Plain English with Derek Thompson - How AI Could Change the Future of Music

Episode Date: March 4, 2025

Artificial intelligence tools for musicians are getting eerily good, very fast. Their work can be maddening, funny, ethically dubious, and downright fascinating all at the same time. TV and podcast composer Mark Henry Phillips joins to describe his experience working with them. We talk about the job of modern music composition; why he's worried AI might eventually do much of his current job; the morass of AI copyright law; and the ethics of creative ownership. But above all, Mark gets my brain whirring about the nature of creativity—how great new ideas, like songs, come to be in the first place. The line between stealing and inspiration in artistic history has always been blurry. Picasso famously said: “Good artists copy, great artists steal.” And that is not just a memorable quote. Many of my favorite musicians were famous borrowers, to put it lightly. Some of Led Zeppelin's most famous songs—such as "Whole Lotta Love"—were such obvious lifts that, after years of court cases, the band agreed to add the plaintiff to the song credits. But analogies to music and art history also fall short to capture the weirdness of this moment. Neither Picasso nor Jimmy Page had access to an external technology whose deliberate function was to slurp up musical elements from millions of songs, store their essence in silicon memory, and serve them up in a kind of synthetic stir fry on an order-by-order basis. Musicians have been writing music with partners for decades, even centuries. What happens to music when that partner is a machine: Will it open up new horizons in songwriting and composition? Or in a sad way, will super-intelligence make the future of music more average than ever? Links: WNYC: "How AI and Algorithms Are Transforming Music"  "In February's Cruel Light (Goodbye Luka)" Full AI song "L.A. Luka (I Wanna Puke-uh)" Full AI song  P.S. Derek wrote a new book! It’s called 'Abundance,' and it’s about an optimistic vision for politics, science, and technology that gets America building again. Buy it here: https://www.simonandschuster.com/books/Abundance/Ezra-Klein/9781668023488 Plus: If you live in Seattle, Atlanta, or the Raleigh-Durham-Chapel Hill area, Derek is coming your way in March! See him live at book events in your city. Tickets here: https://www.simonandschuster.com/p/abundance-tour Learn more about your ad choices. Visit podcastchoices.com/adchoices

Hey, it's Bill Simmons letting you know that we are covering the White Lotus on the Prestige TV podcast and the Ringer TV YouTube channel every Sunday night this season with Mallory Rubin and Joanna Robinson. Also, on Wednesdays, Rob Mahoney and I will be sort of diving deep into theories and listener questions. So you can watch that on the Ringer YouTube channel and also on the Spotify app. Subscribe to the Prestige podcast feed. Subscribe to the Ringer TV YouTube channel. And don't forget, you can also watch these podcasts on Spotify. Spotify. White Lotus. Let's go. Today, AI, music, and the future of creativity.

In the last few years, several generative AI platforms for music have caught my attention. One of them is Suno, which allows you to request a song by typing in a simple prompt. You specify the style, the lyrics, the mood that you want, and the AI will interpret those inputs and produce a musical composition. So let's start with an extremely stupid example. I have several friends from Texas who were distraught by the trade that sent basketball phenom,

Luca Donchich, to the Los Angeles Lakers. Now, if you're listening along, you're not a basketball fan. The only thing to know here is that this was probably the most shocking trade in NBA history, shipping off one of the league's best players at the 11th hour with no warning.

So let's say you wanted to console, or troll, the Luca fan in your life, you might tell the AI to spin up a weepy pop folk song about losing Luca in February to Los Angeles. And within about 30 seconds, you would get this. February's cruel whispered dreams to the wind to Los Angeles heat stray. Left me with a shadow in the cold.

Now, maybe you're impressed by this. Maybe you're not impressed by it. Or maybe your reaction is the mood is all wrong. Dallas fans should be screaming angry about the Luca trade. After all, they lost Luca Donchich for a player, Anthony Davis, who is injured so often that his nickname is street clothes. Well, in that case, you can instruct the AI to try out an angry 1990-style pop-punk screamo song with some punchy lyrics about losing Luca for Mr. Streetclothes.

