The Decibel - How much AI music is in your playlist?

Episode Date: March 20, 2026

What does it mean to create music? Would you be able to tell if the sounds you were listening to weren’t made or sung by a human? Artists and supporters of the music industry are asking some of these existential questions, as technology in consumer generative AI has grown in leaps and bounds and record companies are settling legal battles with AI companies. Josh O’Kane, Globe reporter on arts and business, joins The Decibel to talk about the friction between musicians and AI tools, how AI slop has found itself onto streaming feeds, and what this could mean for the future of music. Questions? Comments? Ideas? Email us at [e-mail] Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Dartmouth musician Ian Janes received an email from Spotify last January. It said, quote, make your release strategy a hit and went on to detail how Ian could promote his new album on the streaming service. But the problem is, Ian didn't make a new album. This is not Ian's music. And you might have guessed from the sound, but this is AI generated. This is Ian James. It sounds. I need you.

Because I can't forget the feel of her body and her lips. It sounds nothing like the AI music posing as his, but it still took him weeks to get Spotify to remove the fakes. Ian is far from the only musician who's experiencing this. Artists already get fractions of a penny in revenue for streams on services like Spotify. The rise of AI-generated music could mean even less money. Meanwhile, the music industry seems to be changing its tune on artificial intelligence. Some major labels have already signed licensing deals with AI companies.

So today, Josh O'Kane is here. He's a reporter for the Globe covering the business of arts and culture. He's been looking into the proliferation of AI music, how the industry and fans are responding, and what it all means for the artists and their craft. I'm Cheryl Sutherland, and this is the decibel from the Globe and Mail. Hi, Josh. Thanks so much for joining me today. Thanks for having me. So, Josh, we've seen examples where AI is used to make realistic looking photos or videos.

But when we're talking about using AI to make music, how does that work? And what are people actually doing? And what does it sound like? So it has been debated over the last three years. Like when you sort of generate something with AI, it is sort of taking little bits and pieces of whatever it's been scraped for the large language model often. And it is sort of recombining it into what you have asked for. So when you're looking at AI music, you're looking at music that has sort of taken bits

of music that have been scraped by the model. And you are listening to a recombined version based on however good the service is that has been used to make the song. So as an example, when I was working on this story, I decided to just set a prompt into the service Suno, which has recently signed a licensing agreement with Warner Music, one of the big three major record labels. And I put in a prompt to say, could you make a song that sounds like Gordon Lightfoot that's about eating 100,000 pies? Like how ludicrous of a prompt could I possibly put in to see what it could generate? And what I got was something that sounded more like weird contemporary folk. It was kind of eerie. Like I think it took the signifier of Gordon Lightfoot. This did not. not sound like the wreck of them in Fitzgerald. It sounded like something that someone kind of quickly threw together to be on like indie pop radio.

Cherry red and apple gold. Stories baked in crust of old. And it was a little eerie. It was a bit strange. But this is, you know, we were only three years into consumer degenerative AI. Like two years from now. And if there's more licensing of catalogs and if you were able to, like if the executor of Gordon Lightfoot's estate says, yes, of course, please do this. I don't know if they will.

Then we could actually get something much closer to what Gordon Lifewood's music sounded like. Okay. And so in that case, when you're talking about this Gordon Lightfoot generated song, was it his voice? It was not his voice. No, it was just a very generic male voice that sounded kind of very radio friendly. Almost as if they took the prompt of Gordon Lightfoot and sort of broke it down. Okay, what does a folk artist sound like? We're common and popular folk artists.

Who do we have that we have licensed to make sound like Gordon Lightfoot? And it's sort of this strange approximation appeared in my ears. And how were the lyrics? They kind of made it a song about traveling down a highway made of pies. Oh, okay. Yeah, you know, listen, sometimes maybe they just don't get verbs yet. I'm not really sure. But it was like a dead serious song.

about 100,000 pies. And I think it was called endless pies highway. I love it. Interesting. Okay. Okay. So these AI tools, they're different than music tools that have been used for decades like Auto Tune, right?

Yeah. So Autotune takes something and augments it. You know, originally Autotune was just meant to correct imperfections in a person's voice in a recording. Circa 2006, 2007, when you have artists like T. Payne and Kanye West, using it more as a tool to just. just we're going to sound like robots. And sort of really brought in a sound

two decades later, it is used as a tool to augment existing artistry. And so, you know, there was, even Jay-Z, I think had a song called Death of Auto Tune. You know, there was pushback to that, but then it became fully embraced. With AI tools, particularly the idea of just generating a whole song, what it is is you're taking what already exists, blowing it up, the entire catalog, whatever the service has pulled from, and recombining existing art into what is allegedly a new piece of art. But the critics would argue that it is not a new piece of art.

