Today, Explained - Fake Drake

Episode Date: April 28, 2023

The Verge’s Nilay Patel explains how a spurious collaboration between Drake and The Weeknd started a copyright fight over generative AI. Holly Herndon introduces her AI alterego, Holly+. This episod...e was produced by Amanda Lewellyn, edited by Amina Al-Sadi, fact-checked by Avishay Artsy, engineered by Paul Robert Mounsey, and hosted by Sean Rameswaram. Transcript at vox.com/todayexplained  Support Today, Explained by making a financial contribution to Vox! bit.ly/givepodcasts  Learn more about your ad choices. Visit podcastchoices.com/adchoices

Transcript
Discussion (0)
Starting point is 00:00:00 When Drake first started getting attention back in 2009, I was skeptical. This guy was on Degrassi, now he's a rapper. But then Drake got undeniably good. But after a while, it felt like Drake stopped caring. Like maybe he was just using a bunch of templates to write his songs. And now, maybe machines. If I had left, she know what she need. All I need, all she bless. The biggest Drake song in the country, as far as anyone can tell, wasn't written or performed by Drake.
Starting point is 00:00:54 How that happened, coming up on Today Explained. That's my stack. Bet you I made the bid so you know that it's going slow. BetMGM, authorized gaming partner of the NBA, has your back all season long. From tip-off to the final buzzer, you're always taken care of with a sportsbook born in Vegas. That's a feeling you can only get with Bet MGM. And no matter your team, your favorite player, or your style, there's something every NBA fan will love about Bet MGM.
Starting point is 00:01:22 Download the app today and discover why BetMGM is your basketball home for the season. Raise your game to the next level this year with BetMGM. A sportsbook worth a slam dunk. An authorized gaming partner of the NBA. BetMGM.com for terms and conditions. Must be 19 years of age or older to wager.
Starting point is 00:01:40 Ontario only. Please play responsibly. If you have any questions or concerns about your gambling or someone close to you, please contact Connex Ontario at 1-866-531-2600 to speak to an advisor free of charge. BetMGM operates pursuant to an operating agreement with iGaming Ontario. This is an artificial intelligence version of Drake and UL is named Toto the X-Player. Nilay Patel is the editor-in-chief of The Verge and the host of The Decoder podcast and... I was a very bad copyright lawyer before I became a journalist. Which weirdly makes you the perfect person to talk about the intersection of AI and the
Starting point is 00:02:28 most popular Canadian R&B goof nugget in the world, Drake. That's a great title for Drake. Tell us where this song came from. I don't know. That's actually a very fishy part of the whole story. So a couple weekends ago, a track blows up on TikTok called Heart on My Sleeve. It is perfectly engineered to go viral. It is an A.I. Drake X The Weeknd collaboration. These two artists have not collabed in forever, so it meets audience demand.
Starting point is 00:03:08 The person doing the track is wearing a bed sheet and glasses. They call themselves Ghostwriter. It has some songwriting to it. The person claims that they've been an industry ghostwriter, an actual songwriter for the industry. So it is well-crafted. Importantly, it is absolutely not clear what role AI has played in making the song, except for synthesizing Drake's voice.
Starting point is 00:03:32 So we don't know if AI made the beat. We don't know if AI wrote the lyrics. We don't know if AI mixed and mastered the thing. We just know that at some point in the process, they synthesized Drake's voice, probably against somebody actually rapping actual lyrics. I came out with my ex like Selena to flex, eh.
Starting point is 00:03:50 Bumpin' Justin Bieber to favor and love, eh. She know what I mean. And that, I think, is where people have gotten particularly confused. It is not as though someone typed, make me a Drake in the weekend song into chat GBT and this thing was returned. It is a complicated songwriting process, and the last step is having an AI re-sing a song for you, basically. There's been a huge community of people making AI music using artist voices for quite a long time. My favorite is a cover of Hey There, Delilah using Kanye West's voice. It is adorable.
