The Velvet Sundown - a non-existent ‘band’ releasing AI music and amassing 750,000 monthly listeners on Spotify - will probably not enter the industry annals for their creative output. It’s decidedly nondescript.
On the other hand, it isn’t terrible. As a result, it may well be that many people have been listening to the group’s tunes as a bit of palatable background music as they potter around doing other stuff. (In reading for this piece, I’ve discovered for the first time in my life, incidentally, that ‘Muzak’ is actually a brand name, rather than a generic term for elevator music. A bit like Portakabin® is a trademarked product.)
The CEO of French streaming platform Deezer says that 18% of the uploads there are now AI-generated and he anticipates exponential growth. The boss of Suno, meanwhile - one of the apps that allows people to generate music out of thin air from a text prompt - says that even Grammy winners are using his technology.
So where exactly does this leave us in the middle of 2025?
Well, there are three fundamental issues for musical artists. The first is intellectual property. We know that real-world music is used in the training of the models that produce the artificial kind. If I ask to generate a reggae song on a particular theme, the AI recognises what reggae should sound like in terms of perceived instrumentation, rhythm, time signatures and so forth. It can only do this because it’s been regularly consuming reggae on an industrial scale.
Artists understandably object to their original tracks being fed into a AI hurdy-gurdy, which cranks out stuff in the same style on demand and fails to compensate them for their talent and effort. There are legal cases on these IP questions that extend into the world of animated characters, works of fiction and so forth, as it’s a phenomenon that clearly isn’t restricted to music. It’s a battle for all creative practitioners.
The second issue is the sheer volume of AI slop that might be produced. Imagine feeds and recommendations on Spotify, Apple and other platforms being so overloaded with artificially generated tracks, that the real work - particularly from new and undiscovered artists - is crowded out. If you’re already relying on a system that might pay you $0.004 per stream, the prospect of fewer people noticing you is immensely frustrating.
The third - and perhaps most complex and unsettling problem - is over the question of whether anyone can tell if the music is real or artificial. Deezer labels the tracks as AI, but Spotify and other platforms sing a different tune. And this starts posing questions that go beyond the business models of the music industry or the ability of artists to make money.
If I know music to be artificial, am I less inclined to listen to it? If I give it a go, does my knowledge of the fact it’s AI-generated affect my appreciation of it?
Artists will live in fear that not only might their work be swamped by AI content, but also that people might quite enjoy the funky fakes. There is even an argument that says AI will be particularly adept at recognising the patterns and structure of music that most appeals to humans, based on a constant feedback loop. If x is included, people spend an extra y seconds listening. A continual process of iteration and refinement then follows, based around streaming patterns and real-world feedback.
Is it too much of a leap to imagine major AI hits in the coming years, with music every bit as good as that produced by some of the top human artists? And if it’s not too much of a leap, how does that make a talented young guitarist or singer-songwriter feel about their own prospects right now?
I hate to leave things on a low note, so I’ll conclude by mentioning an extraordinary and heartwarming use of AI. The country singer Randy Travis, who rose to prominence in the 80s, sadly had a massive stroke which affected his ability to perform. James Dupré recorded a new song last year called Where That Came From and AI trained on over forty of Travis’ original hits was able to transform Dupré’s vocals into those of his hero in his prime. The result is stunning.
As Travis heads into the sunset, his music is thankfully still way ahead of that produced by the Sundown lads.
Such a rich, layered read, equal parts concerning and compelling.
The rise of AI-generated music feels like we’re watching a slow collision between creativity and computation.
On one hand, you have artists battling for visibility, fair compensation, and even recognition that their real work exists in a sea of synthetic sound.
On the other, there's this eerie efficiency: AI serving up algorithm-optimized earworms that do just enough to keep us listening while we fold laundry or write emails!