Dartmouth musician Ian Janes was running through sound check on a hometown stage last January when he got an e-mail from Spotify about promoting his new album. The problem: He hadn’t made one.
“Make your release strategy a hit,” the e-mail began, below the image of an album cover featuring a shadowy figure walking down a foggy street lined with buildings much taller than any found in Dartmouth.
Standing on the stage of a church, his mind toggling between perplexity and irritation, the guitar-slinging songwriter began to dig in. The album was called, in a feat of broken English, Street Alone. Other pieces of the story felt broken, too. The cover looked janky enough that it was probably generated by artificial intelligence. And the track names were ripped straight from popular hits, including Lana Del Rey’s This Is What Makes Us Girls.
Janes took a listen to the song. It bore no resemblance to Del Rey’s version or his own work. It consisted, mostly, of synthesizers, alongside strings and wordless vocals that sounded like they, too, were created artificially. In other words, the music bore the hallmarks of AI software.
Janes would spend weeks fighting to have the album removed from streaming. The ordeal confirmed to him that he needed to deepen his focus on performing in front of real humans. In concert, at least, Ian Janes fans would know they were hearing Ian Janes.
Annotated screenshot of a Spotify album that musician Ian Janes says he didn't write and believes, along with the image accompanying it, was generated using AI.Spotify/Supplied
The rise of consumer-focused generative AI over the past three years spurred, at first, near-universal condemnation from creative industries. Their concerns ran deep. Many AI models were alleged to scrape and ingest copyrighted works so that users could make their own songs, images, videos and writing in recognizable styles. Music companies began taking generative AI companies to court.
In the past few months, however, the dynamic between the recorded-music industry and AI companies has begun to flip. The sector’s biggest companies, including Universal and Warner, have been settling their legal battles with AI services such as Suno and Udio, and have even begun licensing their recording and publishing catalogues to them and other AI companies such as Klay Vision. Canada’s songwriting royalty-collection organization SOCAN, meanwhile, joined forces with American counterparts ASCAP and BMI to start accepting compositions that have been partly generated with AI.
Yet many working musicians continue to see AI as an existential threat, usurping the very notion of human expression with the click of a button. And there is a great deal of skepticism over how much money will actually be returned to artists, given how opaque the AI models’ processes are, and how broad the libraries are that those models scrape. After all, artists only get fractions of a penny for streams on services such as Spotify and Apple Music when their full songs are played.
Musician Ian Janes says the AI ordeal confirmed to him that he needed to deepen his focus on performing in front of real humans.Ian Selig/Supplied
There has always been a chasm between musicians and the music industry. AI is widening that chasm.
Janes’s experience shows that generative AI has already led to surging uncertainty about what music is authentic human creation and what is, in the parlance of our time, slop – slop that is sometimes deliberately marketed as a musician’s authentic work to skim money off their reputations.
“If they copied the names of 10,000 artists like me, that would scale up,” Janes says of the grift.
Things are only about to get more confusing. Suno and Udio let anyone create a song in seconds by prompting their software with ideas of what it should sound like. Up-and-coming producers are dabbling in AI tools to finish a lyric, say, or to convert a guitar sound into a violin. Hip-hop megaproducer Timbaland has gone all-in on the tech, even developing an AI “pop star.”
“It’s like fabrication,” says Mac DeMarco, the Vancouver Island indie singer-songwriter whose absurdism often tips over into an embrace of humanity – such as when he read out his home address at the end of a record so fans could visit him. “Soon we’ll all just be batteries, like in The Matrix.”
Katie Stelmanis, who makes music under the moniker Austra, sees the technology as antithetical to creativity. Her new album, Chin Up Buttercup, is an electronic record, but it’s designed to be performed in concert by her full band. AI, she says, “can imitate what exists, usually in a bland way, but it hasn’t yet produced something that feels innovative and special.”
Musician Katie Stelmanis sees AI as antithetical to creativity.Lamia Karic/Supplied
She adds: “If music does the job of creating culture and defining what’s cool, I can tell you with definitive certainty that AI is not cool.”
Calgary’s Chad VanGaalen has a darker take on the tech, which, beyond usurping human expression, is backed by data centres that require vast amounts of energy and water.
“It’s this black hole of culture for these people who never had a soul,” he says. “If you want to mess around with it and make a Christmas fart record, how many cups of water are you pouring down the drain? It’s a bloodstain.”
Even with just a free trial, it’s extraordinarily easy to make music using AI. Suno, which recently settled with and signed a licensing deal with Warner, lets users create and even edit songs with simple prompts.