It is simply a facsimile. Yeah. And we'll get into kind of this idea around copyright issues. But let's continue on about what this is and where it ends up. So how does this music end up on streaming services like Spotify or Apple Music? So someone just goes and they find a distributor or they try on their own to upload this music. Sometimes they do it on their own. Sometimes they're using their own tools.

They may not necessarily be using sort of the more popular publicly available tools like Suno and Udio. I'm not sure what exactly that a lot of the AI music that we're hearing flooding, streaming service is coming from. but it's basically just someone, you know, I'm imagining someone sitting in their parents' basement, trying to upload something to generate those amazing fractions of a penny per stream, to try to get on popular playlists or in some cases grift off someone's actual name and upload songs under an existing artist's name to try to grift off of their brand to try to get a few bucks here or there. Yeah, okay, and we heard a story about an artist that happened to,

Ian Janes, the musician from Dartmouth. And he had the experience of having music uploaded in his name that he didn't actually make. What kind of checks are there on these services when something like this happens? There's no universal checks right now. I mean, in any large system, you are going to see bad actors try to engage in fraud. Right now, we're sort of still really at the beginning of AI music and the relationship between streaming services and music. So people are getting away with this.

And some people are just saying, I'm going to generate AI music on my own, call myself an artist in doing so, and see if people think I'm a real artist. A great case for that would be the Velvet Sundown, which sort of flooded streaming services six months or so ago. And there's still a bit of a lingering mystery of the origin of this quote unquote band.

Yeah, tell me about Velvet Sundown because they're not a real band, right? Tell me. As far as I know, the Velvet Sundown is just a band that sort of sounds a little bit, sort of 70s, fokey, a little progginess. Even their press photos, I'm using air quotes here in the studio, have this sort of fake sheen on them, like they themselves have been generated using Gen. AI services. You know, they take existing sounds that are pretty popular, and they started getting seated onto popular playlist, and suddenly they were getting thousands, if not hundreds of thousands of. streams on their songs. Wow. And so this is all AI generated. This is what we are told. This is, you know, this is still an ongoing mystery of this particular band. People have come out and said, oh, I am the Velvet Sundown, like they're Spartacus or something. But that mystery, as far as I know,

has not been solved. And someone is making money off these streams. I mean, you don't make a lot of money off streams, but someone's making some amount of money off of these streams. Yeah, let's get into this money part of it. Because in the case of Ian Janes, where, you know, someone has uploaded an album that's not his. Why would someone do this? Like, what kind of money are they making, uploading AI music posing as an actual artist or becoming a fake band? Potentially a fake band. So you, you know, I've been reporting on streaming services for like almost 15 years now, and you get really a fraction of a penny per stream. You need scale to actually make money off streaming music. And what's interesting is so when I interviewed Ian Janes, he described it as

like, okay, listen, like you're not going to make a million dollars off of just copying my name and hoping that people don't notice when, you know, I start showing up with this music I didn't write on streaming services. But if you do that 10,000 times to artists who might not be monitoring their streaming presences very closely, then you're, you know, maybe you make 10 bucks off of pretending to be Ian Janes. You multiply that times 10,000. Then you've got some real scale and some potential money to be given. It's not clear exactly how this works, but it's happening in an increasing amount in such a way that it feels like it's going to be a major factor in how we listen going forward.

So we know there have been issues around intellectual property when it comes to how these generative AI large language models are trained like chat GPT. Is that the case with these music AI tools? So this is what's been really interesting over the last four or five months is that you're actually seeing the big major labels as well as Merlin, which is a group of independent labels, negotiate with these services after previously trying to take them to court. Suno and Udio are the two probably most well-known prompt-based AI generated music services. And they were in litigation with most of the major labels up to about October, November.

Then you started seeing this flood of press releases coming out saying, we have settled with service X. So it's not every label and every service yet. but Suno, as an example, settled with Warner Music, and then Warner agreed to license its catalog to Suno. Now, Warner's actually said musicians have to opt into letting their music be used in the training data for these services. And that's something we're seeing across the labels that are signing these deals. So that the argument from the labels from my interviews with the industry is that, hey, people are clearly wanting to generate their own songs like in a style. of X or Y, recreating songs that they like in different genres, that sort of thing.

