Going Down the AI Music Slop Chute with Camilo Landau

Can people tell the difference between a song made by human musicians, and one prompted by AI? Oftentimes not, it turns out. But that doesn’t make the distinction any less important. Dream Machines hosts Alexis Madrigal and Robin Sloan talk to longtime musician Camilo Landau, who also works in music analysis at The Music Genome Project at streaming service Pandora. Camilo and his colleagues have sifted through millions of tracks uploaded by musicians and users to catalogue their musical elements, all with the aim of curating personalized music recommendations. And he says most of the new music he’s seeing these days is generated by AI. On today’s show, they discuss whether AI music is connecting with listeners, delve into San Francisco’s AI slop music “scene,” and contemplate the consequences of a world where music is produced in the time it takes a finger to tap a screen. 

Guest: Camilo Landau, musician and music analyst

Alexis Madrigal I’m Alexis Madrigal.

Robin Sloan I’m Robin Sloan and this is KQED’s Dream Machines, a podcast about AI, how it works, what it means, how it feels. Yeah, how I feel. And importantly, it’s rooted here in San Francisco.

Alexis Madrigal So the other day, I was sitting with a friend of mine, prize fighter in Emeryville, great bar in the East Bay. And he was kind of blowing my mind about the current state of music. He works at Pandora now, but he’s a long time musician in Latin American music. He’s been like recording and playing music all over the Bay Area and the world since he was like 15 years old. And now that he works at pandora, he’s encountering this strange flood. Of things that is sort of like coming through the doors that is not made by any human being, maybe somebody pushed a button. And it’s AI-generated music that is now sitting there right next to the stuff that was made by someone with a heart and big toes and eyes and a brain. And a guitar. And a a guitar, perhaps, or maybe a theremin. And when we’re sitting at this bar, it was just kind of blowing my mind with this new world. Of AI and music. And it was basically saying something like, close to half of the songs that are getting on the streaming platforms are just AI music entirely. And that like, no one’s actually listening to it. So you have this strange situation where the platforms are being absolutely flooded with these songs. And also they’re just sitting there not even racking up the listens. So please welcome Camila Landa. Yes, thank you. It could be here.

Camilo Landau Yeah, it’s fascinating that the volume is just so enormous. Like you said, flooded with these AI-generated songs that nobody wants to hear, and no one asked for, and no ones listening to. And it’s these prompt-based platforms where you just say, I would like 1,000 melancholy pop songs, and then it just generates them faster than real time. So faster than you could play it or listen to it a single time. It’s actually just printing sound files. It’s not like creating music in the way that we think of assembling these different instruments and we’re recording them and layering them. Not like that at all. It’s like, I’ve seen 100,000 reggae songs, so I know what a sound file of a reggae song looks like and I can print something that looks very similar. I can’t I just cannot help imagine it coming out of

Robin Sloan the like the pasta extruder. Yeah, it’s a sausage, sausage extruder actually. Yeah, yeah, yeah.

Camilo Landau Just a tube, just a tube of audio. For me, that’s kind of the disappointment. So a lot of this I feel is a big disappointment because it’s what could, what could you be doing with AI generation of music? You could be making sounds that we can’t physically make. You could invent, anything you can imagine suddenly can be music. But we’re not seeing anything like that. We’re seeing just, you know, tens of millions of boring pop songs. So as you.

Robin Sloan At your, from your desk, see what’s uploaded day by day, is that your characterization that it’s far more of this kind of automatic slop?

Camilo Landau Yeah, I mean by far more, I mean all, all, you know, like, or maybe there’s like a few dozen interesting ones out of, you know, tens of millions. Okay, I’m going to, uh, point to this Deezer report, uh French, I believe they’re French music streaming service company. Deezers put out, um, a report on, on AI generated music, 45% they’re saying 45% of the music that’s delivered to platforms is AI generated. And this I think this is industry wide. Yeah. So that’s almost half.

Robin Sloan I mean, what a what a curve. This is new in the last two years. Yes. Yeah

Alexis Madrigal And, like, if we were to just hear it, could I tell the difference, do you think? Well, I mean…

Camilo Landau Also, according to Deezer’s report, I think it’s 97% of people. And they did this blind test in different countries. 97% percent of people can’t tell the difference. But even though they don’t know it’s AI music, they don’t seem to be liking it or connecting with it. So a very small percentage of the music that people listen to is AI-generated, even though almost half of it that is being created. Proof.

