Switched on Pop - Is that new song you like AI? Here’s how you can tell

Episode Date: July 15, 2025

The robots have arrived, and they're making protest songs about boots on the ground. When an AI band called The Velvet Sundown fooled over a million Spotify listeners with their psychedelic folk anthems, it raised an unsettling question: have the machines gotten so good we can no longer hear the difference? Charlie puts Nate to the test with a game of "AI or Human?" featuring Wu-Tang deepfakes, phantom instruments, and songs that sound like Dire Straits and Tom Petty had a baby. Along the way, they uncover the five telltale signs that expose artificial music, from juvenile rhyming patterns to voices that shapeshift between tracks. But here's the terrifying part: just six months ago, AI music was unlistenable chaos. Now it's disturbingly competent. And it's only getting better. Songs Discussed The Velvet Sundown - "Dust on the Wind" Post Malone - "Chemical" Crosby, Stills, Nash & Young - "Ohio" The Velvet Sundown - "Freedom Song" Kansas - "Dust in the Wind" The Animals - "House of the Rising Sun" Creedence Clearwater Revival - "Have You Ever Seen the Rain?" The Velvet Sundown - "Where War Remains" Pink Floyd - "Wish You Were Here" The Velvet Sundown - "Ash and Velvet" Buffalo Springfield - "For What It's Worth" The Velvet Sundown - "For the Ones We Couldn't Keep" The Velvet Sundown - "Mirrors in the Smoke" Pink Floyd - "Breathe" The Velvet Sundown - "Rebel Shout" The Velvet Sundown - "Smoke in Silence" The Velvet Sundown - "Marching Shadows" The Velvet Sundown - "As the Silence Falls" The Velvet Sundown - "How Did This Go Wrong?" Hip Hop Intelligence - "Bar Fight" (AI Wu-Tang) Hip Hop Intelligence - "Party with Me" (AI Eminem) Temple of the Acid Fist Records - "Woman Gone Blues" (AI) "Echoes of Twilight" (AI student example) The Velvet Underground - "Sweet Jane" "Whispers of Chaos" (Charlie's AI generation) Mungo Jerry - "In the Summertime" Almost Vinyl - "Phil Wildo's Door to Door Dildos" (AI) Joey Two Legs - "I Shouldn't Have Done That" (hybrid) Bill Evans AI track (untitled, by Nobody in the Computer) Learn more about your ad choices. Visit podcastchoices.com/adchoices

Attention Spotify. Has arrived the new Good Girl Jasmine Absolute of Carolina Herrera, a fragrance intense with character curman and addictive. Imagine a jasmine emvolvent, toffee caramelized and tonka-tosted.

A combination that seduce from the first instant and a wayla. Good Girl Jasmine Absolute hypnotic irresistible. Discover it today and let you knowvelver for

their sensia. Conviette your passion in a new business with Shopify and batte records of the business with the form of the market

with a better conversion of the world. Has heard well. The Mereverion of the world.

The incredible system of Pago of Shopify facilita the website on your social and

in the social and in whatever that's music for your no

you do you do you make your business for a year

per Meper RECords. Welcome to Switchdown Pop. I'm songwriter Charlie Harding. And I'm musicologist Nate Sloan. I've got some bad news. Oh boy. The robots have arrived.

An artificial band called The Velvet Sundown has spoofed millions of listeners into streaming music made largely by machines. And it raises a question. Has generative AI music improved so much in 2025 that we can no longer hear the difference? Let's take a listen to find out. Here's the Velvet Sundown's Dust on the wind. Dust on the wind Boots on the ground Smoke in the sky

No peace found Rivers run red The drums roll slow Tell me brother Where do we go It's pretty good I like the drum sound actually Oh man

The singularity is here. And it sounds a lot like a forgotten Credence Clearwater B-Side from the 1970s, which I did not see coming. Okay. Before we get deeper into the music, let's get to know this band. Band in quotation marks. Yeah, exactly. Perinthes question mark. I don't know what this is. The Velvet Sundown, obviously drawing a hazy resemblance to the Velvet Underground, is a band made up of four fictional AI personas, Gabe Farrow, Lenny West, Milo Raines, Orion Del Marr. They're making vaguely psychedelic 60s-70s protest anthems. Their first album, however, debuted June 5th, 2025. It's called Floating on Echoes. And just a few weeks later, on June 26th, they released Dust and Silence, their next album.