Starting point is 00:04:35 Hey there, Delilah, what's it like in New York City? I'm a thousand miles away because tonight you look so pretty. Yes, you do. Who wanted that to exist? No one. And then it does. You're like, oh, Kanye should just stop whatever he's doing and just cover mid-2000s guitar rock. That would be great for him. Mid is the operative word there.
Starting point is 00:05:00 Yeah, exactly. But these songs have existed for a long time and they haven't gotten viral in this way. So this one going viral and the way it went viral and getting the attention it immediately got and being sort of crypto adjacent, right? The person had links in their bio to a service that was run by crypto people. All of it smells weird, but then the conversation it provoked is very important. So this is where copyright lawyer philosophy, Nilay, comes out. Ah. I apologize. I pre-apologize for it. Nice to meet you, sir. The collision between computers and music is the story of our culture entirely.
Starting point is 00:05:46 Whatever happens to the music industry because of computers happens to everything else five years later you can just see it over time right the ipod came out the music industry was overturned many years later you could download movies music moved to streaming the whole industry was overturned many years later we're all streaming our tv and the streaming services are full bore competition and the thing that is repeated over and over again is that it becomes easier to make music. It becomes easier to copy the music. It becomes easier to distribute the music. And there's not any scarcity left, right? So when you had CDs, they were scarce.
Starting point is 00:06:17 You had to go and manufacture some amount of CDs for some amount of money before you could put them in stores and have people buy them. And so there's natural scarcity to songs and music. You move to digital, now your cost of duplication is zero. The industry gets upended and you can't stop kids from doing it. You get to streaming, the cost of distribution falls to zero. You're not even bundling things into a 99 cent song. You're saying it's an all youat buffet, and we'll do some complicated math on the back end. The last piece of scarcity has been the artists themselves. Taylor Swift is a scarce resource. Economically, there's more demand for Taylor Swift than she can ever make supply.
Starting point is 00:06:56 And now this AI has come along, and it's like, oh, we have an infinite supply of Taylor Swift. We have an infinite supply of Drake. And that fully upends the economics of the industry again, potentially in an existential way. So how are the stakeholders here responding to this song? Another extremely fishy part of this whole thing is he hasn't said anything about it. Oh, Drake. He just knows. He knows how social media works.
Starting point is 00:07:29 He knows that he could just stay out of it and it'll feed the beast. It'll come to some resolution. But he said something about another AI track a couple days before this hit. So Drake has reacted to an AI version of his voice rapping an Iced Spice song. He posted a screenshot of the post to his Instagram stories and wrote,
Starting point is 00:07:46 This is the final straw, AI. How is Drake's label responding? Yeah, so Drake's label is Universal Music Group. And I would say its responses have been equally weird. So first, Universal just released a statement about generative AI. And in that statement, the label said, look, you're on the side of the artists or you're on the side of the technology. Everyone is going to have to choose. And then very importantly, there's a line that said use of our songs to train
Starting point is 00:08:16 these generative AI tools is a violation of our copyrights and other agreements. So this is Universal laying out its position and saying, if you take Drake's catalog and you copy it into one of these programs to train it to make Drake's voice, you have now infringed our copyrights. Huh. But they don't know for sure that's what happened here. They don't know for sure that's what... For all we know, this is some kid who does a great Drake impression. So does that make this song legal or in some murky gray
Starting point is 00:08:47 area? It is absolutely in a totally murky gray area. Universal Music Group was able to go and issue takedowns to YouTube and TikTok and other services. But what they were able to claim in order to get the song taken down was a sample of another song that they own. What's that? There's a very famous artist called Metro Boomin. When he produces a track, he has what's called a producer tag at the top of the track. It's, if young Metro don't trust you, I'm gonna shoot you.
Starting point is 00:09:15 The rapper Future says it. If young Metro don't trust you, I'm gonna shoot you. The A.I. Drake song, Heart on My Sleeve, has that tag at the top. Aha! If young Metro don't trust you, I'm gonna shoot you.
