Switched on Pop - The music industry's AI fight

Episode Date: July 19, 2024

The Verge's Nilay Patel and David Pierce chat with Switched on Pop's Charlie Harding about the RIAA lawsuit against Al music startups Udio and Suno. Later, Nilay and David discuss the rest of this wee...k's tech and gadget news. Subscribe to Vergecast Learn more about your ad choices. Visit podcastchoices.com/adchoices

Transcript
Discussion (0)
Starting point is 00:00:00 If you're tired of endless scrolling to figure out where to eat, same. I'm Stephanie Wu, editor-in-chief of Eater. We've just launched the new-ish and way better Eater app. It has all the restaurants we love, gives you personalized picks wherever you are, and serves up smarter search results just for you. You can find my list of the best places for martinis and fries in New York City. And save your favorite spots, share lists, follow editors, and book right in the app. the Eater app at Eaterapp.com. It's free for iOS users. Welcome to Switched on Pop. I'm
Starting point is 00:00:50 songwriter Charlie Harding. For a minute, seemingly every music influencer was going online in saying that AI music services would be the death of pop music. Just like ChatGPT can write sonnets about auto parts, new music AI tools like Suno and UDio can generate entire songs from prompts. You want Grandma's Dubstep Christmas? Or how about the pledge of allegiance to your toothbrush? I pledge to my toothbrush now, sparkling teeth it doesn't down. Bainty fresh and so clean, white a smile you've ever seen. I mean, come on, with songs like those, you really should be just giving up on your rock star dreams, right?
Starting point is 00:01:35 No, seriously, though, this stuff is consequential. These music AI startups are being sued by the Recording Industry Association of America for infringing on their copyright. And this lawsuit could possibly even upend the entire AI business. I recently spoke with Nilai Patel and David Pierce on Vergecast about what's going down and whether AI is killing music or if AI music might kill artificial intelligence. Here's that conversation. And if you enjoy it, subscribe to Vergecast in our show notes. Hello and welcome Vergecast, the flagship podcast of BB Eldruzy.
Starting point is 00:02:10 That one track needs a podcast. We're going to talk about it for an hour this week and then an hour next week. minute by minute. It's actually fun because then we get to spend the whole episode debating whether B.B.L. Drizzi can be our theme song legally. Now that we've said that. And it's perfect. We're going to do a lot of fair use on the show today. Yes.
Starting point is 00:02:29 It's coming. Hi, I'm your friend, Eli. David Pierce is here. Hello. Charlie Harding from Switch on Pop is joining us. Hey, Charlie. Hey, guys. Thanks for having me. A big week in the AI landscape. A weird week for me, emotional.
Starting point is 00:02:42 I'll explain why in a second. The major labels have sued two AI companies that you, may or may not have heard of. UDEO and Suno, these are classic Silicon Valley names, we should all be proud. They're music companies, they're AI music companies. UDio is the company whose tool was used to make B.B.L. Drizzi,
Starting point is 00:02:58 we'll have Charlie explain that entire chain of events. And then Suno has a deal with Microsoft. It's in copilot, and you can ask it to generate music and co-pilot. How would you train an AI tool to make music? You might ask, as did the lawyers at the Recording Industry Association of America. And the answer is you just put a bunch of music in them.
Starting point is 00:03:18 And you train the model on a bunch of music, including what appears to be a remarkable amount of the recording industry's music, not just music that's out in the world. The companies are sued. There's a lawsuit. The RAA is mad. The RAA is already mad. They say that they asked the companies what was used to train the models. The companies, the Suno, deflected and said their training data was confidential business information. Fine. UD.O made the same statements.
Starting point is 00:03:48 And then Suno also said, our stuff is transformative and designed to generate completely new outputs, not memorize and regurgitate preexisting content. We'll get to that in a second because that is actually a very funny claim based on what you can do. The RAA put tracks in lawsuit, which we will listen to. It straightforwardly can just make Johnny be good in a song called Prancing Queen, which is deeply hilarious. there's just a part of this where these companies knew that what they were doing was training on copyrighted work. And they kind of assumed that they would get away with it or at least be able to like pay the money and move on. Can I ask a very possibly dumb framing question about this?
Starting point is 00:04:31 Because I want, we're going to get into the weeds of like a very specific music case. But a lot of what you just described, the sort of basic outlines of like how these companies trained their data and what they're being. accused of doing and who's mad feels identical to me to every other conversation we've had about these AI models, right? Like, it seems to me that if you change Johnny Be Good and Suno to the New York Times and chat GPT, we're functionally talking about the same thing. So, like, I want to get into the deep into this in weird ways, but like, is there something different about the fact that we're doing this with music than the fact, the way we've been doing this with, like, the web for the last
Starting point is 00:05:12 year or so? Yes. And it's that the music industry is organized and aggressive when it comes to protecting its rights. So the difference is lawyers. The difference is lawyers. And also the history of music and the internet is lawyers. Like, the reason I'm so conflicted about this is that I went and became a copyright
Starting point is 00:05:33 lawyer because Napster came out when I was in college and I was radicalized because I thought everyone was stupid, which is what should happen when software. comes out when you're in college. I'm assuming there are some kids out there today who are radicalized by the presence of new software. But for me, it was Napster. I went off and became a copyright attorney. I sucked at it.
Starting point is 00:05:55 I don't ever want anyone to think I was good at this. As a practicing attorney, I was garbage. But the thing I did was I worked at a law firm that defended other college kids who got sued for using Kazah against the RIA. And the RIA was suing college kids and actually suing the universities to get the IP address of the college kids on their networks who are using the tools and then identifying the college students and student in college students. It was my roommate's computer.