And this band has got some serious attention. Over one million monthly listeners, Nate, Somehow they are a verified artist on Spotify, and their music is being pushed into people's Discover Weekly Playlists. Their most popular track that we just heard, Dust on the Wind, hit Spotify's viral 50 chart in the UK, Sweden, and Norway. And the provenance of this band has been a bit confusing. They've only been around for roughly a month. When they first came out, speculation about this band started popping up all over Reddit, people questioning, is this real, is it not real? A guy named Andrew Frailin, claiming to be the band's manager, got in touch with Rolling Stone and said that this was a giant art hoax, which the band later refuted saying they had no affiliation

with this fake fraudster of a manager. But after spoofing the internet for an entire month, the band came forward on July 5th and confirmed that their music is made with the support of AI. Here is how they describe this project. The Velvet Sundown is a synthetic music project guided by human creative direction and composed voice and visualized with the support of artificial intelligence. This isn't a trick. It's a mirror. an ongoing artistic provocation designed to challenge the boundaries of authorship, identity, and the future of music itself in the age of AI.

That bio, written by ChatGPT, pretty clearly, and performed by an artificial intelligence voice recreation of the late Bert Reynolds. I used an app called 11 Labs to make that, and backing music from Udio, a generative music AI model that I used to try to emulate the sound of the Velvet Sundown. How do you feel about this project so far? The answer to that question is as complex as the narrative that you just gave us. I feel a little uncomfortable that this band initially portrayed themselves as real people and created this web of lies. Is that a fair way to describe it? Sure. Now that they're not hiding anything, I'm a little more comfortable with it.

And I think more of my questions are, how does this work? Who gets paid? Is this responsible? Why do people gravitate towards this music? Do they know it's AI? Do they care? Those are the kind of questions I've raised now that they've stopped pretending to be something they're not.

And I think part of this project is saying that they want exactly that to be a reaction, that this is some kind of art project, which feels like an after-the-fact explanation, if you ask me. And, you know, there are a lot of people who feel like the ethics of all of this kind of work is just completely insupportable, immoral, wrong, no good. You know, these AI models were trained

on copyrighted material, the artist, songwriters, producers, performers of that work never compensated. The two biggest generative AI music models, Suno and Oudio, are being sued by the music labels. And there's no shortage of discourse on what the heck. This is just a waste of vital energy resources and a warming planet. Why are giant data centers being used to make this eh music? Hackney derivative. Yeah. Nonsense.

I think it's a completely fair assessment to say that this stuff is derivative. It's no good. I don't want anything to do with it. I mean, if people are angry about AI music, they have all good reason to. It's insulting to musicians and to listeners when the CEO of Suno goes on record saying that musicians don't even really enjoy making music. It's not really enjoyable to make music now. It takes a lot of time. It takes a lot of practice.

You need to get really good at an instrument or really good at a piece of production software. I think the majority of people don't enjoy the majority of the time they spend making music. If people are mad about AI music and don't even want to hear it, get it. And yet, here we are. This stuff is very much happening. It's not going away. And I feel like if we are going to be able to differentiate human-made music, we need to be able to audibly identify what generative AI music sounds like today. Now I'm very intrigued, Charlie. Let's enter the Matrix together. Great. Let's go back to the first verse of Dust on the wind. You know, peace found.

Rivers run red. The drums roll slow. Tell me, brother. You know, from first listen, I got to say, kind of impressed. It is incredible, astonishing, jaw-dropping. I mean, let us not bypass this too quickly. You can create a song from scratch in virtual seconds with just a few. text prompts using these platforms you mentioned, like Suno and UDO.

It is bewildering that we can do that in 2025. And just a few months ago, like, this technology was not so sophisticated, and now it's here, it's wild. Months ago, these songs were chaos. They would trail off into nowhere. They would have a strange double bridge. The choruses would sound different from one another. The prosody was terrible.

Today, these songs meet our expectations of song structure. They have good prosody. They have the right amount of repetition. I mean, it's pretty amazing. You have melodies that follow chord progressions. You have chords that follow a consistent pattern. Rhythm sections that follow along with the bass and kick playing in time with each other. The instruments match the genres.

The backing track gives space to the vocalist. And when the vocalist drops out, the instruments come back in and a call and response. I mean, it's almost as if people are playing this music. You said almost, Charlie. Almost. What are the clues that might tell us? listener, actually, this isn't people playing together. This is digitally synthesized generative AI. I have five musical tells that I've identified. And the first has got to begin with the lyrics.