Starting point is 00:09:28 So Universal was able to say, hey, that's ours. For sure, that's ours. What copyright law regulates is making copies, right? This is an ancient law. Copyright law is in the Constitution, right? This is as old as it gets. And it is built up very much around old ideas of how you distribute information, which is you have to make copies of it. And as we've gotten more and more into the digital age, we've gotten farther and farther from a one-to-one relationship between a song and a copy of the song, a book and a physical copy of a book. And everyone wants to skip ahead to, does Drake own his voice? Every copyright lawyer is all the way back down here at, where's the copy. And so the copy here is the copy of the Metro Boomin tag. It's the sample. You've made a copy of it. The copy is you've copied all of Drake's songs into your computer to train an AI. The copy is not,
Starting point is 00:10:17 oh, his voice. Because his voice is not copyrightable in that way. And there's a big disconnect here in the law because no one who ever wrote United States copyright law thought that we would get to a place where you could just take someone's voice. How does this evolve? Is this the beginning of a very big problem for groups like Universal Music, for artists like Drake and Taylor Swift and The Weeknd? I think this is the beginning of a historic split between the creative communities and the tech platforms. And what I mean by the creative communities, it's not just the record labels, but Hollywood, publishers like our own, authors. If you can
Starting point is 00:10:58 just take all of our work and copy it onto your computer and then say, it's totally free for me to train an AI to generate more work that seems like you and trade on your likeness, trade on your trademarks, trade on your style. The law isn't there yet. I think there's a lot of questions to be answered. But if you just went and asked a normal person, does that seem fair to you? Does that seem appropriate that there's no compensation anywhere in that chain of events? Most people instinctively are going to say, yeah, that doesn't smell right. And I think that's the split because the tech platforms, Microsoft, Google, OpenAI, they're all claiming that this copying of training data is, quote unquote, fair use,
Starting point is 00:11:42 which is a very complicated part of copyright law that is one intended to be adjudicated on a case by case basis. So every lawsuit is not supposed to create precedent. It's just supposed to be case by case. And then also once it hits the courts, every fair use lawsuit is like a throwing darts at a board. You have no idea what's going to happen. And often the courts get it wrong. So the tech platforms are very confidently saying this is fair use. And everyone who knows about fair use is saying that's not how that works. And then the creative communities are saying, who cares? Pay us. You can see that this is all just going to run right into each other. What about all like the non-Drake level
Starting point is 00:12:22 artists out there, Nilay, of which there are many and they're far more important perhaps because they don't already have piles and piles of money? How does this development in this technology affect them? I think historically what we've found when technology disrupts the high end of the market, it's because the people who don't have access to those tools or don't have all that money are suddenly able to do what the high end of the market, it's because the people who don't have access to those tools or don't have all that money are suddenly able to do what the high end of the market is able to do. So that class of artists you're talking about, they don't have access to all the money that Drake has. They don't have access to all the infinite songwriters that are on every pop hit now. What they do have access to is, oh, I can write some bars and then I can make it sound like Drake rapped them. And then maybe I can use that to get some attention to myself. Maybe one
Starting point is 00:13:11 day I can have access to more instruments that sound differently than the ones that I'm using right now. And you just see how the bottom end of the market is just more excited about the tools because they just have less to lose. We're going to hear from a musician who is thrilled to be incorporating AI into her work in a minute on Today Explained. Support for Today Explained comes from Aura. Thank you. photos and videos directly from your phone to the frame. When you give an AuraFrame as a gift, you can personalize it. You can preload it with a thoughtful message, maybe your favorite photos. Our colleague Andrew tried an AuraFrame for himself. So setup was super simple. In my case, we were celebrating my grandmother's birthday and she's very fortunate. She's got 10 grandkids. And so we wanted to surprise her with the Aura Frame. And because she's a little bit older, it was just easier for us to source all the images together and have them uploaded to the frame itself.