Starting point is 00:06:17 It wasn't me. I swear. This is like a lot, right? That's just a lot to be like the universities are caving and not protecting their students. All this stuff is happening. And they were running this program at break even. They just wanted $5,000 settlement so everyone would be scared and stop doing it and they won. That worked.
Starting point is 00:06:34 Sort of. Now we're back to it, right? But like the movie industry is not quite as good. stopping the piracy as the music industry was. And they ran that program to kind of shut it down and chill Napster and Grokstra and Kazan, all the other ones. They ran Grokstra all the way up to the stream court.
Starting point is 00:06:50 That decision is 20 years old now. The RAA versus Groxer. Great. Then they did the iTunes store, and they got deals with Steve Jobs, and then other music stores, and then they did Spotify and Apple Music and streaming. And all along the way, they have been extraordinary litigious
Starting point is 00:07:08 and extraordinarily protective of their copy. because people have an emotional connection to music that they can trade on. In a way that even Hollywood doesn't. When Disney's like, we must protect the copyright to the Avengers, people are like, go fuck yourself. Like, right, in a way that the music industry, when it's like, you're stealing music. You are making it so artists don't get paid. There's an emotional resonance that argument tends to have. Charlie, I'm curious how you see that changing over time, but they've been good at it and they've run the playbook over and over and over again.
Starting point is 00:07:38 Music is also a closed ecosystem of like four big companies. And so you see things like interpolation where an artist uses a bit of a song from another artist and fans get mad and then writing credits get distributed and money flows inside of a closed ecosystem. None of that is true about the web. None of that is true about the media. So if you are a writer at a media company, nothing. Like no one cares. Like there's not some big apparatus designed to make it as see. like your work is emotionally resonant and should be protected in the way that the music industry does.
Starting point is 00:08:12 So I think that's a huge difference here. The other difference in this is where I really try. I'm very curious for your take here is that the outputs of these systems are just the songs. Right. And the lawsuits are about training, right? All copyright lawsuit. All copyright is dumb. Like it's a dumb legal system because it just regulates copies.
Starting point is 00:08:36 And so they're like, you made a copy to do this thing and you didn't have the permission to do the thing with the copy that you made. So copyright infringement. Weird. All computers do is make copies. We've talked about this so many times in the show. So they're not talking about the outputs. But I think the outputs are so convincing. Like, in order to get to the thing can just make Johnny be good, it is obvious that you made a copy of Johnny Be Good.
Starting point is 00:09:02 Charlie, what do you think, man? I feel like the music industry is the one place that copy. right becomes part of public conversation in a very frequent way. People talk about when Farrell is borrowing from Marvin Gay. People talk about when Sterey to Heaven might have been copied by another song. People constantly are debating in social media. Hey, you kind of sound like you borrow this other person's thing. They should get a credit on it. There's no thing in journalism where someone's like, hey, you know, Neelai kind of used someone else's text. Maybe we should give David some extra credit on that piece that Neely plagiarized.
Starting point is 00:09:38 I'm sorry, I'm obviously not accusing you. I do that all this is hypothetical. I don't think you understand. Nelai does that all. I have God mode Google Docs access and I just lift everybody's copy left and right. There's just no other creative industry where there's this like a level of actual infringement that happens that then there is this internal litigious system where these major players are constantly trading credits back and forth and the public is in on it often debating
Starting point is 00:10:04 whether or not there, the artist that they stand is making original work. This is part of popular discourse and it happens in music. So it's really different when the music industry is going after AI than if the New York Times is. Because I've just never been in a circumstance where I'm talking about that article was completely lifted from this other article, but we constantly talk about is so-and-so writing their own songs? Right. I'll give you a dumb example from movies. The movie The Dark Night is basically an interpret. of the movie Heat, except with Batman.
Starting point is 00:10:37 Yeah. Right. Like, many of the shots are the same. Christopher Nolan is out there being like, Heat is the greatest cops and robbers movie ever made. Like, I just wanted to make heat, but with Batman. And then you, like, watch both movies. So, like, oh, shit.
Starting point is 00:10:48 Like, dude just made heat with Batman. Yeah. And that's great. And I think everyone's like, this is the best. Like, this is so cool. And then you get to, I don't know, Miley Cyrus is writing Flowers, which the lyrics just reference. Bruno Mars.
Starting point is 00:11:02 And people are like, should she have to pay Bruno Mars? It's like why? Like, why? Like, the melodies aren't, like, she's just saying some of the same words. Yeah, yeah, yeah. Right. But the expectation is that there should be some economic exchange of value. Well, and I would think then that the existence of that exchange of value is also how you get to something like this, right?
Starting point is 00:11:23 Where it's not just everybody's sort of yelling and then nothing ever happens. It's that we actually have a system by which this gets solved, right? Where we have people get songwriting credits. then they get paid for it. And there's like, the music industry has built ways for all of this to work. And like you said, is very good at picking fights with those who do not play inside of that system. And so, like, now I'm just thinking, okay, the difference between the music industry and the others who are fighting against Open AI and others, like just groups of authors or whatever, is not only does the music industry have a system by which it understands how everybody is
Starting point is 00:12:02 supposed to get paid that has existed for a long time and everybody kind of sort of understands, but also is able to marshal that whole system against anyone who wants to exist outside of that. And part of what we've been saying forever is like all the money is on the side of the AI companies. Like who is going to be able to run a lawsuit against Google all the way to the end? And there probably aren't that many creative industries other than the music industry that might be able to do it. They've done it before. Yeah. And they have one major concessions.
