This was published 11 months ago
Opinion
Fake Bowie songs will soon be indistinguishable from the originals
Michael Dwyer
Journalist“Oh man, this gonna be dope. Drake’s lyrics suck. AI can write better lyrics than Drake. Now they got his voice.” It was late, I was driving, so I’m paraphrasing my 17-year-old’s gleeful embrace of another ominous development in the world of music.
Drake the man is alive and well. The Toronto teen TV star turned R’n’B phenomenon is still the No.1 most-streamed musical artist on the planet. Drake the brand, though, is dead in my house. Hijacked by generative artificial intelligence.
Google Winter’s Cold, for example. You won’t believe it’s not Drake, but that’s the least of it. “Drake may wanna talk to these AI creators because he hasn’t gone this hard in over five years,” reads one YouTube comment. Translation: real Drake bad, machine Drake good. It’s a sentiment cruelly echoed as you scroll.
The trash talk is just a reflection, perhaps, of a standard mid-career phase most successful pop artists face, in which fans have grown just nostalgic enough to prefer his old stuff to his new stuff. The difference is that now, suddenly, they can make the good stuff themselves.
The “how?” is no secret. Creators will sometimes list the apps they’ve used: “Lyrics created with Open AI … vocals created with Uberduck text-to-speech,” and so on for accompanying images and video. These tools are numerous, free and fast-evolving, with zero technical proficiency required at entry level, though plenty of geek hours involved in perfection.
I’ve been a little rattled, frankly, by the old-school fakes infiltrating my socials of late. The digitally generated likes of Freddie Mercury singing Yesterday and John Lennon ambushing a recent Paul McCartney song are amazing, albeit with hints of the “uncanny valley” digital cinema is fast escaping.
Results are variable. The sandpaper and sneer are there in fake Kurt Cobain’s Black Hole Sun, but the uncharacteristic vibrato is the mistake of a rookie intelligence. And David Bowie would never sing about anything as tediously literal as the pyramids of Egypt or a fast-food drive-thru, as he does in one creation. Even if that artificially generated yelp is kind of … troubling.
Hits for these music nerd oddities are only in the tens or hundreds of thousands so far. The point, of course, is that in another year or two – maybe weeks, given the machine learning curves we’ve been warned about for decades – these cute pastiches will be fine-tuned to the point where an allegedly “lost Freddie demo” will be perfectly persuasive on a Queen playlist.
Joining the rapidly unfurling dots here, the reality of getting Alexa on a whim to “play Adele doing The Dark Side of the Moon with solos by Miles Davis” is surely just a matter of time. The “when?” has little to do with the corporate legal battle lines unfolding as we speak.
The first salvo has been fired by Universal Music Group, which owns the rights to about one-third of the global music market, Drake included. Music trade magazines reported last month that streaming platforms such as Spotify and Apple had been instructed by Universal to remove AI-generated songs from their services.
Melbourne music business lawyer Andrew Fuller tells me that a clause he’s never seen before popped up in a contract for a new band last month. Not only would the label own the musicians’ names and likenesses as usual, but their voices too. Picture a future jury asked not to consider Ed Sheeran’s chord progressions, but the vocal timbre of a robot that’s been listening (god help it) to hours upon hours of his music.
Speaking at Ibiza’s International Music Summit two weeks ago, another Canadian R’n’B star, Grimes, had this to say about it: “Copyright sucks. Art is a conversation with everyone that’s come before us.
“It’s cool to be fused [with] a machine. Why shouldn’t people be able to use my voice? Something cool and beautiful might come from it,” she said, while reasonably requesting that AI co-creators please cut her in for 50 per cent of their royalties.
The uberduck.ai platform (“Make Music with AI Vocals”) soon responded with their Grimes AI Challenge: a $US10,000 song competition open to public voting this week. Compare the speed of this development to the US Senate hearing of AI developers in Washington on Tuesday, which raised very real questions spanning copyright insecurity to human annihilation (no, really), without answering any of them. As usual, the move-fast-and-break-things philosophy of tech-savvy creatives is light years ahead of public comprehension, let alone regulation.
We’ve been here before, of course. Going back to that late-night car conversation, my son’s giddy reaction to looming anarchy reminded me of the day I downloaded the file-sharing service Napster onto my PC in 1999. The room was practically spinning as I stole my first folder of Radiohead B-sides. The digital genie was unbottled. What could go wrong?
We wished no ill on the artists, of course — except for Metallica, the ugly millionaires who took to suing their fans to make a statement about the value of art. They had an excellent point, as it turned out.
Most of today’s 17-year-olds have never paid for recorded music. Most in my experience find the idea laughable. They pay big time at the box office instead, and they don’t bat an eye when artists like Drake perform alone to pre-recorded soundtracks. “That’s just the way music is now,” my son tells me when I ask him where the musicians are.
Did those artists believe that they were indispensable? That pop’s time-honoured cult of personality would be respected and spared in the cruel churn of a marketplace proven to value nothing more than convenience? We may be about to find out.
The Opinion newsletter is a weekly wrap of views that will challenge, champion and inform your own. Sign up here.