Ladies and gentlemen, fair warning, the following is extremely not safe for work. Okay. So what do we make of these songs? They're impressive in their own way. They certainly resemble real songs. I think they're funny enough that I absolutely did send them to Dallas Mavericks fans in my life. But they're more impressive as a fancy parlor trick than as great music. They're a bit like an actor who's magnificent at impersonations,

but far from virtuastic at actually acting. Like, I'm not putting these songs. on any sincere playlist. And so while I've been fascinated by these music AI tools, I wasn't sure I knew exactly what to say about them. Now, that was until several weeks ago when I heard the film, TV, and podcast composer, Mark Henry Phillips,

describe his experience with AI music tools on WNYCs on the media. Phillips explained how a sophisticated user of these tools could eliminate much of the composition work that professional musicians today rely on to make ends meet. What struck me as a funny game is to Philips a dead serious matter, the difference between work and disemployment. Mark is today's guest.

We talk about the job of modern music composition, why some AI tools are eerily good at certain aspects of the job. We talk about copyright law and the ethics of creative ownership. But above all, Mark gets my brain worrying about the very nature of creativity, how great new ideas like songs come to be in the first place. The line between stealing and riffing and interpolating in artistic history has always been blurry. Picasso famously said, good artists copy, great artists steal. And this is not just a theoretical fact. many of my favorite musicians were famous borrowers.

Some of Led Zeppelin's most famous songs, Daze and Confused, Whole Lot of Love, were such obvious lifts that after years of court cases, the band agreed to add the plaintiff's name to the song credits. But analogies to music and art history fall short to capture the weirdness of this moment.

Neither Picasso nor Jimmy Page had access to an external technology whose deliberate function was to slurp up musical elements from millions of songs, store their essence in silicon memory, and serve them up in a kind of synthetic stir-fri on an order-by-order basis. Musicians have been writing music with partners for decades, even centuries. But what happens to music when that partner becomes a machine? Will it open up new horizons in songwriting and composition?

Or in a funny way, will superintelligence make the future of music more average than ever? I'm Derek Thompson. This is plain English. Mark Henry Phillips, welcome to the show. Thanks for having me. How'd you get into music? When did you first catch the bug? Wow. I mean, I guess it was, I think it was the summer between fifth and sixth grade, I made a Beatles mixtape, and I became, I was obsessed with it. I like listened to it nonstop doing gardening work for an entire summer and then convinced my grandma to buy me a guitar and I tried to learn every Beatles song I could. And then I was like, oh, wait, the piano songs. Those are my favorite. So I started.

learning piano, convinced my grandma to get me a four-track recorder, and then, you know, I experimented with layering stuff. And yeah, and then, you know, it just became like my favorite thing to do and kind of still is in some ways. And these days, how do you explain your job to someone, say, at a party? Well, I usually freeze because I don't know exactly how they answer the question, but I'd probably say I'm a composer for films, commercials, and podcasts. I mean, the truth of it is, like a lot of musicians, I have to do a few different things

to piece together a full income. So, you know, if they pry, I'll also say that I'm a podcast producer and editor. But a lot of times I just leave it as I'm a composer, probably because I think that sounds the coolest. Tell me about the first time you discovered an example of AI music that made you feel like this might be the beginning of a professional existential crisis. I saw these large language models and, you know, I think probably a lot like you and, you know, everyone.

It was like, this is cool. This seems like a promise of things to come, but it's not mind blowing yet. It's just like really cool. And then I was like, holy shit, these models are going to be perfect for emulating music, for, you know, the input of music and the output of music is perfect for the way these models work. And so I was just kind of like waiting and poking around thinking like one of these is going to come and it's going to be mind blowing. And so I, you know, kept Googling. And the spring of last year, I, you know, saw one. a new one that was kind of popping up online. And, you know, I went to like the Reddit forum for it. And I found someone posted a track.

And it was basically a rip-off of Toots and the Maidels, an early reggae star from Jamaica. And I love reggae, old reggae. Girl, you know, I'm loving you. Why don't you love me too? The things that we could do together. And I heard it. I was like, this is music.

It's not like, oh, that's cool. Like, this is good music. It has some artifacts here and there, but it's amazing. So for people who don't know Toots, who don't know reggae outside of Bob Marley, what did we just listen to as a musician? What made you stand up and go, this is a little bit spooky?