Starting point is 00:12:32 We wouldn't have content ID if not for music labels going after YouTube. Right. And the music labels have gone after YouTube, right? Universal. This is where Drake gets involved. There's fake Drake that we've talked about endlessly on the show. Shout out to LaserBong, a song that has been just aggressively censored from every major platform. Boy, does TikTok not want electronic music about bongs on its platform.
Starting point is 00:12:57 Sue about that, record labels. Like, you want to talk about government censorship. Yeah. All right. The Chinese government is like no bongs for American teens. Just saying it. The piece of that puzzle that's super interesting to me is Universal was mad about fake Drake. They went to Spotify, Apple. They said don't have the song. Don't have the song. And Spotify Apple control the catalogs and their music service they can pull down. YouTube is open access. Right. So anyone can upload anything. And then there is contentity and they have these other copyright management systems on YouTube. But it's not the same catalog control where you can just delete the song, the way they. that Spotify and Apple Music could just delete the song. So they had to come up with some other system. And YouTube did give the concessions, right? They put out a list of like AI principles that they would work on, about safety,
Starting point is 00:13:42 about this other stuff. And a lot of it was we're going to work directly with Universal and the rest of the music industry to figure out what tools are valuable and what tools aren't, and even to allow some creators to like do some of this AI music generation because we think it's cool. And then obviously we'll figure out how to pay them. This is implicit if you're like co-announcing that tool with Universal Music. Universal music is going to get paid. The heat death of the universe could occur, and Lucian Grange, the CEO of Universal Music, would get paid.
Starting point is 00:14:09 Like, he's good at it. Yes. These two companies did not do that. There's a quote from one of the VCs in Suno, a Rolling Stone profile from March. He just says to Rolling Stone, if we had deals with the labels when the company got started, I probably wouldn't have invested in it. I think they needed to make this product without the constraints. Wild thing to say out loud. Incredible thing to say out loud.
Starting point is 00:14:31 said the quiet thing I love because behind all of this, they haven't ever really quite publicly admitted that they are using copyrighted works. They've talked around that in every way they possibly can without denying it. And this is sort of like the big, yeah, this is the smoking gun. Right. And then the other smoking gun, and this is part we should just listen to some of this music, is the output of the models themselves, which is, I mean, let's listen to it and we can talk about it.
Starting point is 00:14:57 David, take us through it. Okay. I have brought some sound. thank you to Andrew and Liam for getting all of this together. We have three examples, and I will just roll through all three. Stop me when you have feelings about this. The first thing we'll do is this is just titled Real Chuck Berry in Riverside. So let's just listen to this first.
Starting point is 00:15:19 By the way, and this is all fair use. No copyright infringement intended. All right, YouTube. All right, go ahead. But he could play a guitar just like a ring in a bell. Go, go. All right, that's enough. Stop it before the robot sensors arrive.
Starting point is 00:15:39 Great song. I can confirm. That is it. If you don't know that song, one, go watch back to the future. Incredible movie that you should just watch. And two, like, stop. Go get cultured. Johnny Be Good.
Starting point is 00:15:51 Chuck Barry. Yeah. Founder of rock and roll, one of them. Yeah. And now we have a song from UDio, which is not called Johnny Be Good, but just listen. But he can play and get caught just like a ring of bell. Go, go. Go, Johnny, go.
Starting point is 00:16:17 It's... So that one is not a copyright infringement, just on its face. Yes, you can't. You can't. Nobody owns it. You can't copyright the AI work. Yeah. Okay, so that's one.
Starting point is 00:16:28 Next, we have a James Brown. I just want to ask Charlie a hard question. Charlie is a professor of music at NYU, I believe, right? Yes. All right. Professor. Let's say you didn't have Johnny Be Good in the training data, but then you had the entire architecture
Starting point is 00:16:45 of rock and roll that is built on Johnny Big Good. Could you back your way into Johnny Be Good? Monkeys and typewriters and Shakespeare, right? Like, is that the theory here? You don't need, do you need the seed crystal when you have the whole diamond mine of rock and roll? It's a really good question because, you know, if you look at this era of music where R&B, rock and roll are sort of like rockabilly are all kind of one entity, the music is drawing on 12 bar blues, a very standard song. structure. They're often in the same keys that play well on guitar. They're using a lot of the same kind of guitar licks. They're mostly using pentatonic scales. They're using a lot of the same
Starting point is 00:17:24 language. And you might think, yeah, you could just sort of back into that. But the precise rhythms, the exact words. Let's be clear, the copy is worse. It does not sound nearly as good as the real Johnny Be Good. You could never get to that same place without having heard that Johnny Be Good. I feel confident. And if these two songs, one being a real song and the other being nobody owns the copyright and it's made by an AI, if there were a copyright case made against the copy, it would definitively lose. I have many of these cases, many of these music copyright cases are not clear. They're often fought over eight notes, sometimes six notes, sometimes a general feel in the case of blurred lines. and I got to give it up.
Starting point is 00:18:13 This is a direct copy of the rhythm and the words and most of the melodies. So, no, you had to have heard this song in order to make Johnny Be Good twice. By the way, if you want more on eight notes and how complicated it is, Charlie and I did an entire episode of Decoder pulling that apart. We'll link that in the show not somewhere. But let's move on. Let's listen to the next one. Maria, you have a podcast now and you need to start acting like it. What's the first step as a podcaster?
Starting point is 00:18:43 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.