I think this is the place that most people can hear that something is just a little bit off. What we hear is a constant barrage of single syllable perfect end rhymes. Dust on the wind, boots on the ground. Smoke in the sky, no peace. found. Rivers run red. The drums roll slow. Tell me, brother, where do we go? Ah. It's juvenile rhyming. No multisyllabic rhymes. No. No internal rhymes. No. No slant rhymes, even. They're all perfect rhymes. Interesting. Right. I feel like I always use Post Malone's chemical in my coursework as a great example of all of those things that you just mentioned. A very lighthearted, wonderful pop song that displays a great amount of sophistication in its idea of rhymes. You know, I think with rhyming,

what we're trying to do is write songs that are memorable. Rhymes help stay in our memory. But when they're overly obvious, sometimes they become cloying, annoying, boring as a slant rhyme. I'm failing. Thank you. Thank you. Thank you. But check out what he does. Outside of the party, smoking in the car with you, seven nation army fighting at the bar with you. each line in the chorus with a self-hyme, the word you. It's not as strong and obvious as a slant rhyme. Maybe it's too obvious. And instead, the whole phrase is basically one giant set of rhymes outside of the party,

seven nation army, smoking in the car, fighting at the bar. So he's burying his rhymes internally in the stanza. And then when he gets to the title line, he bends his vowels. He uses cursive singing in order to sort of force a rhyme. Let it go and... Camacool. I can't let go. Or his attempt at getting a rhyme with, Tell me what I got to do.

Because I can't let goo. It's chemical cool. Yeah. You sound like a malfunctioning robot right now. Right, right. That's kind of where the humanity of good lyric writing is, is that the things almost match up, but they just don't match up in a really nice way. They are just unpredictable enough. You don't get that with this AI music. Okay, so the rhyme scheme is suspiciously perfect and regular. Yes, and they're not just regular and obvious and rhymes.

The lyrics are mostly meaningless. Like, what does the chorus of Dust on the Wind even mean? Well, I would say it's an anti-war protest song, but in a very vague way, it doesn't really take a sigh. It doesn't really describe who you're protesting against even. No. But generally, I wouldn't say it's meaningless. I would say it's just kind of anodyne in its lyrical content. I would call this a style of writing where you are trying to target rhymes, where the rhyme is more important than the meaning.

Raise your hand, don't look away. That's a fine line, right? There's an image to that at a protest. Sing out loud. Make them pay. Wait, hold on a second. Why are we making our enemies pay when the next line is March for Peace, not for Pride, let that flag turn with the tide?

There's internal contradictions to like, is this peaceful protest for a larger calling, or are you going to make them pay? And what is the deal with let the flag turn with the tide? Yeah, I got nothing for that one. What's going on with this mixed ocean metaphor? Yeah, yeah. It doesn't actually mean anything. I mean, if you compare this to a classic protest anthem like Crosby Stills, Nash and Young's, Ohio, you get a much more human song.

Right, we have Nixon's men gunning down protesters at Kent State, an actual historical event, and CSNY making commentary about that moment. There's such a specific narrative, there's such an urgency to the song, which they wrote very quickly after the murders at Kent. state and released very quickly as well. Dust on the wind, by contrast, feels like it has no urgency. It's a protest that feels like, you know, take your time. Whenever you get around to it, maybe, you know, stop the war, but, you know, no rush. I think probably even cheapens meaningful songs like Ohio. They have another song called Freedom Song, which shows that the lyrical ideas quickly run out. Sing it loud. Don't turn away. Pieces more than words we say. Raise the light. Disolve the fear. Freedom's song is ringing clear. Perfect example of saying nothing,

perfect and rhymes that are more important than the actual content of the music. So lyrics are a really obvious giveaway to either amateurish writing or definitely AI writing at this point. Okay, so that's our first tell. What about our second? Second tell is obvious borrowing. Now, humans get away with this as well, but oftentimes you'll hear an AI generated song and you'll say, that feels really familiar. I mean, just take the title, Dust on the Wind, Ring a Bell?

Yeah, Dust in the Wind by Kansas. Yeah, yeah, yeah. Strangely, the songs don't really sound alike. No. Here's Dust on the Wind, or fake song. Well, it's a real song. Oh, my God, we're already in this ontological nomad land. Here's Dust in the Wind, a real song, pure, organic, fully human.

No, that's like some like post-fleetwood Mac acoustic folk-y stuff. But dust on the wind does bear some resemblance, I think, to the animals' house of the rising sun. Here's us on the wind. You put that on an electric guitar in the hands of the animals. I definitely hear the similarity. Also, kind of generic, A minor guitar strumming. What about a comparison to the band I mentioned earlier, Creedence Clearwater Revival?

I hear the vocal style of their lead singer John Fogarty and the acoustic guitar strumming and the drum sound and the subtle electric organ of a CCR tune. like, have you ever seen the rain? If you take the progression from House of the Rising Sun and map it with the texture and vocal tone of CCR, I feel like you're starting to get the Velvet Sundown. I mean, I feel bad comparing these songs

because, yes, there is a little John Fogarty twang. It's got that rasp, but we didn't even mention this, but like the AI vocal sounds a little weird. Not that it's pitchy. It's just like kind of inconsistent, dynamically. Sometimes it's quiet. Sometimes it goes into this like, stronger chess voice, head voice, but kind of in the same note and in the same range,

gets lost a little bit. Sorry, John Fogarty. But I think there's other examples in the Velvet Sundown's repertoire that might even better demonstrate the ways in which the AI is obviously borrowing from its training data. Take, for example, the very beginning of their song where war remains. Boom booboo. Is that the clue here? That little opening baseline?

