“There’s some good stuff. Two [tracks] in particular were very, very good,” the musician Grimes told Wired’s Steven Levy in a just-published interview, regarding her experiment with releasing an A.I. model that lets anyone make songs with a simulacrum of her voice. “They’re so in line with what my new album might be like that it was sort of disturbing. It’s like, ‘Who am I, and what am I here for?’ On the other hand, it’s like, ‘Oh, sick, I might get to live forever.’ I’m into self-replication…That would be the dream. A self-replicating pop star.”
It's no surprise to see the techno-optimist considering the upside of digital immortality—her latest (and unfortunately far from finest) single is called "I Wanna Be Software," after all. But many of Grimes’ fellow artists are still hung up on that first thought, about A.I. essentially making them redundant after training on their works.
When fellow Canadian Drake saw a fake song by A.I.-Drake go viral earlier this year, he wasn’t impressed, commenting: “This is the final straw A.I.” His label, Universal Music Group, had the track removed from streaming services by lodging copyright claims and talked about its artists’ rights being violated.
But now Universal seems set to embrace the trend. According to the Financial Times, the music-publishing juggernaut is in talks with Google about creating a tool that would let people use its artists’ voices or lyrics to make new, A.I.-generated tracks. Copyright holders would get paid (how much isn’t clear; Grimes takes a 50% cut of royalties generated by A.I.-Grimes songs) and artists wouldn’t be part of the scheme unless they opt in.
Google espoused quite a different approach in its response to an Australian government consultation on A.I. regulation. The Guardian reports that Google argued generative A.I. systems should be allowed to freely train on copyrighted content under a new fair use exception in Australia’s copyright law, and that “entities that prefer their data not to be trained in using A.I. systems” should be able to opt out of this happening.
This would apparently be a bit like online publishers using the robots.txt web standard to tell search engines such as Google not to crawl their sites. “We believe it’s time for the web and A.I. communities to explore additional machine-readable means for web publisher choice and control for emerging A.I. and research use cases,” Google said in a blog post last month—anyone wanting to join the discussion on that proposal can join the mailing list here.
I think the web-A.I. analogy has its limits. The mass indexing that was fundamental to the web’s development was all about sending people to the source—an obvious win for publishers, which is why I find the idea of Google or Meta paying publishers to send them traffic so daft. Google and Microsoft show citations in their Bard and Bing Chat responses, but that’s really just a fact-checking thing. To a far greater degree than with search results, people use chatbots to get a definitive answer rather than a springboard to somewhere else.
Asked by Wired’s Levy for her opinion on artists resisting their work being used as A.I. training fodder, Grimes replied: “We do need to change the legal and economic structure. But if you’re an artist, how could you not find it beautiful to be building the soul of an alien?”
Depends on who owns the “alien” and what the builders are getting in return. More news below.
Want to send thoughts or suggestions to Data Sheet? Drop a line here.
David Meyer