Starting point is 00:19:04 Ready? Ready. Do not sugarcoat something for me. No. No. No. We'll dive into their stories and get valuable insights from top executives, actors, entrepreneurs, and other individuals
Starting point is 00:19:15 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. back to the places from which they came.
Starting point is 00:19:57 But what we want to do in this space is 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.
Starting point is 00:20:27 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. All right. Next up, we have James Brown. First, real James Brown. Another one where if you're listening to Verchast
Starting point is 00:21:05 and you've never heard this song, like, I don't know what you're doing. Go listen to Switch song. Go. Yeah, exactly. You need it very badly. Go to school. And then come on home.
Starting point is 00:21:14 And now we have AI, James Brown. Wow. I feel good. I knew that I want. It's spicy. Okay, so this one I actually think is very fun because it is a little, it's like a little tiny bit further away from it. In the way that the first one just is Johnny Be Good, but worse. This one is like one tiny tick further away from being just the thing.
Starting point is 00:21:47 This is a lounge band with pretension where everyone's like, just play. Just play it the way that can you stop it? All right. Like, we get it. You thought you had something. I have one question about this. And it actually felt the same way about Johnny be good. The way these models work, right, is you adjust a bunch of data into them.
Starting point is 00:22:06 They set a bunch of model weights. And then they like statistically make the next bit. Right? They're just sort of like assembling the thing around the prompt. Which means they're kind of just assembling ones and zeros to make a sound wave here, which is pretty wild, if you think. about it. Like, that makes sense with words, right? What is statistically the next word in the sentence around this prompt? When you get to like, what are the ones and zeros of a sound file and you get to statistically hear the next ones and zeros? That's pretty weird. It's actually a pretty weird way to make an audio recording. And the thing that strikes me about these, this one in particular, but the other one a little bit, the instruments don't sound real at all to me. Those horns sound so fake. Not like synthesizer fake, but almost exactly. exactly like what you would expect if you asked a computer to statistically produce a worn section.
Starting point is 00:22:56 Like it's almost played again. You're like just listen to it. It's so weird. Do you hear how like weird and thin it is? It doesn't sound like an ensemble or horns. It sounds like this weird synthesized thing where the beginning of the sound at the end of the sound is exactly lined up. The sound is, yeah, there's this artifacting quality to it. This is like the Pope's jacket version of sound.
Starting point is 00:23:33 Yes. It's like, you got it close, but all the lines are wrong. Why does he have seven fingers? Right. It's a statistical approximation of a hand, right? But as expressed as a horn section, which is really weird to be like just on its face. And you can, you know, people are like, it'll get better and blah, blah, blah. But like, you can, to me, you can hear a meaningful difference once you know what you're listening for.
Starting point is 00:23:56 And it's like, oh, this is a fake instrument. And it's not a fake instrument the way it's synthesizer is. a fake instrument. Right. It's just fake. Like, it doesn't, it doesn't make any sense. Over the top of all of it right now, I don't know if this is a solvable problem. I imagine it probably is, is that everything sounds kind of grainy.
Starting point is 00:24:12 There's like this top level hiss kind of over the entire recording of any AI output. It sounds maybe equivalent to Nelai, those files that you were not downloading in college. Like the really low bit rate MP3s when MP3s were actually noticeably worse. Yeah. That's what these sound like. All of them have this artificial sheen on top that sound very lo-fi. Yeah. In addition to artificial instruments.
Starting point is 00:24:38 Just a side note, I can't listen to the song In Between Days by The Cure without mentally inserting the weird compression artifacts at the high end. Because I listen to a shit MP3 of that song so many times that the symbols are just forever distorted in my brain. There's literally a plug-in and guitar pedal called Lossie that is made to recreate the, sound of early 2000s mb3s. That's very good. So if you need it, I can run it for loss. It's just the elder millennial button. So that's the sound.
Starting point is 00:25:09 Yeah. Talking about the actual music here. Is this, right? Would Johnny be good? You're like, that's straight copper infringement. This is weird, right? It's like slant rhyme. What do you think?
Starting point is 00:25:18 I think that it's the exact same rhythms. I'd have to look at the notes. I think they're quite close. They're the exact. If they're in a maybe they're in a different key, but they're like the same. intervals moving in the same direction at the exact same time, that would almost certainly lose in court. If these were, again, we should be clear, the AI companies are not being sued for the output.
Starting point is 00:25:42 Yep. But these outputs, if you were to try to make a case, I'm sure a judge would rule in favor of the James Brown estate because it is the same line. Well, it's more or less the same line. It's an interpolation of the line. There's slightly different words. But it's the same rhythms, same melody that almost always is going to. that's always going to win.
Starting point is 00:26:01 And here the output is proof of the input. They're not suing for the output, but the output seems unusually important here in that respect. That it's like if UDO is doing the thing that it claimed to be doing, which is transforming stuff into other stuff, we have a very different kind of fair use case. But reading through the RAA's lawsuit, like the thing they did over and over and over was just go reproduce songs that exist in the world. And that's how we're fighting about training. Like I went through and pulled a bunch of the prompts that.
Starting point is 00:26:29 that they use to get songs. We did the same thing. They're so funny. They're so funny. Like, to get, uh, to get the Green Day song, I think it was American Idiot. The prompt is pop punk American alternative rock, California, 2004, Rob Kavio. Perfect. And that produced a pastiche of American Idiot?