Alexis Madrigal Like the proof is in the pudding there, you know, right? Like people don’t, they don’t like it even if they can’t tell the difference. That’s like.

Camilo Landau Yeah. I mean, it really made me rethink what is music, right? Because I spent a long time learning how to play music good, right, and like mastering skills because I thought that like, yeah, if you can play it good, then that makes better music. Now, I think I was wrong. Right. The point of music is not to play a music good. The point a music is to have a human expression and connect with other people through human communication. And that’s the piece that AI can’t do.

Robin Sloan Oh, that’s so interesting. It’s almost like a unintentional natural experiment where they said, okay, let’s take out that technical competence piece and just do that. Let’s say like, cause you can do that before. To have technical competence, you had to also be a living, breathing human musician with like a heart and ideas. And they’re like, oh no, no, no, we can do it without that. And let’s see. And now we have the result of that experiment, which is that it is somehow hollow. Yeah, yeah.

Alexis Madrigal I mean, how do you even know, you know, from your position, like how do even know that it is AI generated? Like how, are there tells? Are there like detectors? Like how does that work?

Camilo Landau Yes, there are tells, there’re detectors. Well, sometimes they’ll just say, composer, UDio, Suno, something like that. So they just tell you. Yeah, and then sometimes it’s, oh, here’s a brand new artist who just dropped 30 albums and 1,200 songs or something like that and let’s look them up online. Oh, they don’t exist. There is no artist. They had a really busy weekend, okay? They were locked in. So some of the tells can be something like, all right, well, each song has a different singer, right? Like male, female, different range, tonalities, vocal qualities. Each song is a different style. So OK, oh, you got a hard rock song followed by a country song followed by electronic music, and completely disjointed, right. And so there’s another one. Fully produced songs that sound like incredible studio musicians, incredible recording. But maybe they didn’t master the record or something like that. Like, why would they have spent so much money on this production? And it still has this kind of like gauzy sound to it a little bit. And then it’s other sort of like, remember with photos you see like the third hand and stuff like that, so sometimes it’s stuff like that like the track starts super abruptly or ends super abruptly. Or… This section comes in in a super weird place where people wouldn’t put it. No human musician would ever score it like that. And then sometimes it’s almost the opposite. It’s like too formulaic and perfect. Yeah. Yes. AI has solved the wrong problem here. The wrong problem was never that we don’t have enough music, right? We’ve had plenty of music. We have more music than we could possibly listen to. But that piece of connecting with the artist, I think, is actually completely fundamental, right, because best case scenario, And it’s highly likely that an AI will produce a song that fits really well on a particular playlist and resonate really well with people. Cause it’s producing tens of millions of songs. So just statistically, it’s gonna be one that’s gonna fit really well. So you’re listening to whatever you listen, you hear a new song and you like it. Oh, that’s kind of cool. What is this? And you go and look it up and check it out. And you realize, I’ve been fooled. I’ve be tricked. I’ve bee deceived. That’s the feeling that you get. And, and cause that is the impetus. You want to connect with the artist and there’s no artist and it’s wild that.

Alexis Madrigal You would try to put out music without that stuff, because it’s so clear there’s that drive to know about where did this, what soul did this song come from? It’s like that gets kind of like maybe the core lie of the current AI era. Like the thing that we didn’t have to do, but that we did do was to like give these the chatbots and these things like an eye, to let them pretend to be a human. Like everything else could work exactly the same, and I feel like, oh man. I wish we hadn’t done that, you know? I wish they hadn’t been like, yes, I am a being in here that you can sort of connect with, because it is, it’s fundamentally deceitful, I think, about what’s happening behind the curtain. Sure, yeah.

Camilo Landau I think so. I mean, I think, yeah, like the tool itself isn’t the problem. It’s, somebody has to be making those decisions and that’s not happening, right? Like it’s the human decisions are what make the art, right. And, and there is no decision here. You’re just prompting a thing and then it’s going to make the decisions, right I mean we have music that’s completely electronically produced by humans using machines. Yeah, right, like, but somebody’s making those decisions somebody’s deciding this is where I want the hi-hats to come in. Or whatever it is, and that’s not what’s happening with AI. It’s like, print me 5,000 songs, and then in five minutes later, you have, you know.