This threatens my professional ego. I have been hired to make a reggae song that's supposed to sound like 1969, Jamaica. You know, I can do a pretty good job of making something. It's certainly of the same spirit of it. But it doesn't feel like it was recorded on a hot summer night in 1969 in Studio One, which was Lee Scratch Perry's,

like, original studio. Like, you know, with like this brother duo on drums and bass, like it doesn't have the same vibe. And this one, just the vibe is like, it's good. It's like really good. And it's like you kind of need a time machine to recreate something like that. And then the vocal performance, you know, yes, there are weird parts here and there. But by and large, the vocal performance is virtuosity. just like him.

You've said this technology has given you existential dread about your job. I want to provide a really clear apples to apples comparison of human work and AI work. So on the WNYC segment, you talked about scoring a scene where a couple is getting to know each other.

And this is a cute, romantic, slightly goofy scene. Tell me what you did next. So, yeah, I guess I thought first, like, what's a good target? You know, what does this scene need? And, yeah, what came to mind was kind of like an indie drama comedy from the 90s, and I thought of John Bryan, who's a favorite composer of mine. And, you know, so I didn't go back and, like, listen to his catalog,

because I was afraid if I do that, I might do something too close to it. And I kind of just, in my head was like, what's a John Bryan song sound like? And I just started playing it. So in a way, I was trying to rip off John Bryan, but I knew my memory is so bad that whatever I create right now is not going to be a direct rip-off of John Bryan. It's just going to have a similar vibe to it. And that's a melaton playing the main melody, which is kind of this, it's an instrument from the 60s that I think has a kind of

like toyish feel to it so you know it gives it like a cute flavor to it but I think it's kind of a bittersweet melody and it just kind of has that you know end of a romantic comedy feel to it and I heard Huckabee vibes too exactly is what I got from it is it's

there's a goofiness there's a bit of irony in it. You know, it's not the most straightforwardly sincere sound, if that makes sense. But there's a cuteness to it. A knowing winky cuteness to it is what I get from this soundtrack.

Totally. And, you know, I didn't know, since, you know, when I was making the piece for On the Media, I kind of went back and I was like, what is this? Like, I know this is close to a John Bryan thing. And I went through his scores. I was like, oh, it's I heart Huckabee. Like, that's definitely what it's close.

closest to. So in this case, you had a job, and to do this job, you prompted yourself, come up with a John Bryan-style soundtrack for a movie scene. And here's where things get weird. Let's imagine that you're not Mark Phillips composer. You are Mark Phillips movie producer, and you want to score a romantic comedy scene with a John Brian vibe, but rather than hire a musician, you just tell the music AI to spin something up. Now Mark, you've done this exercise. What is the AI's version

of a John Bryan soundtrack? This is what it came up with. And like, those are real strings. Like, I would need to hire L.A. session string players to have something that good. And there's a French horn. And it's not like what I use

when I use a French horn. It's a real French horn It sounds like a real French horn player. So in a way, I just can't compete with that. I can't, with the budgets of the projects I work on, I can't hire amazing session players in L.A. So, Mark, I think people could debate whether your John Brian Riff was better

or whether they prefer the AI version. But there are, in fact, much more impressive examples. In one job, you were asked to score a scene that had an Alfred Hitchcock vibe. And so you went to AI to see if it could cook something up that sounded like a Hitchcock film. So a classical orchestra, playing something anxious and evocative for a thriller.

What happened? Well, yeah, it produced a, like, within like an hour, I produced six tracks that were just unbelievable. Here's the first one that it produced. So what I was, What I have to ask here, as I'm listening to like the trembling strings and other instruments just emerging to add little like pointilless spookiness in a way that just sounds quite human, have you gone back to look at whether Bernard Herman has been totally ripped off here? Like is this to a certain extent just playing something that exists as a fully.

completed soundtrack to a Hitchcock film, or do you detect, do you know that there's true originality here with, I guess, originality in heavy quotation marks? So my take is, it is definitely, definitely stealing quote-unquote voices. That is the strings of a Bernard Herman recording. that is the, you know, there was a harp being plucked in the background and a trumpet. Like, those are the instruments from real Bernard Herman recordings. I don't think the melody and the performances are stealing from an actual composition for very long. I think it probably only happens for like a microsecond and then the AI kind of goes in a different direction.

than the original composition does. So I think it's stealing the voice, but it's just kind of randomly coming up with melodies, orchestrations, arrangements that are perfectly like Bernard Herman, but not the exact same thing he did on Psycho or, you know, whatever. In your WNYC segment, you said that you're genuinely afraid that this technology could replace you.