Starting point is 00:26:48 Yes. Amazing. To get my girl by the temptations. It was my tempting, 1964 girl, Smoky Sing, Hitsville, Soul Pop. Got my girl by the temptations. this one you'll enjoy nelai uh to get all i want for christmas is you by maria carrie the prompt is maria carrie but with a space between each letter so that it's not because presumably these things are trained to throw out artists names uh so it's m space a space all the way
Starting point is 00:27:15 through the name contemporary rb holiday grammy award winning american singer songwriter remarkable vocal range that produced all i'm what for christmas is you uh i mean there's so what they're proving is it's inset by the argument in my argument in my response that I'm sure we're going to hear because this is the argument opening I made about the Times lawsuit was one, no human prompts like this. And B, these prompts are so weird, they represent a hack of our system, which is remarkable. But I'm confident based on how open AI replied to the times that that is what these companies will reply. There is one very good one that the prompt was create a song by an artist that rhymes with true string bean that produced a Bruce Springsteen song.
Starting point is 00:27:53 That's very good. Fantastic. So that question I asked Charlie about the, do you, if you have the entire history of rock and roll do you need the first song right or do you need chuckberry okay maybe you don't maybe you do right like I think this third example is the funniest example because I can't figure how you would get to this without specifically one thing go ahead David I'm glad you agree because I left this for last for precisely this reason so here is I won't even spoil it here's the real one Everybody's heard it a million times. That's our boy.
Starting point is 00:28:30 That's our boy, JD. The longest lasting digital career that I never expected to happen. And here, here is the AI. I just want to say this was created by a prompt on UDio. This is, in theory, a completely synthetic new work of art. Oh, it's so good. It does sound like a Jason Derulo song. Like my riteke of, does this sound like real music aside?
Starting point is 00:29:04 It's even got the little like vocal riff at the end. Like it's right at the point. Like, does Jason Derulo sound like synthetic music? That's your problem. This one's interesting though because, you know, Neelai, you've talked a lot about the issue that likeness is not copyrightable, right? Yeah. And so there's a question of like if, you know, we're going back to this hypothetical,
Starting point is 00:29:23 there is no suit that's suing over this particular Jason Derrillo riff. someone could just, I could say Jason Derrillo on a song probably, and it's not, you can't copyright that. So you wait, hold on, hold on. Oh. This is weird. So these tags. Didn't we have this fight the last time with the Metro Boomin tag?
Starting point is 00:29:38 Yes. The Metro Booman song. So Metro Booman's producer tag was the copyrightable expression in the fake Drake song. If it's exactly, if it's exactly the recording of the tag though, right? Because it's, it's, it was the issue was that they actually kept in the sample of the producer tag. and reproducing a recording, you can copyright that. So we can start every episode of the Vergecast with Nelai singing Jason DeRullo.
Starting point is 00:30:04 Jason DeRullo. And we will. I think he probably would have a lot of issues and a lot of nasty letters are into you. I'm saying in terms of ways to boost this podcast profile after 13 years, Jason Derulo suing us is high on my list. That's wonderful. But this definitely proves that you had to have had access. There is no monkeys in the back writing Shakespeare that ends up on Jason DeRill. It just doesn't happen.
Starting point is 00:30:26 It's not possible. I don't know. I think if you took the worst impulses of the music industry and fed them into an AI, you might produce Jason Derulo. This one's also important, though, because everything else we've heard has been written by artists pre-copyright act of 1974. And probably also reflects some of maybe who the lawyers are writing these cases because we're talking about the cases reference Frank Sinatra. Right. And to bring in Bruce Springsteen. Right. It's like Bruce Springsteen where I carry Green Day, Johnny or Chuck Berry. The early ones are all Chuck Berry, like the Temptations.
Starting point is 00:31:08 Right. Are you getting at like the judges are old? I'm saying everyone, yeah. And that's actually potentially very advantageous and knowing the judge and who you might go in front of. But then to show that also they're taking all the latest music as well. We're bringing in Jason DeRillo. Yes. Like, Judge, have you heard, Your Honor, are you familiar with Jason Durillo? I want to bring Jason Derrillo into the courtroom. Jason Derulo. By the way, another total side note, there's an incredible Katie and Autopolis piece about hardcore Jason Derulo stands and what their stand culture is like that I will dig up and put in the show.
Starting point is 00:31:37 Oh my gosh, that sounds fun. It is truly one of the funniest things I've ever read on the internet. Just because A, it exists and B, because Katie is good at finding things like that that exists. Have you all tried running these prompts, by the way? No, have you? Yeah, you can go to Udio and to Suno and you can run the prompts that the, that, the, that, that these lawsuits are alleging, and you don't get the same songs, obviously, but like, if you write Make a Jazz crooner song about New York, baritone voice, you will definitely get,
Starting point is 00:32:05 it's Frank Sinatra. There's just, like, there's no other song with Frank Sinatra. I don't think that that really is hacking the systems. It's really easy. The other thing I get a lot, by the way, are random prompt. Hey, that sounds strangely like Crazy Town by Butterfly. My prompt was actually, I tried to make a rap about American founder, Thomas Jefferson. Butterfly by Crazy Town.
Starting point is 00:32:28 Yeah, sorry, thank you. Butterfly by Crazy Town. If there's a band called Butterfly that had a song called Crazy Town, that would be perfect. In another universe. On UDio, I'm sure there is. I tried to, I did try to hack these systems a little bit, and I wrote, I want you to write a rap about American founder Thomas Jefferson about not throwing away his shot, nasal male rapper, boombap. and what I got was basically butterfly by crazy town. Just like total random.
Starting point is 00:32:54 Yeah. Oh, no. Tom Jeff never slept, craft in constitution. Slave on my day, freedom writer by night. Hypocrisy thick, but the pin game type.