Boom booboo. The band, the weight, maybe. Nope. It's the opening guitar line to Pink Floyd's Wish You Were Here. Your classic rock knowledge is... I'm very disappointed. Check it out. Same turnaround.

Okay. Okay. Let's try some more. How about their song, Ash and Velvet? Does it remind you of anybody? Not the melody. It's the timbre. This is like if somebody got a time machine and made a new old Peter Frampton song.

That talk box sound, pure Peter Frampton. I'm going to try one more on you. See if I can. convince you, how about the song for the ones we couldn't keep? Literally just that first note. Wait, wait, I do recognize this. Okay, you're prompting this AI to make a song based off of protest music from the 60s and 70s. There's no more prime example than Buffalo Springfields for what it's worth.

Oh, yeah, wow. Right, just that very first note. Let me hear the sundown again. Yeah, it's not a harmonic, but I do. I definitely hear the comparison. All right. Last one. Mirrors in the smoke sounds like... You got a smile on your face. You got this one? We're back to Floyd, right? Yeah, yeah. Dark side of the moon. Time. Breathe. One of those.

Breathe. They all sound the same, but it's breathe. Yep. Oscillating between the minor one and the major four. That's wild. Okay. So we've got two signifiers. One, amateurish lyrics that overly rhyme don't mean anything. Two, obvious borrowing, from the training data. And three, now, I hear a lot of inconsistency. Hmm. Now, if you're going to be a band, yes.

With a lead singer, you need to have some kind of identity. Yes. I don't think it works in your favor. If every single song, it sounds like there's a different singer. The opening song of the Velvet Sundowns Floating on Echoes is called Rebel Shout.

And it kind of sounds like a Robert Plant sound alike. Boots on the ground again. Lot of recycled. material, yeah. Go to the next song, smoke and silence, it's a new voice. We went from a tenor to a baritone, and when you go to the next song, Marching Shadows, it's like a blues singer has entered the room. Much more nasal. Much pure tone. These are different singers, and even the guitars sound kind of different. There's no

consistency of identity. So if you see a band that has an album full of different singers, you're either listening to, now that's what I call music, volume 1970s, or maybe an AI band. Okay, so that's three things. Okay. The next is song form. Now, song form used to be a greater tell. Like, chorus two wouldn't sound like chorus one. But I think that you can still find ways in which an AI is showing its cards when you hear something like a weird trailing off ending that goes into nowhere, which happens on Dust on the Wind. We come out of this guitar solo. It's like we're going into new verse.

Wait, what just happened there? No way. It's almost if the AI was like, well, we just use the word end. So we got to call things quit. That is so funny. It happens all over the place. Like it happens on their song as the silence falls. Which kind of sounds like if Dyer Straits and Tom Petty had a baby.

Again, long guitar solo, followed by. There she speak through a guitar line. Weird. Also another totally new voice that we've never heard before. That's so unsettling to just have it end abruptly like that. Yeah, these abrupt endings, you might also get unexpected interludes or double intros. In fact, dust on the wind has two intros. So songform can be a tell. Yeah. I think the next level of AI giveaway can be a little bit harder to hear. But if we tune in our ears, we can hear

all kinds of hazy artifacts sounds that are kind of just frankly indistinguishable ghosts of instruments that never existed. Like, listen to their instrumental interlude that sounds way more like a contemporary crumbin joint than it does like a 1970s protest anthem. Listen in between the notes as they trail off. In the reverb, there's like these other like almost like little horns and things that come in. The guitar will go, duh, and then there'll be those of strange sound that is not clear. We hear the same thing on their song floating on echoes. I want you to tell me what instruments are we actually hearing.

Obviously drums, guitar, bass. But what is the background layer? It almost sounds like a conch shell or something. There's just this in the background. That's really weird. Things are blurry, like on the song, How Did This Go Wrong? Which is a song that asks a question, which I think we know the answer to. What am I hearing? What went wrong? There's a very bizarre phantom sound in there.

It sounds like a string orchestra, but they're playing with the wood of their bows rather than the string or something. That is very peculiar. Yeah, so something is obviously wrong here. You can hear these sort of blurry non-human instruments. They're often in a background layer when things get a little bit more fuller in the mix. And so we have here funky lyrics that overly rhyme in me nothing. We've got obvious barrow. We've got inconsistency from song to song, messed up song form, and these hazy, smoky artifacts.

