AI music has hit one of those moments where individual news stories come thick and fast, creating a bigger picture of how rapidly the technology (and the business models and debates around it) is evolving. Four stories today offer the latest snapshot of how the sands are shifting.
In a nutshell: artists are experimenting with generative AI; artists are also being experimented on with it; money is pouring into the industry from investors; and regulators are scrambling to keep up with developments.
First, artist Grimes’s desire to make her voice available for AI-assisted musicians and developers really is more than just tweets. She’s launched her Elf·Tech website, where people can register; upload their vocals, and receive them back as a ‘GrimesAI-1 Voiceprint’ to use in their tracks. Stems of her voice have also been made available for people to train their own models.
As trailed in her tweets, Grimes is asking for 50% of sound-recording royalties from any tracks commercially released, with an option to handle distribution through Elf·Tech for $9.99 a year. “Grimes is now open source and self replicating,” she tweeted. “plz excuse any hiccups or bugs – this is all new and we’re workshopping it!”
Elsewhere in AI music-land, Drake can have a break from being at the forefront of cloning controversies for now: it’s time for The Beatles to have their latest moment. Listen to this YouTube video with a recording which has taken Paul McCartney’s 2013 track ‘New’ and added in backing vocals from an AI John Lennon.
They’ve done the reverse with Lennon’s posthumously-released ‘Grow Old With Me’ too. “We love you, lads. No copyright infringement intended. This is an AI creation,” explains the ‘New’ video listing. We’ll see what the rightsholders make of ‘no copyright infringement intended’ in the coming days, but given the current industry climate, takedown notices seem likely.
(To expand on a point we made last week, beyond the legal status of such tracks, there’s an ethical/moral question here. If anyone’s going to use AI to bring Lennon and McCartney back together, shouldn’t it be the choice of McCartney and Lennon’s estate? Getting permission would be a stronger proof that you truly love the lads… However, these tracks are a pointer to what could be achieved by artists who want to explore – with all the permissions – these kinds of ‘what if’ scenarios with late bandmates.)
Away from hobbyist projects, the funding continues to increase for commercial generative-AI companies. The latest fund investing in them has a strong music link: Sound Ventures was co-founded by veteran music manager Guy Oseary. It has just closed a new Sound Ventures AI Fund backed by $240m.
“We anticipate that AI is going to play an impactful role in everything we do — including entertainment, an industry we know well,” said Oseary in a statement. Sound Ventures already has stakes in two of the biggest players in generative AI: OpenAI and Stability AI.
(Talking of OpenAI, it has just picked up its latest tranche of investment: $300m valuing the company at $27bn-$29bn.)
Finally, to the prospect of new regulation of generative AI in the European Union, with proposals that music rightsholders are likely to find to their liking. The European Commission has been drafting its AI Act for two years already, covering all manner of applications of AI technologies.
Now it’s got a provision focused on copyright, which could require generative-AI companies to (in Reuters’ words) “disclose any copyrighted material used to develop their systems”. According to the report, this was a compromise reached after some lawmakers tried to include a full-scale ban on copyrighted material being used to train AI models.
Separately, Reuters has some good background on how the new provision was “hammered out” in just 11 days by a “small group of politicians” striving to catch up with recent developments in generative AI.
The music industry, with its Human Artistry Campaign, has been arguing that “the use of copyrighted works requires permission from the copyright owner. AI must be subject to free-market licensing for the use of works in the development and training of AI models”.
In theory, the new European rules should be welcomed by rightsholders. In practice, they’ll have some questions. If this legislation has been drafted in a rush, will it get proper scrutiny to iron out any issues before it becomes legislation?
How will the rules on disclosure be enforced? And will European action spur similar laws elsewhere in the world, or create safe havens for training where the rules are much more lax?