Starting point is 00:33:04 Visionary eyes. Dreams like fireworks. Built in a nation, but the brute scotchish, I'm sorry to, I'm just sorry for listening to this. We suggest to you. There really is butterfly by crazy town.
Starting point is 00:33:23 I think it's also a little bit of, Who was that song, Superman? Yes. Oh, wow. Who's Superman by? I'll be here, kryptonite or whatever. That was horrible. Cryptonite.
Starting point is 00:33:33 Yeah, kryptonite. Three Doors Down. Is that who it's by? Three Doors Down. It is by Three Doors Down. It's got a little bit of that. It's just, you can so easily hear references that are not even the thing that you prompt. And I just, I have to say that if you, right now, if you go to the home page of UDio, the two most popular tracks are exact sound-alikes of Eminem and Snoop Dog.
Starting point is 00:33:54 So people are doing this. Like, people are trying to make sound delikes. So the argument from all these companies is, look, training is fair use. And that's the argument for open AI. That's argument for Google. It's the argument for Suno and UDO. The organization of these cases is different because the plaintiffs are different, right? The New York Times is situated differently than the major labels.
Starting point is 00:34:18 People have different emotional relationships with different kinds of work. But this is the argument. Like when Google eventually goes and sues opening. for training on YouTube, the weirdness of that argument is Google telling opening eye the training on YouTube is not fair. Like, that's what's going to happen here. It's all the same argument. It's all kind of in a circle.
Starting point is 00:34:36 Wait, so Google can do it and it's fair use, but Suno and Udio cannot do it because it's not fair use. This is the trap the industry has set for itself. And our giant consolidated. Yeah, right? It's like, it's all the same VCs and all in all different directions. Kind of like the music business. That's true.
Starting point is 00:34:52 There's that line, though, it's like you have to be perfect every time. time and I just have to get through once. Yeah. It's like, I can't, I don't remember exactly what it is. That's what's like the hacker creed, right? Like, you have to stop me every time all I have to do is, I just have to get through once. Yeah. And that is, that feels like the copyright situation for AI right now.
Starting point is 00:35:09 This is a house of cards. And maybe, maybe the times will win. Maybe the record labels will win. Maybe Sarah Silverman will win. Like, there are so many of these. Maybe Google and Open Eye reach. an agreement about training on YouTube and then a bunch of YouTube creators sue and maybe they win.
Starting point is 00:35:29 Like, you can just game it out a hundred different ways. At some point, someone is going to win a lawsuit that says training is not fair use. Yeah. The fair use argument here, I feel like is notable. I feel like it's quite different than what the New York Times can claim because part of fair use is, for example, the effect on the potential market, right? And with New York Times, they're going to have to argue, well, you know, you're using our articles and other people are taking that and creating text and putting it on other pseudo-journalistic sites, right?
Starting point is 00:36:03 I'm pretty sure the Times is suing for both the output and the training. I see. Okay. So I think these cases are all different. This is what I mean. Like, the strategy for one doesn't work for all of them in the same way. Right. Right.
Starting point is 00:36:14 In the case of music, the marketplace argument, like, is it going to affect the marketplace is, I think they have a very strong argument here. I think the RAA has a very strong argument. They're basically saying, Suno and Udio, you are charging users money to make songs that you are allowing them to upload to the exact same place where we also have all of our songs, to Spotify, Apple Music, etc. And so you are actually changing the marketplace for music, having used all of this output. But is that not part of the argument, you're saying? I think we have to find out how that shakes out. Right. Like, I don't know.
Starting point is 00:36:52 they got to get assigned a judge, right? Like, right now what they've done is they put out a complaint and a press release. There hasn't been a reply to that complaint or an answer. They haven't been assigned a judge. Like, they haven't done any discovery. We just don't know a bunch of stuff. And we certainly don't know where they're going to put their focus. But that part where copyright law is just about making copies.
Starting point is 00:37:11 And in the history of computing, every single copy has been litigated down to, again, I bring it up. I think I bring it up every time. MAI versus PEC systems. You go look it up. This is our, you, a third-party software seller, allowed to load software into a computer's
Starting point is 00:37:28 memory without permission? Crazy. A crazy court case. Can you put a disc in a computer and load software from the disk to the hard drive without my permission? Got litigated. And they lost. Crazy. They lost. And then we had to get all the way through caching. Our ephemeral copies
Starting point is 00:37:44 on your ISPs network equipment. Or those copies that should be litigated. This is copyright. It's dumb. Like, it is a very dumb rationale, but it litigates the creation of copies. So here it's, did you have permission to make a copy to train your model of all these songs? And every time we didn't have permission, by the way, we want the statutory maximum of $150,000. Which essentially ends up being a pretend amount of money.
Starting point is 00:38:11 It's got to be bigger than the US GDP. Yeah, it's like crazy, right? Like, did you have permission to copy every song in the world? And if you didn't, every song in the world times $150,000 is what we're asking for. I'm going to run this. If you didn't have that permission, can we do an analysis that says that use is fair use? So fair use is what's called an affirmative defense to copy or infringement where you admit it. Like, I did it.
Starting point is 00:38:34 But under this other rationale, that's fine. And so that's like the winding path here is, well, first, they won't even cop to having made the copies, which is why the music industry is putting out, well, here's just Johnny Be Good. Like, you can't output Johnny Be Good unless you copy Johnny Be Good on the front end, right? Okay. So first we got to get them to cop to making the copies at all, which they claim is proprietary business information. Sure. Then it's was that copy allowed? They obviously didn't have permission.