Starting point is 00:14:35 And because we're all connected over text message, it was just so easy to send a link to everybody. You can save on the perfect gift by visiting AuraFrames.com to get $35 off Aura's best-selling Carvermat frames with promo code EXPLAINED at checkout. That's A-U-R-A frames.com promo code EXPLAINED. This deal is exclusive to listeners and available just in time for the holidays. Terms and conditions do apply. The all-new FanDuel Sportsbook and Casino is bringing you more action than ever. Want more ways to follow your faves? Check out our new player prop tracking with real-time notifications. Or how about more ways to customize your casino page
Starting point is 00:15:13 with our new favorite and recently played games tabs. And to top it all off, quick and secure withdrawals. Get more everything with FanDuel Sportsbook and Casino. Gambling problem? Call 1-866-531-2600. Visit connectsontario.ca. Holly Herndon is an artist based in Berlin. She makes music mostly with her computer, and she generously made that version of our theme song you just heard.
Starting point is 00:15:56 So I've been making music since I was a child, but when I went to college, I decided to start using the laptop as my primary instrument, and that was a little bit weird at first because a lot of people didn't consider a laptop an actual instrument. But I realized that it was this very personal instrument that knows more about me than any other instrument before. So I tried to kind of harness some of the newness
Starting point is 00:16:18 that this tool slash instrument could enable in music making. I was trying to find a way to make laptop performance interesting because a lot of people would say, oh, you know, if you're playing your laptop on a stage, I can't tell if you're checking your email or not. Like, how do I tell if you're actually playing an instrument? And so I started singing using my voice as a kind of data stream, but what would come out the other end
Starting point is 00:16:42 often didn't sound like a human voice. It would sound like whatever digital processing I had kind of put on my voice. I always say that's my biggest attribute is that I'm a mediocre vocalist so that it forced me to create all of these really unique digital tools to make my voice really unique and interesting. So that's kind of how that all started. And this is how Holly got into making music with AI. I'm still doing a ton of digital signal processing, but when I learned about machine learning that you could kind of create the model, create a model of a sound, for example, my voice, then I realized, wow, okay, then I could ask this voice to kind of perform things that my physical human voice couldn't normally do. So I'll also
Starting point is 00:17:33 play you an example of her singing in Spanish, Bésame Mucho by Velazquez. It can sing, you know, really complicated musical passages in multiple languages that I don't have the vocal training to accomplish. Thank you. I can't speak Spanish, by the way. Sorry, you have an AI voice. Tell me more about that. I do. Yeah. So I've been working or experimenting with AI since around 2016. Wow. So like way, way longer than Drake and The Weeknd. Way longer than Drake and The Weeknd. Yeah. I mean, I've been, I, yeah, I kind of, I gave a TED talk in 2022 about this kind of like vocal impersonation or as I like to call it, identity play. Could someone else go on tour as me with
Starting point is 00:18:25 my permission? Could I be in a thousand different bands in multiple languages? And what would that even sound like? But yeah, when I started experimenting with AI around 2016, I realized pretty early on that your model is your data. So it really makes a big difference what you train your AI on. And I really only felt comfortable training my AI on my own voice because that was the IP that I had access to or on people who I could get consent from. So I put together a small choir,
Starting point is 00:18:59 small ensemble, and we collectively trained an AI together and we performed with the AI. Wow. Yeah. And so I have a newer version now called Holly Plus. Like Disney Plus? I don't know what Disney Plus is, but I imagine that's a streaming service. Okay, no, not like Disney Plus, like a digital version of me.
Starting point is 00:19:32 Holly Plus is essentially an AI model of my voice. There are actually multiple models. Some of them are really natural sounding with my, you know, that can enunciate language and others are kind of more just a kind of textural audio. I also did a cover of Dolly Parton's Jolene that we made a video for. You can find on YouTube where my AI voice is singing one of the greatest songs of all times.
Starting point is 00:19:58 Wow. Wait, I want to pull it up right now. Is that okay? Yeah, I mean, I... Is it Holly plus like plus sign or Holly plus like PLUS? Plus sign. I got it. I'm going to play it. Wow. Cool video right out the gate. What is going on? This is my friend's team Rolfes. They created the video using a lot of machine learning, you know, models trained on East Tennessee imagery where I grew up because we were kind of creating a kind of fake past for myself.