How would it replace you? I mean, yes, I think it could replace a lot of my work. You know, commercial stuff, film composing, podcast composing. You know, why would you hire someone who charges you, you know, a fair price for the time it takes? and for sort of the licensing of the music, when you could use one of these things, produce 10 tracks in an hour, and conceivably use them for free.

I think you can use them without getting sued. So why would you hire someone when you don't even have to have that process in your workflow? If you're making a commercial and you want to have a song that sounds like, you know, the Rolling Stones, you could have 100 songs that sound exactly like the Rolling Stones in an hour, and your editor can start cutting to them. So I think it's going to happen.

I think there will be some delays because people will be afraid of the legal consequences, but that'll happen in films, podcasts, commercials. It's going to happen in all these different realms. There are three layers here that are. interesting to me that I want to make sure we hit in this conversation. The first is the practical layer, which is, should musicians use this technology? And if they use it, how should they use it? The second layer is moral or ethical. Is this technology legal? And even beyond legal, are these tools right to use? Do they rob us of something? Do they rob the original makers of

music that's been fed into these generative AI systems of something. And then third, I think that these technologies raise really fascinating questions that exist at the realm of culture and even something that touches on philosophy. And I want to make sure that we save some time for those. Let's start with the practical. You had to come to Jesus moment with this technology that has for now solved at least some of your existential dream. What was it? Yeah, so it added a feature where you can upload your own music and then extend it. And that was, yeah, it was a mind-blowing experience because I uploaded just a little 12-second

stinger I did for a commercial 10 years ago that I had totally forgotten about. And this is that. It turned that into a full. full track and here's where it switches to the AI version. It really feels like where I was imagining taking the track and it did little things that I had envisioned but did not put into the prompt like a little drum machine here or, you know, a synth here and it's just really amazing. I mean, the other thing that is really cool about this to me as a musician is I can then

take this and use it as a blueprint for where I take the song. So maybe I don't use the AI audio, but I could pull up some synths that sound exactly like this and play that part. And maybe I put my own little spin on it while I'm playing it. Or maybe I ignore this, this and this, but I do that part and that part. So it's like, it's like having, it could be like having a writing partner. And you can take some of the ideas, reject the others. And I think, like, for a musician, that's where it gets really exciting. It feels like you could use this to kind of never get stuck on the song again. And I think all musicians have the problem of where to go next. I'm stuck here. It's 50% done. It's 60% done. And this feels like a little thing that could get you over that hump.

Well, just to stay in the realm of the practical here, there is so much music that's written alone, right? It's somebody in the attic with a piano, in the basement with a guitar. They're just by themselves, staring at the wall without thinking, moving their fingers, moving their mouths, making sounds. They don't have a writing partner. And, you know, AI is not John Lennon. It's not one-one-thousand-one of John Lennon. And certainly using it doesn't make you Paul McCartney. but it sounds to me like you're saying

the existence of these tools that create a kind of sandwich, you know, human to AI, back to human sandwich. As absurd as this might sound to some people, and frankly, it sounds a little absurd to me too. Like, it does create the possibility of a partnership where you write this little guitar riff,

you write a little piano piece, you wonder, is this a thing? You record it, you feed it to the AI, it adds other instrumentation, and something new pops out that then you can run with. Is this kind of how you see working with AI now? Yeah, if I move forward with working with it, that's exactly what I'm going to do.

And it's exactly how you write with another person. I mean, I was just texting with my buddy Graham yesterday, and he's an amazing guitarist. And the way we would collaborate sometimes is I would say, you know, what would Steve Cropper do? What would George Harrison do? And, you know, he's listened to enough of the music that he gives sort of his spin on what they would do. And that's what the AI does. Yes, it's not John Lennon, but it kind of knows what John Lennon would do in this situation if he had reached this point in a song. Before we get ahead of ourselves, let me just dribble some cold water on all of this.