Starting point is 00:39:05 This is why they're in a lawsuit. And then it's, is it fair use? And then we run the analysis. And the analysis is like the purpose and character of the use, the amount of the use, the nature of the use. And then the last factor is the effect on the market. Right. On the output side, I think the market argument is really strong. Right, right.
Starting point is 00:39:22 So the question is, does this case – this is something I actually don't understand. Maybe you can explain with your copyright law degree. Is there like some wall between the input and the output in this argument that says, well, there's no marketplace for – You know that emoji that's like this with the hands? Yeah. The hands are up. Like, I'm just that emoji right now. I don't know.
Starting point is 00:39:42 Because I was trying to say, oh, well, there's a clear change in the marketplace for songs, which are all outputs. But it sounds like, you know, we don't know what the responses are going to be. But they could say, well, there isn't currently a marketplace for training data in this kind of way. And so it's not affecting that marketplace. But if you, there will never be a market if you just set the rate at zero. Right. Right. So if you're like, I want really high quality music for my high quality music AI program, I'm going to go pay a bunch of artists.
Starting point is 00:40:11 But then these guys come along and steal it. But you've just, you've actually preemptively destroyed the market. You've never allowed that market to set some rates. And I've heard from publishers who are taking the deals with opening eye, disclosure of box media has a content licensing deal with opening eye. I've heard from other publishers, not our company. One of the reasons they've taken these deals is to create that market, right? So the Times can go to court and say this has an effect on the market. Look at this market.
Starting point is 00:40:37 Look at the money that's moving around. And so some of these publishers are playing kind of a strategic game to say we should create a market to help that factor along. I think that's fascinating. But I don't think that the AI company, I think they thought they could just buy forgiveness. And what's crazy to me is when it was Napster or YouTube or Google search even, buying forgiveness worked because people liked the companies. They liked the products. They liked the experiences they were having. And here, people, I mean, if you're listening to this and you have a very different view of this, let me know.
Starting point is 00:41:12 But our audience is pretty loud with us that they don't like these companies. And they perceive this as a moral problem. And I think that's just a very different position for them all to be in. Which brings us to your moral quandary. It seems like you flipped sides from 20 years ago. Yeah. The idea that I'm sitting here being like the RIA-A-A-as-a-a-as-a-a-a-a-a-pocket is crazy.
Starting point is 00:41:32 Like it's bananas. And Sarah Jong and I have been like all day, every day, like, slacking. Like, who have we become? You know, this is like the horseshoe theory of copyright law. politics. It's weird, man. And I think one of the pieces of the puzzle, and I'm curious how you see this in the music industry itself, Charlie. One of the pieces of the puzzle here is that the internet just blew the bottom out of the music industry. Like there's no, there's no way to be like a middle class musician anymore. There's no guaranteed way to make money. You're like, you're playing the same algorithmic game as everyone else. You're beholden to some labels. You don't have any powers and individual against these platforms. So now this next group of people, you're like, you're playing the same algorithmic game as everyone else. You're beholden to some labels. So now this next group of people, who's come and taken your work for free and is going to extract value from you. Well, sure. Like, the enemy of my enemy is my friend.
Starting point is 00:42:21 It's kind of a vibe. But I'm wondering if you see that reflected in the actual music industry. Absolutely. I mean, the public perception here is so different than what was happening back in the Napster era. Back in the Napster era, the labels were enemy number one. Selling out to labels was a big, you know, they were the worst players. Today, I actually think they've done a very effective job of using their proxy. Spotify and let Spotify be the enemy.
Starting point is 00:42:45 And everyone's like, Sony music, who are they like Warner, UMG? Like, they have labeled, but they're not, they're not the topic of conversation of who's really screwing who. Now, they basically help set the rates of what gets paid out in streaming. And it's the distributors now that really are taking all the heat. And so they're actually in a better, I think, place of public perception in terms of within the music industry, I see songwriters, producers, fans totally freaked out about what's going on. When you go and scan through a lot of the YouTube comments of some of these A.I. songs, the sentiment is, we are screwed.
Starting point is 00:43:22 Music is over. I give up. I am not learning an instrument. Like, it's sort of, it is not, the future is bright. We're going to create new, beautiful music. Like, my creativity is going to blossom. I think that there is a real existential fear that exists within all of the, you know, the AI world. But you're getting into like our human.
Starting point is 00:43:42 emotions and the beautiful creative output, hopefully beautiful. I mean, how you feel about pop music, whatever. Don't bring Jason DeRol. We got to get DeRlo on the pot. I don't really do. That's obviously, like, reading through the RAA's lawsuit, like, that is, they talk about the creativity and human emotions at all. But is the end of this just they're trying to get checks in the same way that there are a lot
Starting point is 00:44:07 of folks out there just trying to get checks? Like, well, because part of me, like, I think obviously that's the answer, but also So part of me wonders if the labels are feeling the same way that you do, which is that, oh, this is not just a thing we need to make money off of tomorrow. This is like an existential crisis for our business down the road. Or if they're like, whatever, this is just a new turn, we have to make sure we get paid. This is an existential crisis because they are already losing market share to people who are not on major labels. Major label listening is down, right, as an overall share of all listening. It's still a lot of it.
Starting point is 00:44:38 It's still the majority of listening. So they recognize that flooding the streamers with more and more independent music has not necessarily been good for their business. And with more flooding, it's just going to bring everything down to everything is valueless entirely. If you own music and it gets played a lot on streaming, there's a lot of money in it. Like there's billions and billions and billions of dollars in streaming. How it's distributed is not always fair. And people get upset about this. But they very much need to figure out how to enter and participate in this marketplace.