Starting point is 00:20:31 Dolly Parton is a huge, I don't know what to say, her figure looms large there. She's a national treasure, but even more so in Tennessee. Exactly. Yeah, like if I got all A's on my report card, I would get to go to Dollywood kind of thing. This sounds great. So this isn't you singing. This is AI that's been trained on your voice. Exactly. So that's the kind of really cool thing about this is the term that I came up with with my partner for this is called spawning, teaching the AI sonic properties of one sound to create a new sound. So this is what allows HollyPlus to create like a really wide range of vocals that I did not sing based on a set of pre-recorded phrases that I did sing. So it's kind of teaching the sonic logic of a sound. As I always do with any YouTube video I'm watching, I scroll down to read what people are saying about it. Oh, no.
Starting point is 00:21:53 And it's just near universal love. People are saying stuff like, I'm super excited about this surprise drop, remarkable, unique, and fresh take on a classic. But then there's this one i found oh no that's maybe the most interesting it says your voice is incredible and then someone replied it's not a real voice it's all ai and then the original person replied, no, it's not. And then the other guy replied, yes, it, look, it, foo. I don't know what that means. But there's a debate. People don't know if it's real or not.
Starting point is 00:22:35 I think this gets at something I wonder about when I hear there's a new AI Drake song. It's like the question becomes to me, like, does it matter? Do you like it or do you not like it? Yeah, I mean, I think for a lot of people, it doesn't matter. It kind of brings into question like, what's the role of the artist? You know, what aspects of art creation are okay to outsource or can become kind of a mechanical process? And this is a question that's not necessarily new. I mean, Warhol's been asking this, you know, decades ago. And so it's kind of a continuation of that question. I think the big point here is about consent. If people are going to be performing with your voice or your likeness, it would be really great if there's a system for the artist or the voice that's being
Starting point is 00:23:25 used to be able to consent or to be able to tell if certain works have been approved by that artist or if certain works are kind of more a part of fandom so that you can differentiate the kind of artist's voice from the automatically generated voices. Well, let's bring this back to Drake and The Weeknd then. It sounds like if this is all about consent, you would not approve of some internet artist, whoever it might have been, taking Drake's catalog, feeding it through some AI and coming up with some fresh Drake tracks, if that is indeed what happened here. You know, in an ideal world, Drake and The Weeknd would have their own models that they could make accessible to the public. And then you could see which works Drake and The Weeknd maybe approved or which works were kind of just like freely works in the wild. But ideally, the artists whose voices are being used would have some kind of say in it. I do think it's one of those kind of things. It's like,
Starting point is 00:24:31 it's going to be really difficult to control. So I'm not really in favor of a kind of punitive IP debate around this. I think that it will be more interesting if people are more permissive with their IP. That can actually unlock really fun and cool stuff. Like, what if The Weeknd and Drake actually really loved the song and then decided to then add a verse to it, and then that could go on their official Spotify playlist or whatever. Like, what if there was some sort of interaction there, and then they could interact with their fans, who could perform as them in certain ways. I think it can unlock really weird and interesting collaborations.
Starting point is 00:25:06 Of course, the devil's in the details. That was Holly Herndon. Her AI alter ego is Holly Plus, kind of like Disney Plus. Find them both at hollyherndon.com. Earlier in the show, you heard from Nilay Patel. Find him at theverge.com. He's also got a podcast called Decoder. If you're into tech podcasts, you'll be into Decoder. Our show today was produced by Amanda Llewellyn. It was edited by Amina Alsadi. It was fact-checked by Abhishek Artsy and mixed by Paul Robert Mounsey. The rest of the team at Today Explained is comprised of Halima Shah,
Starting point is 00:25:45 Hadi Mawagdi, Miles Bryan, Victoria Chamberlain, Siona Petros, and Laura Bullard. Happy birthday, Laura. Extra help from Jolie Myers and Patrick Boyd and Michael Rayfield. We use music by Breakmaster Cylinder and Noam Hassenfeld.
Starting point is 00:26:01 Matthew Collette's our managing editor. Noel King hosts the show with me. I'm Sean Ramos for him. Today Explained is distributed by WNYC. We are part of the Vox Media Podcast Network. you

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