Do you have any sense that musicians actually want to work with these tools? Like since your WNYC piece, did musicians write to you to say, sorry, dude, this is completely alien and messed up? Or were they interested in using AI as an extension of their creativity? I will say since my on the media piece, I would like, I would say 30 to 50 musicians have reached out to me sort of very excited about the prospect of using it this way. And, you know, some, you know, one was a famous electronic musician who, like, is, I've listened to for a long time. And, you know, it was, like, really interesting to see how excited musicians were by it. But they all had this exact same. It feels different. You know, this isn't like a new piece of technology, like a new keyboard or a new, you know, plug-in, digital plugin. It's, it's, it's something. different. And so everyone is excited by it. And I think musicians are inherently, ooh, there's a new tool that can make good music. I want it. I want to play with it. So when you and I were emailing back and forth about doing this show, you proposed what I initially considered a pretty weird

experiment. You said, Derek, if you want to make a little piece of music and send it to me, I can use my AI tools to build something out of it. So, you know, I dabble on the piano a little bit. I don't play anything very good, but I can sort of play triads. I can hunt and peck little melodies. And last week, I was, I happened to be at a recording studio of finishing the second half of the audiobook of abundance. So I'm in the studio. They have beautiful pianos there. I sat down and I played something that I'd come up with. And this is like a little interpolation, a little riff on one of my favorite pieces of classical music, a concerto by Shostakovich, who is a 20th century Russian composer that I like. So I recorded it. I email it to you,

and I said I imagined it could serve as the basis of a Danny Elfman-style Tim Burton fantasy or a Pixar fantasy tale theme. And here's how it goes. And then you fed it to the AI, and it does this. And then you let the AI keep massaging the melody, and this happens. Mark, what did we just hear? Yeah, I think we heard it doing what it does great, which is it hears what you were doing. It hears the sound of it. And it just continues it. And it doesn't need much to see kind of what the piece is. It still feels like it grew out of the piece you wrote.

And I will say that kind of one of the limitations of where it went was that, you know, you recorded that with your iPhone, I'm assuming. and it's a pretty like rough mono recording. So like it can't, it's not going to go to, you know, a full orchestra because there aren't any recordings that sort of start with iPhone demo and then go to like, you know, the New York Phil Harmonics. But if you had started with a good recording of your piece, it would have gone in a different direction. It probably would have become even more lush and, you know,

well produced. So, you know, it's limited by what you put in, but I think like from a musical standpoint, it took it, it made something that feels like it, but, you know, much more complex and intricate and amazing. When I think about this song and the degree to which I wrote it, I feel very weirdly like two things are true. One is that I did not write this. I didn't come up with the inspiration of the original arpeggio. I certainly didn't finish the track that you just played. But that sits alongside this very strange other thing that when I sent this audio file to my wife,

and my wife said she liked it, I reflexively felt a little bit proud. I felt proud about something that I'm not entirely sure I even did. I wonder, A, whether that makes any sense, and B, whether it touches on the emotional, legal, moral, artistic messiness of creating art with AI. Yeah, I think that's it exactly. And yeah, I think, like, an even grayer area would be, what if you took that piece and then tried learn it and you learn bits from it, you know, you probably, I couldn't play that myself. And so you would probably play like a more simplified version of it. And then you might make a little tweak, oh, it goes to an A minor here, but I want to go to the F. You know, and so then what's that? Is that

yours? It feels a lot more yours than that piece, but it's not entirely yours. And I think that that point, right there is where music is going to be in five years. I think all musicians, well, let me take that back, I think a lot of musicians are going to start using it. I think it's going to start getting baked into Pro Tools, Ableton, Logic, and, you know, it's going to be this spell checky thing. But it's a lot different than spell check when it's helping you come up with a bridge or helping you come up with a melody. But I think it's going to be. be just in the water very soon because it's so good. And why, like, how could musicians not use this tool that could make unbelievable music and help them make better music than

they could without? So there's a practical question. If people use this technology, how should they use it? And I think our humble examples show how AI can essentially amplify the little brain seeds of musicians. But there's a very important and very distinct question, which is, should people use it? Is there something morally or spiritually or creatively corrupting about the use of these tools? Some people will hold out. A lot of people will. And I think we should as a society. But it's just like, it's there. And I think people are going to start using it. And yeah, I don't think there will be, you know, top, you know, I don't think there will be number one hits that are just purely AI generation, someone typing in a prompt and what the AI spits out. But I do think that people will use it

like a writing partner and it's going to produce amazing results. And like that's the thing that's so scary is I know a lot of musicians who are like, fuck this. This is bullshit. It's theft. It's awful. I hate it. And a part of me feels that. way too. But what's blowing my mind is just how good it is. And so like you have to respect it at the same time that you scare, that you fear it. And if I think of a 15 year old today, who is an amazing musician, who's, you know, the next, you know, Tame Impala, the next, you know, Paul McCartney. And they have this tool in their arsenal. I have, think this would probably make them even better and could make them produce a million times more.