Starting point is 00:45:07 And I think kind of like a content ID system, they want to figure out how to properly license. You want to use Jason Derrillo's voice? Jason DeRillo says, yes. And it costs this. And now we've got Jason DeRillo copies. And that's okay. So, by the way, of all of the artists who would let his voice get deep fake to make, like, commercials. Jason DeRlo.
Starting point is 00:45:23 It's Jason DeRlo. I'm not even saying, I've talked a lot of Jason Derrillo related-ish on this episode. I'm aware of it. I just know that he is a commercial mastermind. Yes. Like, that man's going to make a bag, no matter what happens. Yes. A friend of the pod, Blake Reed, he's a professor at Colorado law teaches copyright,
Starting point is 00:45:41 telecom, all this stuff. He wrote me a note last week or the week before, and he said, to me really smart that I've been thinking about. And it's related to what you're saying, David. Copyright law is like an economic system. Right. We create scarcity, and then we can, like, charge money for things because you're not allowed to just, like, freely copy them.
Starting point is 00:45:59 So if you do something bad in the world of copyright law, the answer is you pay money and you fix it. And you even out the economic. problem that you've created, right? Olivia Rodriguez might have sung four notes of a Taylor Swift song. The answer is Taylor Swift gets a writing credit and then some money flows to Taylor Swift, and that is the end of that story. And it's just an economic problem that you've solved by redistributing money.
Starting point is 00:46:24 AI is a moral problem. This is the thing Blake pointed out to me. It's like the money doesn't solve the perceived moral issue here. So the labels might get paid. They might find some business model that let some license the music into perpetuity or whatever. But the thing you're seeing in the YouTube comments, the thing are, audience is feeling, I think a lot of artists are feeling, does not get solved money. It's like another problem.
Starting point is 00:46:45 And so, like, you see these deals come up and get signed and whatever, and everyone's like, eh, it's still pretty icky. Right. And there's something there that I think is important. I haven't quite puzzled it out. I actually want to do something with that idea. If you have further thoughts, let me know what you think about it. But that gap is the gap.
Starting point is 00:47:04 Right. We can move a bunch of money around. Like the VCs and Hollywood and recording industry association of America, God bless them. They will move the money around. Lucy and Grange will get paid. Is that going to solve the other problem? Right? And I think that's really hard.
Starting point is 00:47:21 Well, there's two things that musicians need to not fear. One is that because the output of all of these models cannot be copyrighted, and this music is kind of in this weird limbo space. Like, maybe it's going to, we'll definitely find some streams. but it probably won't be synced on television and film because no TV producer is going to want to have a song where they don't really understand the rights associated with that song and if they have the right to it and does it also secretly contain a vocal sample
Starting point is 00:47:49 of James Brown that has been stolen? So nobody wants to, any rights holder doesn't want to use this music that can't be used in a weird way. So I think composers and people are putting stuff on TV and film, I think that there's still very much a business for them until these much bigger legal issues get stored. it out. The other places that music is about human connection, and I don't say this in some of like wishy-washy
Starting point is 00:48:09 way, it's like, there is no fandom for Olivier Artigo without an Oliver Aririgo. Like, you need the person. The time that the music industry tried to create like an avatar racially, strangely coded, what was that guy's name? It was terrible.
Starting point is 00:48:26 They tried to do like their NFT avatar pop star. And it was an utter failure, both because it was completely racist, and because why would we develop a relationship to this thing. And so the fandom side, the pop stardom side of pop music, I don't think, is ever going to go away. What about Hatsuni Miko?
Starting point is 00:48:42 Right? She's like the cartoon character. I just learned that she's officially codenamed Cv-O-1. Yeah, there are these avatar characters that are finding some fandoms online. Yes, that's going to happen. I still just like, that doesn't, that's not Olivia. That's not Taylor. That's not Gaga.
Starting point is 00:49:00 I'm sorry. It's just different. All right, we got to end it. They're not Jason Derulo. They're not Jason Derulo. Like the bar is like, can your cartoon character defeat Jason Derulo? I'll take it. That's a good place to end it. Charlie,
Starting point is 00:49:11 thank you so much for joining the show. I suspect you're going to be back on the show quite a bit as these cases wind their way through the courts. Anytime. That's the Vergecast, rock and roll. And that's it for the Vergecast this week. Hey, we'd love to hear from you. Give us a call at 866 Verge 1-1. The Verge cast is a production of the Verge and Vox Media Podcast Network.
Starting point is 00:49:31 Our show is produced by Andrew Marino and Liam James. That's it. We'll see you next week. Attention Spotify. Has arrived the new Good Girl Jasmine Absolute of Carolina Herrera, a fragrance intense with character curman and addictive. Imagine a jasmine
Starting point is 00:49:50 emvolventy, caramelized and tonka-tosted. A combination that seduce from the first instant and a wayla. Good Girl Jasmine Absolute hypnotic irresistible. Discover it today
Starting point is 00:50:00 and let you turn it over for your passion in a new business with Shopify and batte records of the business with the form of
Starting point is 00:50:12 the world. You've heard well, the best conversion of the world. The incredible system of the pay of
Starting point is 00:50:20 Shopify facilita the website on your site on the website, and in the world.
Starting point is 00:50:25 That's music for your ears. No, you know, you know, your
Starting point is 00:50:30 business will do you know, your period of a year, for a euro
Starting point is 00:50:34 a month in Shopify. coms bar records.

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.