Some limits do exist: When I entered the prompt “make a song in the style of Gordon Lightfoot about eating 100,000 pies,” I was told that Suno wouldn’t reference specific artists. And yet it took cues from that reference anyway, creating songs that “start with soft, fingerpicked acoustic guitar and smooth storytelling vocals.” Eventually it built a song about eating a six-figure quantity of pies; the tune had an eerie grace.
It’s not clear how much money this trial song, which Suno dubbed Endless Pies Highway, will return to Warner artists, nor what they would receive in the aggregate from the nearly 100 million users the company says have tried the platform. Such is often the lingering mystery of court settlements.
“Who will be the first of the major labels to put up their hands and say how much money they got, or is it all under non-disclosure?” asks Catherine Moore, an adjunct professor at the University of Toronto who studies the music industry.
The Globe and Mail contacted the Canadian offices of all three major labels – Universal, Sony and Warner – to discuss this moment in AI. In return, The Globe received an offer to interview Patrick Rogers, chief executive officer of Music Canada, the record-industry trade association that lobbies for their interests.
Patrick Rogers, CEO of Music Canada.Mike Highfield/Departure/Supplied
“To go from completely stolen, completely scraped, to introductory licences on this stuff, gives reasons to be positive,” Rogers says.
The key word there, he implores, is introductory. He likens this moment to when record labels began figuring out how to license their music catalogues for streaming and other, more specific use cases, such as for Peloton Interactive’s exercise machines. Those licences evolved; the first wave simply got the industry a foot in the door, legally.
“We are at the announcing stage, not the fine-details part,” Rogers adds. He imagines the music market bifurcating in the immediate future, as some fans and artists seek out and create deliberately AI-free content, while others embrace the tech’s potential.
“If we can make as many of those tools licensed, legal and authorized as possible, that’s the goal,” he says. “Music aficionados will have the other fights. They’ll be welcome to have those fights, and it’ll be easier and better to have those fights, as long as everything is licensed and legal.”
He knows some musicians are frustrated at labels’ embrace of AI.
“We absolutely do need to hear from the on-the-ground artists, but that doesn’t immediately make my members the villains of the story,” Rogers says. “There are no better people to negotiate these contracts with than the people who have access to the greatest catalogues in the world.”
Robert Kyncl, Warner Music’s chief executive, wrote in a blog post in late November that the label would only sign deals with AI companies that committed to artist opt-ins – as opposed to automatic participation – and that the company would licence its catalogue in ways that respected “the value of music.”
Though he did not provide detailed information on how much money artists will get from licensing their music to Suno or other platforms, “the more interactivity users have with the music they love, the more value is created,” Kyncl wrote. (Disney recently signed a similar licence with OpenAI’s Sora video-creation service, hoping fans will share social videos.)
The music industry’s main focus so far, Rogers explains, has been on AI’s inputs: the copyrighted pieces of music it breaks down and rearranges into new pieces. How the business treats those new pieces – the outputs – is yet to fully crystallize. For example, there is not yet an industry standard for whether AI-generated music should be labelled when uploaded to streamers, which now drive most of the world’s paid listening.
Illustration by Kyle Smart
SOCAN, though, is in the early stages of wrangling with the messy output that AI does generate, and the extent to which humans are involved.
“I don’t think there’s enough licensing in place yet, and I don’t think there’s enough transparency yet, as far as what goes in and what’s coming out – so I still think there’s a lot of work ahead,” says Jennifer Brown, SOCAN’s CEO. “Here’s the thing. We did this with digital. We’ve seen this with every single new tech. Things start slow; they start experimental.”
SOCAN’s own research, with the firm Pollara, has found that 87 per cent of Canadians believe it’s important that “the music they listen to is created by humans – not AI.” Still, the tech has become enough of a creative tool for musicians that SOCAN and its North American peers, which collect and pay out money for songwriting and composition copyrights, have begun registering songs that are partially AI-generated. Fully AI-made works are still no go.
“We want to protect human creativity,” Brown says. “That’s the reason that this organization exists – that music exists.”
There are nuances in SOCAN’s initial policy that will be bound to frustrate some musicians. When someone registers a song, they don’t need to explain how AI was used, which could create loopholes, such as if the vast majority of a track was created with software such as Suno or Udio. Brown says this policy choice is because of subjectivity: It’s hard to quantify AI use if it only, say, helped complete a line in a single verse.