I think all of these reasons, though, are also why this band worked. Because if you're trying to make a 70s psychedelic protest band with four members, that might be the perfect test case. Music that is a little bit lo-fi in an era where we might have heard more perfect rhymes, where you have multiple singers in a band. There's a little bit of plausible deniability in this music. And, you know, this is also maybe just a little bit before the era of the launch of a lot of independent music. Like, it's feasible that there was an archive of vinyl that was sitting in someone's basement that someone pulled out. and digitized just last month and was like, check out this amazing unreleased music that you never heard.

But that's not what the band was advertising at first. They were saying they were a new band who were reaching back into the past. Yeah, yeah, yeah, right. And yet, I totally agree with all your points here. It is a canny kind of way to slip in these AI songs into an existing playlist. Because like you said, it feels kind of plausible

in a way that another genre or style might not. Well, here's the issue though. I think the word plausible is doing a lot here because that's what we're trying to figure out. Is it plausible that this was made by a human? And I actually was tricked yesterday as I was doing my listening all the way through their two albums. I don't recommend it. I was fed another song where I was like, oh, this is the worst AI song I've ever heard. I mean, is this not a song with sort of faded in the world one day?

I got caught up in a traffic jam talking about all the bombs that were dropping. I mean, is this not a song with sort of faded? lo-fi sounds. They are clearly borrowing from Sweet Jane by the Velvet Underground using really noxious lyrics about war, how we're bombing someone from another land, and then I just fade away. It all felt too obvious, clearly AI. Yet when I went back through all of the Velvet Sundowns albums today being like, what was that song that I heard? I realized that Spotify had auto played me the next song by a real person. And this song, Stubborn Faith, was human, and I thought I had been duped by an AI. We should never discount actual human's

ability to make music that sounds like it was created by AI. You don't need a computer to write an incredibly forgettable, oceous, predictable song. We've been doing that arguably for centuries. Yes, I've written plenty myself. Since the dawn of the popular music industry. Yeah. Okay, so I was duped, and what I want to do is to figure out whether or not I can dupe you and all of our listeners in a little game we're going to call a I. When we come back, we're going to find out whether or not the music you're hearing is written by an AI or by a human. Yeah, we might workshop that title in the break, but I'll see you on the other side, Charlie. Maria, you have a podcast now and you need to start acting like it.

What's the first step as a podcaster? Well, you have to ask lots of questions. I'm Maria Sharpova, and I'm hosting a new podcast called Pretty Tough. Every week, I'm sitting down with trailblazing women at the top of their game to discuss ambition, work ethic, and the ups and downs that come on the path to achieving greatness. I have a few pretty tough questions for you. Okay. Ready? Ready.

Do not sugarcoat something for me. No, no. We'll dive into their stories and get valuable insights from top executives, actors, entrepreneurs, and other individuals who have inspired me so much in my own journey. Pretty tough is your front row seat to the women who have demonstrated the power in being unapologetic in their pursuits. I hope you'll join us. New episodes drop Wednesdays on YouTube or in your favorite podcast app. Immigration may be Donald Trump's signature issue. President Trump is now targeting predominantly democratic cities for ice raids and deportations. Dozens of protesters clashing with

immigration and customs enforcement agents in Minneapolis Tuesday. We will begin the process of returning millions and millions of criminal aliens back to the places from which they came. But what we want to do in this space. It's talk about America and politics beyond the current president. So what do most Americans think about deportation and border security, period? I think that Americans are definitely against the kind of violent displays that we've seen in the street from ICE. When it comes to the question of deportation, the answer is more complicated. My sense is that people want border at the border. They don't like the idea of having no idea who's coming into the United States at any given time.

the view on immigration from the bottom up instead of the top down. That's this week on America Actually every Saturday in your audio and video feeds. AI? Still got it. Couldn't come up with anything better. Okay.

Let's do it. Here's the roles of the game. Maybe it's a little bit of a touring test. Do you have the ears to tell whether or not something is human or not? I'm going to play six songs. Are you ready? Ready as I'll ever be, Chuck.

Okay. The first song we have is called Bar Fight. It's the return of the king. Bring the heat. burned in the ring. This is murderous. Yeah, you've heard of us. Still doing this thing one decade after the next. Still packing the tech show some respect. We collect in a debt. Get torn with the sword if you ain't protected that neck before. Cook up them raw metaphors. You can smell it at the door. Swipe at your jaw. You can fight for your life. Okay, I'm going to use the criteria you just gave us in the

first half of the episode here, Charlie. So if I get it wrong, it's your fault. Do I hear simple rhymes? No. I hear slant rhymes. I hear multisyllabic rhymes. I hear internal rhymes. So that's kind of suggest it's not AI. I don't think I heard that artifacting. I was actually really impressed by the opening, which was kind of weird and chaotic in this way that felt very human to me. I'm going to say this is not AI. It's AI. God damn it. This is music by the YouTube channel Hip Hop Intelligence whose Patreon says that they are on a journey to keep hip hop alive, assisted by the latest technology as it continues to evolve. And on that track, you are hearing AI models of the Wu-Tang Clan, hip-hop intelligence has made new 2025 albums of many artists, including Eminem.

