It started with a rising country band that hit 576,000 monthly listeners on Spotify for their debut album. The four handsome cowboys had an air of nostalgia about them, and a perfect collection of star-quality names too; Gabe Farrow, Lennie West, Milo Rains, and Orion “Rio” Del Mar. The only thing stopping their meteoric rise was a lack of physical existence — none of these guys were real.

By now, we all know the story of Velvet Sundown, a band that was revealed to be a complete AI fiction. Listeners suspected it, and Rolling Stone confirmed it. The mastermind was one pseudonymous Andrew Frenlon, who went as far as he did because he was “interested in art hoaxes”. What you might not know is this guy wasn’t real either, and the true Velvet Sundown, whose identity still remains a mystery, were pissed off, telling Rolling Stone: “We ask that reporting on us be based on verifiable sources — not fabricated accounts or synthetic media.”

This was an AI hoax within a hoax. Frenlon, or whatever their real name is, went on to publish a lengthy account of how this whole farce was pulled off, just to point out who the real genius was in all of this. “Suddenly, I had the crazy idea, what if I inserted an extra layer of weird into this story?” they wrote. “This is an example of social engineering… humans are usually the weakest link in any technological system, and often quite easily fooled.”

Events like these paint a picture of AI music as a big bait and switch, a space where listeners are left debating whether what they just heard is real or not. The funny thing is, you don’t need to fool anyone. One study showed that 60 per cent of music consumers would willingly listen to AI-generated music, while another survey revealed that 46 per cent of artists would consider using AI music tools in the future. The problem is: an even larger number are unwilling to admit it. In the same survey, more than 50 per cent of artists said they would hide their use of AI, fearful of how their audience might react.

All of this secrecy around AI music poses some interesting questions: Who exactly is creating AI music? How many AI bands are there on streaming platforms? How many people actually listen to AI music? And crucially, are these AI bands earning royalty payments?

If Spotify knows these answers, they aren’t willing to tell us. Without a clear policy on AI music, Spotify is starting to look a lot like the Wild West of AI, and harbouring a decent population of AI-generated country bands to boot. In spite of this, Spotify believes that the numbers don’t justify concern. “[There is] infinitely small consumption of fully AI-generated tracks on Spotify… there is no dilution of the royalty pool by AI music.”

The tools to identify AI music do exist, however, and Deezer is using them. It discovered that 20,000 AI-generated tracks were uploaded to its platform every day, and that 70 per cent of streams for AI music are in fact from bot accounts. Once again, AI music has led us to another case of fraud. This time, however, AI-generated music definitely was taking a bite out of the royalty pie.

To protect its royalty pool, Deezer have moved to tag music on its platform with an AI label while also barring AI music from landing coveted spots on curated playlists.

From all of this, we’re left to wonder—are we all being fooled by jokers like Frenlon? And should we believe Velvet Sundown when they update their biography (post-scandal) to assure us that: “This isn’t a trick, [it’s] an ongoing artistic provocation designed to challenge the boundaries of authorship, identity, and the future of music itself in the age of AI”?

The AI music debate is often positioned as technology vs. the luddites too stubborn to embrace it, but this argument is a flagrant red herring. History tells us that artists are the first to put new technology to the test. Kraftwerk didn’t hide the fact that they were making music with synthesizers, and Run-DMC weren’t ashamed of using innovative sampling techniques, which then delivered a golden age of hip-hop. History tells us that artists aren’t afraid to test the limits of technology when there’s powerful music to be made. So it’s highly unusual that AI music technology continues to be concealed in its creation and distribution — not to mention the underlying data that makes generative AI possible.

Several years into the “AI revolution”, and notably, all we have to show for it are poor quality riffs on existing music genres with nothing coming close to the explosion of genres that accompanied previous technological revolutions. Of course, that’s not to say that AI technology isn’t making a difference.

Stem separation software is hitting near-perfect levels of precision, and the same AI technology has unexpectedly led to the creation of powerful and novel new DAWs like RipX. Music companies are exploring the use of AI in mastering and reverb plugins, while indie developers are pushing the creation of completely new sounds in AI timbre transfer software.

AI isn’t inherently bad technology, it is possible to train AI models in an ethical way. But by letting it remain a shady secret, artists and listeners are becoming increasingly sceptical of any potential benefits it has to music —and you can see why. Since being caught, Velvet Sundown has rebranded as a “synthetic music project”, meanwhile, Suno, the software likely used to create the AI album, is being sued for copyright infringement “on a massive scale”.

It’s not too late for AI to find its legitimate place in music history, but it’s high time we bring it into the light of day so there can be greater transparency around its use. With that we can dispel the shadowy practices that concealing its real impact. There are many artists ready to experiment with AI, and plenty of listeners who will be the first to judge how good it really sounds. But only with an open culture between everyone involved will we find out what AI can actually offer music, and what it can’t.

The post Fake artists, real opportunities: Hoaxes and scams are holding back the potential of AI music appeared first on MusicTech.