If that's going to happen, what the labels want is maximum income for themselves and the artists that they represent. They're sort of using the argument of technological inevitability, which is the same argument made by social platforms 20 years ago, that people are going to use this. So we want to be in a position with the most bargaining power to license these. And so starting four or five months ago, we saw the end of lawsuits in the beginning of licensing. So, I mean, the argument here basically from the music industry is that if you have a license with these companies, then you're actually going to make a deal and actually get money when they're using some of their artists, basically. Exactly.

And one of the really interesting things that I still, you know, I asked everyone about this. And no one's really sure how this is going to shake out. As I said earlier, when someone's dream is your song, you get a fraction of a penny. Yeah. Now, let's say you've got a catalog of a million songs, and then you blow that catalog up into a million pieces and they're recombined into a song. What is one millionth of a fraction of a penny? What it really leads to is like no one really knows if artists are going to be compensated in a meaningful way for this.

And that's something that's going to shake out maybe in earnings reports or potential future lawsuits where, you know, settlements become public and then we'll actually get a sense of the actual financial viability of all of this. If you can't beat them, sign and licensing deal with them. I mean, that's kind of what happened after in the music industry before, they post Napster, right? Yeah, yeah. There was a lot of pushback against Napster.

Then, okay, well, what about the iTunes Music Store in 2003 in that era? Like, okay, well, people are going to download music, why don't they just charge 99 cents a track? The music industry has always adjusted to these sorts of things and they're going to continue to adjust this. There's obviously always tension at the start and then there's a negotiation and then the contract change over time. We'll be right back. So, Josh, when we look at who's affected by the advent of AI and music, I can think of three groups. There are the record companies, which we kind of talked about already, the musicians themselves and listeners. So I want to talk about what this means for the musicians themselves.

First of all, are there any musicians embracing AI in their work? Yeah, you're seeing different versions of that kind of slowly unfold into the public eye. I think there are a lot of musicians who are actually using AI tools in the studio that haven't gone public with it. I try to reach out to a whole bunch of different artists. And the vast majority of people that I did talk to who were willing to discuss the issue were deeply, deeply against the idea of embracing AI. Now, there are definitely people who have made announcements that say, hey, we're going to try to use this technology for good. One example would be Grimes, the electronic musician. She has basically offered a royalty split.

If you want to emulate her voice on a track using AI, that's great. She gets some royalties. You get some royalties. Everybody wins. She's fine with that. She said so at least in 2023 or so. The artist Caribou on his latest album under that name from 2024 used AI to manipulate his voice in a copyright-friendly way so that you could have a different range of expression on songs like, only you or come find me. His voice is being manipulated across those recordings.

And it's a very interesting, occasionally controversial approach. Okay. What have you heard from artists who are against AI's use in music? So they're pretty emotional about it because what this cracks open for musicians is the endless debate between are you creating a piece of art that reflects human expression? Or are you creating a product that is meant to be consumed as a form of entertainment? This debate began long before the rise of generative AI and it will go on for many, many, many, many, many years. Huge tension there. Yeah.

And so a lot of the musicians who do want to talk about it are very unhappy with what's happening. You know, I talked to Mac DeMarco, who is this great singer-songwriter, originally from BC, who's back in BC now. He's got songs like Chamber of Reflection. He basically described AI music to me as like, are we sort of heading towards a future

where human beings are just being plugged in, like in the Matrix? That were just batteries for some other, machines version of expression. Chad Van Galen, the musician from Calgary, I think it's best if I just read his quote that he gave to me,

where he described AI music as this black hole of culture for these people who never had a soul, end quote, and then beginning of new quote, 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. Wrong words. Yeah, these are people who believe that music is fundamentally human expression and then what is being used is an energy and water-intensive technology to create a facsimile. I mean, we talked about money a little bit already, but how do musicians feel about that part of it? Are they concerned about what this means for pay? There's so many different

ways that you could look at that. I mean, you could look at the fact that more AI-generated music on streaming platforms, which is the predominant way that music is listened to outside of, say, TikTok or YouTube. And if more and more AI stuff gets on there, that's just, they're crowding out songs that were written and recorded by human beings. Now, again, we don't know how much they might get if their own music has been used. That's the base for training these AI models, and maybe they'll get some amount of money for that.