Here is Party With Me. Back when Vin Diesel was March Sinclair, this is how we used to make the party scare. We used to take shots with no other intention, but to take off to another dimension. Okay, that one I would have gotten. There was artifacting at the beginning. I feel like Eminem has such a distinctive style. I could tell immediately it wasn't really him. Some of the rhymes fell flat.

The Wu-Tang one was harder because, you know, they were making references to protect your neck. I was like, oh, this is someone paying tribute to Wu-Tang. Yeah. Interesting stuff. So, not doing great so far. Let's try round two. This is Woman Gone Blues. Okay.

I heard, now I'm like, not sure if I'm fooling myself. I heard some kind of phantom sounds in there. This one, you know, without vocals or the full song form, it's harder to apply some of these criteria. So I'm just going to go with my gut and say, Also based on the weird title that it's AI. Yeah, yeah, it's AI. This is by a YouTube channel called Temple of the Acid Fist Records, who have been basically every month,

we've been releasing these archives of North African Jam 70s music. Again, an era of maybe it's easier to borrow from because it's lo-fi quality. It also reminds me of an actual project called Awesome Tapes from Africa, which is a blog started by a Fulbright scholar who, while traveling West Africa brought all these unpublished tapes back to the United States, then partnered with artists to release things that hadn't previously been released and to share in the profits. So I was like, maybe there is this cool acid fist records company that is making all these interesting recordings. I honestly was duped when I first heard these because I just thought maybe it was a bad MP3.

Okay, so you've gotten one out of six so far. Let's try the next song. Echoes of Twilight. Memory stitched in threads of gold. A whisper hard the silence. This is the easiest one yet. That's AI. Yeah, yeah, yeah.

That's a weird Paul Simon takeoff. The guitar strumming is all out of whack. Yep. His voice gets distorted at moments. The vocal harmonies when they come in sound nothing like Simon and Garfunkel. The lyrics are weird. Yeah, this one I feel pretty confident saying it's AI. Yeah, no, you're right.

And frankly, this is an older one. This was a song that one of my students had brought to me in class. I do an assignment where students have to use one of these AI music generation tools to create a song. They write a prompt. They get a song out of it. They then have to grade the quality of that song and write a better one. And this was a prompt that was simply 1970s folk song. And it is a clear rip-off of Paul Simon's voice.

This is what we call overfitting in the world of AI, where the output, of your prompt is too close to the original training data. I went and tried to make some other songs just today, and the same thing happened to me. Here's a song that I generated called Whispers of Chaos. Was the prompt that you gave it to make music that sounded incredibly inchoate and not together? No, I actually was trying to make songs

that were in the same style as The Velvet Sundown, but instead it gave me kind of a regurgitated version of in the summertime by Mungo Jerry. More like Bungle Jerry. Okay, okay, okay. That just sounds like slop to me. That's just hot trash. Okay.

All right. You're getting better at this. I believe I'm two for six now. Let's go to number four, please. Okay, I've got a song for you by Almost a Vinyl. My name is Phil Wildo. And I was wondering if you'd like to browse my selection of Dildo. All shapes and sizes to suit your preferences.

It's breaking my brain. And it's definitely AI. Yeah. A lot of AI novelty songs using crude humor in traditional song forms. But this one is fascinating to me because this one came from one of our longest time listeners. I got a message from our friend Katie Richmond. She'd been listening to our episode on Sinatra, which, featured Seth McFarlane, of course, the creator family guy, amongst many other animated hits,

and he had this album of Snotra covers, and inspired by our episode, she went and listened to his album, Lush Life. And then, when the album ended, Spotify played music from this album, Fools Golden Oldies, Volume 4, which includes Phil Willows, Door to Door, Dildos. And I'm like, how on earth did this happen? And the only explanation I have is Spotify's trying to figure out, okay, you're listening to Sinatra by Seth McFarlane, who's known for sometimes crude humor. Yeah. There is no song that follows that combination except for this AI generated slock. It raises a question which has come to me throughout this episode, which is like, how much human

intervention are there in these AI creations? Like, did a human have to tell the AI like, hey, call this guy Phil Wildo? Right, right. Would AI make up a person like that? Like, yeah. We probably don't have the answer to these things, but it is something I'm curious. curious about. So we can table that for another day. There's all kinds of instances where AI is used, where an actual vocal that is then replaced with an AI vocal, that's an issue. Maybe in the

Wu Ting sound alike, perhaps the person actually had written the lyrics or had them generated and then improved the lyrics. Totally. All right. Right now, you're three out of four. Let's hear the next one. This is Joey Two Legs. I shouldn't have done that. I shouldn't have done that. I'm sorry. I'm sorry. I'm sorry. What's your feeling? This one is a little trickier, but I'm going to go AI. Either that, or it's maybe like a Swedish jazz band singing in English for the first time. But there's something a little awkward about it, lyrically, I think. The trombone solo sounds a little too perfect to me.

