at five key takeaways from this lawsuit filing.