And I don't know. I think they'll just grow up using it. And it's going to be just a part of music very, very soon. You're touching on aspects of culture and philosophy that I really do want to hold until the end. So before we do that, let's talk about the law. The state of play in AI music is that the major labels have filed lawsuits against several of the major AI startups in this space. Those lawsuits are ongoing, and from my reporting, it is not 100% clear how the courts will ultimately come down on this issue of fair use and copyright in generative AI. So in the absence of any hard and fast ruling here, I think we should just talk about it. Do you think these tools are legal? No, nowhere even close. I mean, I think

clearly there is clearly they have sucked in basically all the music ever recorded and that's how they've created this model. Their argument is that their model just listened to music just like you or I listen to music and then create a piano piece. But I think it's entirely different. And I think you can look at the license of the music. You know, when you buy a piece of music, you're allowed to listen to it, but you're not allowed to put it in your commercial or your movie. You have to pay a separate license for that. And there is certainly no license when you put your piece of music on Spotify that you're granting some company to create an AI model off of it. So I just think it's pretty clear cut that they do not have the right to do this.

but, you know, that's being litigated right now. I think the RIAA is suing the two big companies and trying to, you know, stop them. But I don't think it's going to stop the AI music generators from existing because, honestly, I think the music labels are going to create their own or at least do some sort of partnership with the pre-existing ones. The YouTuber Cleo Abram, who's covered this space, has a lovely way of formalizing this question

of legal responsibility in music. She has this two-by-two box of copy versus inspiration, pay versus not pay. And there's two squares of this box that are obvious to practically everybody. Everybody acknowledges that if you copy something to produce a paid product, you pay the person you're copying.

That is practically universally recognized as what you should do or what is legal under copyright law, it's also generally understood that if you're merely inspired by somebody, you don't have to pay them, right? If you interview modern artists, if you interview, you know, Kendrick Lamar or Taylor Swift,

and you say, who's inspired you, and they give an answer, you're not going to immediately say, oh, why haven't you cut them a check for the fact of their inspiration? Nobody thinks like that. AI does scramble this an interesting way because it creates this black box

where inspiration and outright theft are intertwined. And a point that you made that I think is really important is that if I use a Spotify link in my commercial, that is if I make commercial use of a song, I have to pay the person who owns the song.

But these AI companies are, in a weird way, making commercial use of music not by copying it, but rather by putting it into this weird black box where inspiration and theft and synthesis is all mixed up and jumbled. So how do you think these questions ought to be resolved? I think you're exactly right. It's on the input. We can't look at the output and say, litigate every song. Oh, this sounds too much like that. And as an aside, I think that part is kind of of haywire currently with real live music. I mean, without getting too deep into it, the whole Marvin Gay blurred lines decision, I think every musician is baffled by that and terrified of that.

And for the listeners who aren't familiar, there was a Robin Thick song produced by Farrell that was clearly inspired by Marvin Gaye. But if you listen to it, it's not the same melody, it's not the same chord progression, it's not the same beat. There is no sense. going on. If you played the two songs, you know, on your keyboard and drums at home, they would not be the same song. But there was a legal decision that basically said, this is too close because we heard that they were inspired by this Marvin Gay track. And it's just so dumb as a musician. Because like, what is the blues? I mean, it's literally the same chord progression. And there are a million blues. songs that are in E or in A. And, you know, if you, you know, whole genres basically are impossible to play if you're saying, oh, this is too close to that. You know, yes, if you are literally the exact same melody with the exact same chord progressions, with the exact same lyrics, yeah, that's stealing. But all we can do is look at the input and say, do they have the right

to suck in every song ever into their model. Is that the same as you and I listening to every song? And I think it's just so clearly not. It's not a person. It's not listening. It's, I don't know what you call it, but it's like sucking it in and absorbing it and putting it onto its neural networks somewhere.