Alexis Madrigal Robyn, so there’s none of that scene, there’s not of that like playing music good and transmission of all these things. But you went to what Passage for. You

Robin Sloan Yeah, maybe. Of AI music. Maybe, maybe it is the seed being planted, you know, as this, as the stuff emerges. I saw a digital poster for a scene, for a show in San Francisco, and it was called the Gen Punk Slop Show. Slop, right there in the title, not hiding it, celebrating it. It was in a big, cruddy art warehouse in Soma, you know, kind of hard to find. One. They had a stage set up, but the host, who’s also an open AI researcher, just deeply embedded in the AI scene and obviously a great enthusiast, announced that we were gonna hear a whole night, hours and hours full of songs, that’s sort of a sequence of performers, were they even performers, that they had organized and all this music had come straight out of the AI machines. And this was something we heard. That was magical, what a thing to hear music from nowhere. They started out with a set glitchy, pitched up, hyper-pop, like super high tempo. I will tell you, in those first moments, it was a wild thing to see a group of people dancing, I mean really dancing in that bad Burning Man way. It’s a music that came from nowhere, music without a musician, music without Bye! A reason behind it other than a prompt. This goes on through the night and I have to say, on these big speakers, it sounded good. I mean, it was technically very- But that beat drop. It was very competent. It was smooth. It shined in the right places. One of the producers at the end of their bundle of songs leaned into the microphone. There’s the only words we heard from them the whole time. They said, shout out Claude. Yeah. These are the new superstar DJs. Yeah. Claude does your taxes and produces your EDM. What a world. Superstar DJs! Yeah. And then I kept expecting there to be someone who maybe had, I don’t know, pieced parts together or used AI samples. But no, it was all these just like soup to nuts AI generations. And the thing that struck me the most, maybe the glimpse of something truly new in this event, was about midway through getting close to midnight, One of the performers came up, I say performer, but I don’t even know if that’s what they were doing. The figure took the stage wearing one of those great, shiny, reflective helmets. That’s the helmet? Yeah, kind of like dead mouse or something, one of these cool new cyber DJs. So they stood up to the front with a laptop, and they’re actually projecting their screen up on the wall behind them, and it was Suno. So just completely shamelessly, unapologetically, They were clicking through their generations. You saw the prompts, you know, dark EDM with skittering beats. And it just was, it was wild because already, you know, we had the figure of the DJ who would stand at the front of a stage with their laptop. And people would say, oh man, what is he even doing? He’s just like pressing play on an MP3. So this person was just pressing play on an mp3 that they hadn’t even made. And in a weird way, it was kind of radical. It’s like, okay, what are they now? What is their role? And is this the new performer of the 21st century? You get to wear a shiny helmet and press play and.

Alexis Madrigal Life is good. It’s a scene of some kind, you know, that essentially like celebrated the sausage. Yeah.

Robin Sloan You’ve never, boy, you have never seen more people more excited about AI Slab.

Camilo Landau I mean, I almost think that if it were sort of like live generation, that’s kind of more interesting because that’s almost the promise of AI that is not being explored much, which is why even print these songs in the first place? You can actually create the perfect song for this particular moment. I can check out your playlists, your play lists, my play lists. Know that we’re all in proximity. We have our Geolocation, we know what the weather is here. We can have our Apple watches on and can be reading our heart rate and biometrics. It can be adjusting in real time. The tempos of the songs based on all of those factors. Why bother printing that? That could be the perfect song for us three right now, right here. And that goes more to what the heart of what is music. It’s vibrations in the air. That might not be good for anyone else ever, which is what we’re seeing where nobody’s listening to these songs, right? But why not just like make something that’s for here and now. Vibe composing. Vibe, composing!

Robin Sloan I didn’t prompt that and so I left the show, walked out into the San Francisco drizzle, thinking that maybe that was the most interesting thing I’d seen, that it creates this new stand.

Camilo Landau The freedom to just create without responsibility, because you can always just pass it off. I didn’t actually create it, something else did.

Alexis Madrigal You do see like if AI can just produce a pop song perfectly, instantly, every time, the value of doing that as a human being just feels like it goes to zero, you know? Right.

Robin Sloan Yeah, of course, I want to believe that’s right, but I do think we underestimate the power of things being easy at our peril. You know, the last great music technology scare, you know, synthesizers created as simulations of, you know strings and, you know, brass and everything else. But like then. Drum machines. And the drum machine, right. And again, and as you all know, invented as accompaniment, you know, it was like, oh, you can’t afford a drummer, get this little box to go. Right

Camilo Landau But that organ, it’s attached to the organ.