“There are some that really don’t want AI to be part of anything that is within the system, and I totally hear that,” Brown says. But AI companies are proliferating anyway, she argues, and only a handful have started signing licences.
Globally, regulations around AI and copyright are struggling to keep up with the changing landscape. Some countries and jurisdictions have “text and data mining” exceptions in copyright law that can make it more permissive for AI models to scrape creative works. Groups such as SOCAN are trying to ensure no such exceptions happen in Canada.
AI has the ability to generate songs that sound like popular artists. In doing so, those songs – especially if uploaded to popular Spotify playlists – also divert casual listeners away from real artists who want to make a living off their art.
“It’s not a level playing field. You can have generative AI, not generating royalties, competing with a human songwriter,” Brown says.
A detection tool shows The Velvet Sundown's press photo is AI-generated, and it has visual inconsistencies that are the hallmarks of an AI image.Supplied
Sure, Janes got the fake music under his name taken down, but if someone uploaded a few hundred fake albums under real artists’ names and got 10,000 streams apiece, they could make decent money off of other artists’ work. “Ghost” artists have for years already flooded streaming services with lower-cost music that help their platforms’ profitability: It’s not difficult to hide these new evolutions of ghosts in plain sight.
Last year, a psych-rock “band” with the moniker the Velvet Sundown began racking up streams before it was outed as allegedly AI-created. The country charts were hit next, with Billboard determining that fast-rising songs by Breaking Rust and Cain Walker were, at minimum, AI-assisted. In mid-November, a third of its top-10 digital song-sales chart was made up of songs that used AI.
Vancouver company Beatdapp, which was formed last decade to help artists more accurately track streams and royalties, just released a suite of AI-detection tools to help the industry crack down on fraudulent claims of human creation. Claiming more than 96-per-cent accuracy, it analyzes sonic details such as timbre and timing to assess whether AI was used.
“Hammers can build houses, and they can also rob jewellery stores,” says Morgan Hayduk, Beatdapp’s co-CEO. “Some participants in this ecosystem will be good actors, and others will not be. From a platform or label standpoint, you need to be able to see both.”
Two of Canada’s most acclaimed electronic artists, Caribou and Grimes, have already embraced AI. Dan Snaith, Caribou’s mastermind, manipulated his own voice to sound like others’ on his 2024 album Honey, including a rap verse that music-criticism website Pitchfork said “edges toward racial ambiguity in a way that feels queasy.” The latter revealed software in 2023 that let users mimic her voice for a royalty split.
Canadian fiddler, Ashley MacIsaac likens generative AI to the spread of digital recording and editing software over the past few decades.Rob Gurdebeke/The Globe and Mail
Cape Breton fiddler Ashley MacIsaac is also excited to use AI as a tool with his next record – even after having a concert cancelled last month because a Google AI search summary falsely described him as a sex offender.
He likens generative AI to the spread of digital recording and editing software over the past few decades – or even DJs working a crowd with other people’s music. “It’s good for creativity,” MacIsaac says. “It’s hip, it’s modern.”
David Usher, who has been making music both solo and with the band Moist for more than three decades, sees the point of the music industry’s embrace of AI – given the rise of tools to, say, create a trumpet sound when you don’t have a trumpeter around. He worked last decade to develop an AI tool for lyric writing, and now runs an AI-embracing venture studio called Reimagine AI, based in Montreal.
But he cautions that it’s a much different technological advance than, say, Auto-Tune, which once stunned the industry. Creative fields have long compensated workers in relation to the scarcity of high-talent labour, he explains; generative AI is bound to reshape that market.
“I think we’re going to see the rise of human spaces,” Usher says. “I have no interest in creating music with artificial intelligence. I use AI every day as a CEO, but in music, I want to play for my friends, and for people who want to listen to it live.”
Like many musicians, he hopes that future regulations will help enforce how AI-generated music is treated in the marketplace. Until then, the market will keep changing.
In B.C., DeMarco, who just released an album called Guitar, sees the industry’s embrace of prompt-generated AI songs as a step toward an embarrassing-sounding future.
“There’s going to be new designations in the music industry, like, ‘We’re going to bring in this fire prompt writer,’” he says.
Janes, meanwhile, believes that the evolution of commercial pop music has already conditioned the listening public’s ears to soft-edged, synthetic sounds, paving the way for generative mishmash.
“What we’ve got to do as creatives, if we’re going to survive, is make something for which there are no prompts,” he says. “You can’t scrape the data off something that hasn’t been done.”