And that's a different thing. I want to hold on this image of a black box. In your radio segment, you had a very beautiful yet strange insight. You said people don't know how their favorite songs were made, but musicians also don't know how the process of writing music works. Musicians aren't conscious of what they're doing. They mess around. They discover the song at their fingertips at the tip of their tongue.

they're just sort of guessing what the next note is. Those are your words. They're guessing what the next note is. You must have been aware when you said that, how close to the bone that description of creativity is to generative AI, which is often described as prediction, next token prediction.

And it raises a really interesting question about creativity. I don't even know if there's a psychology of true. psychology of creativity, but certainly the philosophy of creativity. As a musician who uses AI, do you see a profound similarity between your creation process of listening, remembering, creating, and the creation process of artificial intelligence, which is pre-trained, has a memory of sorts, and synthesizes that memory in response to prompts? Totally. Yeah, I think it's about sucking in as many references as possible, knowing the appropriate references in the moment, and then just doing

something that has the vibe. And yeah, there's a guessing process in music that is way different than photography or, you know, art or writing. You know, when you're writing an article, you're not going word by word, just, you know, what word comes next? You have an idea, you have an idea for a paragraph, a section. Music, I think, is a lot different. I think you really just, on a certain level, are guessing what the next note is. And I think there's degrees of that.

You know, free jazz takes that to an extreme. But any songwriting process, yes, you have your limitations, you have the key you're in, you have the genre of music you're playing. But ultimately, you have to kind of take a leap and guess the next note, whether that's with a pen to paper or just while you're fooling around. So I think it's, it lines up perfectly with the AI process. And that's why I think the V1 of AI music is so much better than the V1 of the chatbots or the image generators. I do think those other mediums will catch up.

But I think that, to me, music is a year, three years ahead of these other ones. And so I think I'm having the existential crisis that basically everyone is going to be having in, let's say, two years. How do you think this is going to change the process of the experience of making music? It seems to me like in a world where you don't have these kind of automatic feedback loops from an AI, independent alone music writers are living inside their head waiting for something to happen, waiting to hear it, and that feedback loop is entirely internal, right? They have an instinct, the instinct is represented by a sound of the piano or the guitar or voice, then they have taste, they judge the sound, and then they adopt or they change or they keep going.

But it seems like the world that you're describing is a world in which writing music is more like being an editor, more like being a manager, because you are creating something that then you're feeding into a system that you can evaluate at a distance, evaluate almost as if it's someone else's work, which is so interesting to do, and then incorporate the output from that AI partnership to pull back into your own work. Do you see this as changing in some real, good, bad, meaningful ways the process of writing music. Yes, I think it's definitely going to change it if people start using it. And I will say, you know, depending on the genre, the type of musician, so many musicians are going to say,

I will never touch AI generators. And, you know, I don't know if I'm really going to incorporate this into my workflow. I'm just sort of exploring it more out of curiosity and just because I'm impressed with how good it is. It doesn't mean I want to use it moving forward. And again, like the thing that is so tantalizing for me is I have hundreds and hundreds of songs that are just sitting on hard drives half finished. And I know there is a dozen songs on there that are really good. And if AI could help me finish it, that's really appealing to me.

Because, like, a song spending, you know, a song sitting on a hard drive from now into eternity and never being listened to again, that's worth nothing. And so if AI turns it into something and it's something different, something a little weird because it, the process involves AI. Yes, that's weird, but at least it's something. But I think the gray area becomes like, if that just becomes your process. And I think there is a huge difference if it's for commercial purposes. You know, if someone hires me to do, you know, what's called in the industry a sound alike, you know, hey, we want an MGMT song for this commercial, but we don't want to pay MGMT come up with something. I don't know. Using one of these to generate some starting points, that's really appealing, you know, because they'll be able to come up with something that isn't.

too close, but still has the right feel, and I could build off of it. So that's really tantalizing, because to try and come up with something with the right vibe purely on my own would just take much, much longer. And a lot of those jobs, it's like, hey, we need something tomorrow, you know, a fully produced track that fits this commercial. And so, yeah, it's really hard to imagine if the tools exist and they're legal to use just not using them. What I really like

about that answer is that you're holding on the table two possibilities for the future that really pull us in different directions. One is that these tools are simply too weird for artists to feel comfortable using.