Robin Sloan Yeah, and then, but then it was, but, but then it, you know, repurposed and adapted and, and sort of gave rise to wild new genres. I mean, these were tools that kind of had a grain and they pushed back and they both, um, demanded and sort of invited mastery and boy, Suno just wants you to say, give me a song and I’ll do the rest. Maybe that there’s a bit of danger there and it’s something that is new. I mean

Camilo Landau So the thing with the drum machine and the synthesizers and stuff, yes, there was a scare, right? People are like, oh, now people aren’t gonna hire drummers. This is gonna kill music. Oh, that’s so squared off. It doesn’t have any of the human swing and feel. That all turned out to be, I think, wrong, right, because people found this new tool and started using it for super cool, interesting stuff. They created whole new cultures and subcultures and whole new art forms, hip hop. You know, I mean, you know, people are like, oh, wait, you’re just sampling somebody else’s, you’re not even playing music, you just sampling it. No, they were creating new stuff on top of that, and they were using a new tool to create new art forms, right? They started connecting it with whole larger subcultures, but we’re not seeing any of that. Yeah.

Alexis Madrigal Is it different this time? That’s ultimately the thing. That is the question.

Robin Sloan It feels like it’s different this time. And there are reasons that it could be different. Again, this sense of ease, you know, there’s a difference, right, between a musician and a user. I feel like a lot of folks are stuck in the mode of being a user, there’s also a difference between a tool, even if it is a high-tech tool, a drum machine, a synthesizer, that you have in your garage or your basement versus a service that exists only on this far-off server controlled by this company and they change it whenever they want, however they want. You know let’s say you learn how to use it. You’ve become a virtuoso and then they announce the big upgrades to know 7.0 and you’re like wait Wait a minute. Where did my piano go? So these are I think these are real risks. It’s it is kind of new new territory

Alexis Madrigal I mean, we’ve talked a little bit about drum machines and the way that drum machines and that form of technology created new genres of music and new realms of culture. You know, there were other musical technologies through time that also had kind of this fascinating impact. You know one, recently there was a big show of Bernie Krause’s work at the Exploratorium, you know, recording these soundscapes all across there. The world and really showing the way that these ecologies as they degraded, their soundscapes also degraded. One thing that felt so important about that show, because it is fundamentally about climate change, is that music, even though it was essentially nature and a microphone, and the selection apparatus, a human making a decision there, it was really deeply political in its orientation. It wasn’t just sounds to make you feel good, it was sounds to you realize we’re destroying the earth and it’s species.

Robin Sloan And the drum machine, you know, this would begin as the most robotic, you know, tick, boom, tick boom, powers them probably the most political music. It’s been made certainly, certainly in the 20th century. Yeah. Incredible.

Alexis Madrigal Do you think there’s a pathway? Does AI music have an inherent politics? Is it just because it’s slop, it’s just conservative in some deep sense, or could it have some other politics?

Camilo Landau Well, I mean, one interesting thing is that you can always, in many of these platforms, you can insert your own lyrics or suggest lyrics. So there actually is quite a bit of right-wing country AI music. Like hard-right, like racist and stuff, right? Which is, I’m mean, problematic in many ways, but. But so typical. It’s just like expect.

Robin Sloan Well, and sort of, you imagine the strange liberation of the person saying, oh, I could never get anybody to sing these before, but the robot will sing whatever I want. I mean, that’s pretty grim.

Camilo Landau Yeah, but there’s also something liberating, I think, about being able to create any kinds of music, right? Like, you can create whatever you’re imagining, you should be able to create it. Okay, well, maybe it doesn’t work perfectly like that today. But that is the promise, is that, like, you have an idea and something in your imagination, and if you can get it out, if this can help you get it out, then that’s cool. It’s not exactly what’s happening, but there is a promise there. There is a promise there.

Alexis Madrigal Some liberatory little grain of potential in there that could take off. I mean, uh, maybe just to wrap us up back into sort of Pandora and what you, we know you don’t speak for the company, but just if you like, imagine what streaming services might look like in response to what is likely to be an increasing onslaught of this. It’s like half the music is probably the best it’s going to get. So, all right, five years from now. Now it’s. 99% of the music that is submitted to, you know, Pandora or Spotify or whatever is all AI slop. Like, what happens?