Yeah, I'm going to go AI. And Joey Two Legs, that's another one of these bizarre AI concatenations. Joey two legs, according to their post on Reddit, so I'm sorry, I didn't do nearly enough fact-checking on this one, but this artist claims that they wrote the lyrics, used Suno to make the music because they didn't feel that they had the chops they needed to actually write the songs, did extensive editing using Suno and the things that it came up with, and then hired real musicians to play the songs. Really? Wow. Yeah. So kind of a hybrid. Interesting. So I'll give that one to you. I've got one more for you. and I'm not going to even give you the title this time.

Not a lot of information. You don't get any lyrics. A little bit of noise, but could that be like a deliberate recording technique for this? Oh, yeah, by the way, some of those Velvet Sundown recordings sound like there's tap hiss in the background. I don't know if that was added by whoever's controlling it or what,

but you can't always tell by that. It's not always very precise, but then some of that might be deliberate by the musicians. This is a tough one. I think it's quite lovely, actually. I mean, I'm just going to kind of psychologize you and say that it's AI because you would get a sick pleasure by making every one of them AI. So yeah, that's what I'm resorting to.

This is my favorite one. This is Bill Evans and the AI. This is a YouTube channel called Nobody in the Computer that does really fabulous sort of experimentation with what AI can do that is maybe truly creative. And what they've done here is they've taken recordings of Bill Evans and trained their own data set just on Bill Evans' recordings. It then outputs sort of like le generative music that has the weird artifacts, the haziness, but then converts it into midi data so that you can then hear it played on a more properly sampled piano. And so there's a lot of human intervention

going into it. Moreover, it's kind of just leaning into the way that the current state of AI will often create weird hallucinations. Or maybe this, you know, obviously, you know, Honestly, this song is a year old, and that year ago AI hallucinated a lot. And there's something special in the mistakes that only the AI would make, just in the way that we sometimes create beautiful composition ideas from our own mistakes. There's something nice in this very idiosyncratic approach to using these tools. So I really like nobody on the computer, I think, really worth watching the videos of how they do this. So in typical Charlie fashion, you have sort of pulled the rug out under me.

and giving me six AI songs of various styles and various methods of creation. And it leaves me feeling like some of those tells you described in the first half of the episode, we won't be able to use those pretty soon. Yeah. At the rate of this technology evolving, I anticipate there will be zero daylight

between the original and the copy. And that raises some interesting philosophical questions, doesn't it? It really does. Yeah, I think that this episode will sound dated very quickly. Yeah. Where I believe that these tools will get better at making human-like rhymes, not just perfect n-rimes that are single syllable. I think that the audio quality will definitely improve.

I think it's going to get easier to copy these things. Maybe that's interesting. But, you know, doing this episode, I feel like my perspective on what this music means has really shifted. And I'm not sure it's for the better. I really am starting to believe that AI music might split audiences in two. That for most people, music in the background is fine. Yeah.

And that's okay. I got no issue with you if that's what you're doing. And then there's the connoisseurs, you know, the listeners to switch down pop. I think about it this way. Music is going to become Doritos. Okay. Cool Ranch or... No, we are talking the original orange Dorito made out of Yellow 6,

yellow 5 and red 40 artificial dyes, dies that according to a new Texas law called the Make Texas Healthy Again law requires labeling that says that this is not recommended for human consumption. You're suggesting we have a similar label for AI generated songs? I think that's not a bad idea,

but no, what I'm suggesting is that I think most people are going to keep eating Doritos. Ah, okay. Even when they could be going to Whole Foods and buying the late July nacho cheese chip made out of non-GMO and organic yellow corn. If you look at it, the tortilla chip business

is a $37 billion industry, but only 9% of those purchases are organic. And, you know, the U.S. is leaning towards healthier options according to the I-Mark group who publishes data on this stuff. But within the U.S., Doritos is the biggest seller. 1.14 billion units sold in the past year. That's a head of Tostitos who only sold 332 million units of tortilla chips.

What's up with this? What am I talking about here? I'm not sure, but now I'm really hungry. So, yeah, where's this going and then let's break for lunch? The Doritos theory of music? I think that for most people, delicious, snackable food, which is made to be an approximation of real food, is going to be perfectly good. And a lot of people are going to enjoy it just fine.