That, as you said, writing music is not like writing paragraphs. nobody thinks today certainly that spellcheck intrudes on the authenticity of essay writing nobody thinks that but working with AI to write music does in a small or maybe even enormous way feel like an unwelcome intrusion to some people an unwelcome intrusion into a process of creativity that's that feels to the artist like it's meant to be in the space of four humans only. I think that is absolutely a possible future.

But there's another possible future where these tools, because they're so good at the small tasks that you've specified, at finishing incomplete work, at filling out, unfilled out ideas, will be ingrained in the music writing habits of young people today and will see AI move into the world of music not because a bunch of 40 and 50 year old suddenly adopt it, but rather because they'll be demographically replaced by musicians who today are only 15 or even five years old. And that's where I want to close is taking that

possibility seriously. I'm very persuaded by this idea that is sometimes called stuck culture theory, which says that something's happened in the last few decades in film. in celebrity, maybe to a certain extent of music, where cultural progress feels more frozen than it used to be. It's like the one factory that I always go back to lean on here is that in 1996, almost all of the top 10 films in America were an original screenplay. And today, really, for the last 10 years, practically very few of the biggest blockbusters are original screenplays. It's sequels, adaptations, and reboots as far as the eye can see. And, you know, stuck culture theory can extend to TV as well. The idea that Gen Z is still watching the office and friends,

and there really aren't a lot of shows made the last 10, five years that have, you know, achieved that sort of pantheon status. I wonder if AI will make music more stuck as well, because if AI is essentially training the future of musical creativity on a perfect understanding of music's past, we might get stuck in the grooves of perpetual refinement in a way that keeps music

from evolving the way it's historically evolved. And I wonder how much stock you put into that sort of cultural story. Yeah, I think about that a lot. You know, the image is a snake eating its own tail.

I mean, it's like if if AI is going to be in the music that we create and then, you know, and then that will be fed back into the AI and then, you know, will be worked back into the music that's created after that. It's just going to be a feedback loop of AI. And I don't know if it's ever going to be 100% feedback, but I think it'll grow. You know, maybe five years from now, AI is, you know, let's say, responsible for 5% of kind of coming up with new things in the song. Like it's sort of the writing partner model.

I think that will only increase and I think it will increase at an exponential level because of this process. So that really scares me because, yeah, I think humans will have less and less of a role in the writing. At the same time, yeah, I mean, maybe people use it. it in really creative ways, and it's not, it actually creates a really new, cool product that wouldn't have happened just with human evolution. But this is where I get sort of a little gooey where I'm like, it's just not pure anymore. You know, from like the dawn of time

up until last year, music evolved by like humans listening to other. humans and making music together. And it feels like now there is going to be this algorithmic element into it. And that's really scary to me. I don't know, maybe, yeah, maybe it makes better music, but there's still something icky about that to me. Mark Henry Phillips, thank you very much. Thank you. Many thanks to Mark Phillips. I think my main takes to Mark Phillips. I think my main takeaway from this interview is the distinction between inputs and outputs with AI. I think a lot of people that consider themselves or present themselves as AI skeptics tend to be very critical of the outputs of artificial intelligence. They'll say, look at this piece of writing,

it's so wooden. Or listen to this song, it's such a cliche. But where I see AI being sneakily effective and ultimately potentially transformative is as an input, it's the way that it allows software programmers to accelerate their coding. It's the way that it allows white-collar workers or writers to accelerate their research. And it's the way that it allows musicians or composers potentially to work on something, a little ditty, send it out to some artificial AI assistant, and then get something back that's just a little bit better or even just a little bit different than the thing that they deliver to the AI.

That's AI as an effective input, as maybe we'll see in a few years, a kind of universal input to the creative economy across white-collar work and writing and music and all sorts of idea generation. AI as the all-purpose input, I think, is a really interesting idea to play with. And I'm very grateful that Mark helped me concretize it. We'll talk to you Friday. Has heard been. The

Meregore Conversion of the world. The incredible system of Pago of Shopify facilita the

business on your website, and in whatever that's music for your

ears. No, you're more and you do you do your

business is a great to be a euro per per per

per per record's Thank you.