Camilo Landau Yeah, I mean, it’s interesting. So there’s a few things. Spotify and Warner and the major labels and UDO or Suno, one of the AI companies, there was a lawsuit, they came to an agreement, which essentially they acquired, they become the same company. And so it’s. Spotify purportedly is working on a feature for like basically AI remix of any song in there, in these certain catalogs that they control. So the idea would be like, I don’t know, you could be like okay, I want this Daft Punk song, but in the style of Lofi Beats or whatever. Yeah. You kind of just like create these sort of AI remixes on the fly using this particular catalog that’s licensed to do so. I mean, I kind of don’t f-

Robin Sloan think that that’s what people want. I mean, you know this well from the front lines of Pandora, the way we know that people listen to music. Every listen of every track is not an emotional resonance. It is this lull. Elevator music.

Alexis Madrigal Music for everything yeah

Robin Sloan Yeah, yeah. Yeah, exactly. You know, yeah this music it’s the it’s the final the final victory of music

Camilo Landau in this functional music space, I do think there is like, I kind of think that’s the sweet spot for this AI music, right? These lo-fi beats for studying or whatever. You don’t, you’re not listening to the songs. You have it on the background. You’re just trying to get a vibe. AI can do quite a convincing job of doing that, you know, a stream, like a constant stream of lo-Fi stuff. And it just, I pretty sure it was all. Just to drown out your roommate, Tony. I’m pretty sure was all the AI generated. Hey, if it can do a perfectly good job of this and people don’t really care, then fine, whatever. On the other hand, there’s also plenty of human created lo-fi beats that, you know, plenty like, you know, millions of hours or whatever, like, so why not prioritize that? And, you know, is, is the AI doing it better? No. Is it doing it competently? Sure. But is it doing a better job? So no. So then why use that? Why not prioritize the human stuff? And now the AI does that endlessly.

Robin Sloan Simulations of simulations of simulations

Alexis Madrigal It is a grim, grim vision of the world, so I like some of your other visions.

Robin Sloan In our conversation, I feel like we’ve come up with between three and five actually exciting paths and gestures, and it just so happens that we’re stuck with the very worst one at this moment, but that’s okay.

Camilo Landau Yeah, which is that the vast, vast like almost all majority of AI generated music is really just used for fraud to try to trick people into playing they’re just putting out massive volumes of music just in the off chance somebody might accidentally click on one or one of them might get on something and they can make a few cents.

Alexis Madrigal It’s like the worst arbitrage, like tokens into streams. And they’re trying to make like one fraction of one cent on each of those things. That is depressing. Um, well, um, Camilo, I feel like I have learned, uh, so much about the strange state of AI.

Camilo Landau I mean, I think there are a lot of interesting AI tools being used in music, in production, right? I mean there’s AI mastering AI, you know, like, okay, here’s a thing. I know what a finished song should sound like and what an unfinished song sounds like and I can help you make your unfinished song sound sparklier and louder, right. Great. I mean it’s not as good as hiring a real mastering engineer, but it’s, that’s a good tool for musicians to be able to have to make their music sound better. Great. Yeah, yeah. But there’s still somebody making music in that instance, right? There’s somebody assembling, somebody making decisions. A human making decisions, I think, is the important element.

Robin Sloan The core, right? It’s almost a filter. You can say like, did somebody have an opinion or a vision or one tiny idea? I’ll take it.

Alexis Madrigal There it is, lowest common denominator, we have found it. One tiny idea.

Camilo Landau I think people connect because they’re connecting with the artists, with the music, and if that’s not how this music is being made, then I’m not convinced it’ll work.

Alexis Madrigal I had like the worst possible scenario where it’s like, do-do-do, do-doodle-do. Five years later, we’re here. It’s like the most popular song in the world is AI, you know, and we’re like, ah. We’ll come back to Camilo on that.

Robin Sloan We got to trust and hope that listeners will hold the line. It’s actually heartening in a way that, you know, all the machinations are certainly at the platforms. We’d love to be able to serve music without authors that you didn’t have to pay for. But if listeners won’t take it, they can’t get around that. So, hey, let’s all stay real. Stay positive. Let’s stay all very persnickety in our listening.

Robin Sloan KQED’s Dream Machines is made by humans and hosted by me, Robin Sloan and Alexis Madrigal. Our series is produced by Anayansi Diaz-Cortez and Derek Lartaud. Sound design by Brendan Willard. Jen Chien is the executive producer and Ethan Toven-Lindsey, our editor-in-chief. Support for the production of Dream Machines comes from the Krishnan Shah family, Dorothy Marsh and other generous KQED members. Special thanks to Chris Egusa, Annie Fruit, Paul Lancour, Vivian Morales, Zaldy Serrano, Xtine Tiñoso, Hazel Tesoro, and Alex Tran. And of course, thank you to the Close All Tabs team for letting us visit their feed this month.