But again, it's not clear how that math works yet. But if the charts and if streaming playlists start really filling up with this, they're simply going to be crowding out. human artists at the expense of people who are genuinely trying to create human expression for a living. Okay, so we talked about the music industry. We talked about the musicians themselves. Let's talk about the listeners. Is there any regulation that requires AI music to be identified? As a listener, how do you know if you're listening to something that was partially or entirely made with AI? Right now there's no one major labeling system that is required through any regulation.

and I mean with regulation, it's hard to really work. I mean, Spotify is sort of based between Sweden and the U.S. The French service Deezer does have pretty strong labeling systems, but that doesn't necessarily mean that a company in France is going to have its rules guaranteed by another country such as Canada or the United States. So it's kind of a Wild West out there. Some people don't know, even if there's labeling out there, the average person just clicks play in a playlist,

and they're just going to hear songs that they like, and maybe they'll click Fave on one or two songs, and not even realized, never clicked that song, and see if it's even labeled. So it really is a Wild West out there. Yeah. And I mean, when it comes to other AI-generated content, I'm thinking about videos, images, sometimes we're duped. I feel like sometimes I'm duped, which is, you know, embarrassing when you are.

But so when it comes to music, is it easy to tell if something is AI, or is it easy to be duped? It's becoming increasingly sophisticated in music. And one of the really interesting factors, actually, that Ian Jains had brought up to me is that over the past 20 years, the evolution of digital technology has given any recording this sort of real polish to it. That doesn't, you know, human voices are very, very often auto-tuned now. And that's fine. But what it does is it means that sort of like that's slightly inhumanness you might hear in a voice or in the way that a guitar is recorded or a synth line that may have been played directly through a computer. it has this polish to it that is easier for an AI to replicate.

And so it is not that difficult for a song to sound like AI. It doesn't have that sort of jarring feeling that you might notice in a video or a photo or someone has six fingers. It's harder to tell. How new of an idea is this, like an artist that isn't a person? That's a great question because it reminds me in particular of two great bands. We've got the gorillas out of the UK, and then we've got Canada's own Prozac, a purveyor of such great songs, A Strange Disease and Sucks to Be You back in 1999. And, you know, these are, you know, artists that were cartoons that were sort of proliferating through music videos. They didn't have a human face in either case, really for a little while afterwards.

You knew that there were human artists behind it, but part of the entertainment was that they were these avatars. And if you look at artists such as the Velvet Sundown, like they've got these AI, apparently AI generated avatars that represent the apparent members of the apparent band. But there is this sort of very slight difference, which is that, you know, Prozac and Gorillas were humans who put up a mask. AI generated bands. Yeah, there's a human who's saying, I want this to sound like this or that, but the actual creativity is not there. And you could argue that artists such as guerrillas and Prozac were sort of another interesting leap in human creativity because there was a whole sort of backstory that was being generated. They were creating lore with these bands. You can generate lore with something like the Velvet Sundown.

But at the end of the day, it's, you know, because we still don't really know 100% about that alleged band, we don't really know what level of human creativity went into that music. I want to talk about solutions here. Is there any kind of policy change around copyright that would offer better guardrails around AI? Yeah. So different jurisdictions around the world have different lines in their copyright legislation around text and data mining and whether there's exceptions in copyright law. Groups in Canada, such as Socan, which manages copyrights for songwriters and publishers in Canada, are actually trying to ensure that there is no exception to copyright law around mining data for the sake of. of using it for generative AI. This is sort of a heated debate in different jurisdictions around the world and whether or not this should be exempted from copyright law.

And I don't know what the framework is going to wind up being two months, let alone 10 years from now. To end here, Josh, what does it mean for the music industry if we're outsourcing creativity? How do we see AI changing music and our experience of it? You know, when I talked to a lot of musicians over the last few months about AI and music, a lot of them took a step back and said, we need to reinvest in live music spaces, which was something that really stuck with me because it came up in a number of conversations. You know, a great example would be the band Austra. When I was talking to Katie Selmanis, who's sort of the mastermind of Austra, she and I were talking a great deal about like, she makes electronic music. Yeah. But she records it and writes it in such a way that could be performed with a live band,

that there's still this communal nature that brings people together to experience music. And I think you're going to see, I mean, there's also a venue real estate crisis we can talk about in another episode. But people are really talking about like they want to commune with their actual fans, their actual audiences and to really double down on that. And I think that gives a little bit of hope. Yeah, yeah. sounds like this might lead to the rise of more live music. I would hope so. Well, Josh, this has been really interesting.

Thank you so much for being on the show. Oh, thanks, Cheryl. That was Josh O'Kane, a reporter for the Globe, who covers the business of arts and culture. That's it for today. I'm Cheryl Sutherland. Our intern and associate producer is Finn Dermot. Our producers are Madeline White,

Rachel Levy McLaughlin and Mahal Stubber. Our editor is David Crosby. Adrian Chung is our senior producer, and Angela Pichenza is our executive editor. Thanks so much for listening.