I mean, frankly, the entire music playlisting world has shifted away from genre as the predominant marker of music and towards context. Are you working out? Are you sleeping? Are you chilling? You got a dinner party jazz playlist that it doesn't matter what's happening in the background. Lean back listening. Mindless snacking. A lot of people want this stuff. And I think that this generative music is going to find a very large audience. I don't think that this is the beginning. What does this mean for musicians, though? I think that they're going to have to do everything they can to prove their humanity. Being consistent, not having like five different singers. They need to have a clear identity. Their images on Instagram can't look like. like AI slop like the Velvet Sundowns. They have to have meaning in their work that connects with an audience fundamentally. They're going to have to be human. A lot of people have just come out and said, oh, this music sucks, this Velvet Sundown, AI generated music. Rick Biotto, my frenemy, you know, made a whole video about it.

And everyone in the comments is like, oh, this is bullshit. This sucks. This sucks. And I'm like, it doesn't suck. It's very listenable. It works. Like, don't dismiss it because it's AI. You have to take it seriously. Just like you have to take Dorito seriously.

you said, it's this, you know, multi-billion dollar industry or whatever. Try, before we go, I know we've been here for a while. Can I offer one other litmus test for determining whether something's AI or not? Definitely. Try to make it yourself. I went to Suno. I thought, can I recreate the Velvet Sundown? I gave it the following prompt.

Anti-war lyrics, male tenor vocals with slight southern twang, prominent acoustic guitar, 1970s, medium-tempo rock groove. It gave me Shadows of the Cannons. Dust in the air, it don't settle the boots on the ground. Blood stains the floor. What's with all the boots on the ground? Now, I got to be honest, I don't think this is the Velvet Sundown. Like, this is a little bit more like if the Velvet Sundown transitioned to country in the 1980s.

Musically, it's not analogous. but lyrically, it's shockingly similar. Yes. I tried to dial it in a little more. I said, make this acoustic guitar driven, give it a slow rock groove. I added in the qualifier a 70 swamp rock. And that gave us muddy hands of war.

Boots sinking deep in the mire of lies. Dust choking out the blue of the sky. Blood runs thicker than the river's down. here it goes Oh, it's bad. Okay, okay, turn it up. This is not Doritos. This is inedible.

Okay, more boots, more dust, more blood. Clearly, Suno has a profile. I feel very confident that Velvet Sundown use Suno to generate their song based on those lyrical similarities. But I was not able to totally replicate the sound. I would probably have to waste a lot more water

and keep trying different combinations, which I'm frankly not willing to do. But presumably you could get there eventually, or maybe there's some other secret sauce that we're not aware of. But, man, hearing those lyrics, I was like, oh, okay, there they are. Boots on the ground, dust in the air. There it is. This is much worse snack food.

Ultimately, this is not Dorito music, because Doritos are way better than this music. Let's be real. This stuff is pretty janky. It's bad. Like, it's not tasty. It's like those knockoff Doritos you get in the party mix with the pretzels and the cheeseits. It's like, all.

little, you know, air sets. Yes, exactly. So, you know, maybe this has given me a lot more hope in this current moment for the musicians. We do this to form connection so I can hang out with you, so our listeners can hang out. That's why we listen to music. Yeah. And this background junk that might fill up some of our life, I'm probably just going to tune it out, unless it fools me. Okay, Charles, before we get to the credits, I know everyone wants to hear them desperately,

but I have to share. We've been discussing bad lyrics today. And arguably we were discussing bad lyrics the previous week, too, when we were talking about Benson Boone's Moon v. Ice Cream. So clearly this is something on our minds. And I actually want to do a whole episode with you, Charlie, breaking down the worst pop lyrics of all time. In order to do this, we're going to need listeners help.

Yeah. So what I propose is go to our substack. You can find it in the show notes. And you can also find it on our website. switch on pop.com. We're going to start a chat that will be dedicated to probing the worst lyrics of all time. I think we should keep them human generated. Yeah. Rather than, you know, lumping in the fill will-dose of the world. And hopefully in a week or two we'll have a collection of the most offensive lyrics of all time. And we can go through them and try and understand

what makes them so bad. Great. And I'm enormously excited about it. So this is so fun. Substack, lyrics, chat, post your worst offenders there, folks. Let's do it. Switched on Pop is produced by Rana Cruz, edited by Art Chung, engineered by Brandon McFarlane, illustrations by Iris, theme music by Zach Tenario and Jossi Adams of Arc Iris. We had additional production this week from Charlotte Tang. We're a member of the Vox Media Podcast Network and production of Vulture. It's part of New York Mag.

You can subscribe at nymag.com slash pod. Find us on social media at Switched on Pop. Again, check out the substack, weekly newsletter, show notes, or switchedonpop. We'll be back next week. We're flipping the energy around. Talk about the best theme songs in podcasting. So if you have any good podcast theme songs, also let us know. We'll see you next week.

And until then, thanks for listening.