The Ballad of ‘Deepfake Drake’

[0] From the New York Times, I'm Sabrina Tavernisi, and this is The Daily.

[1] When an anonymous music producer used artificial intelligence to impersonate Drake earlier this month, it jolted the music industry.

[2] Today, my colleague Joe Koskarelli on the rise and fall of one AI song and what it means for the future of music.

[3] It's Friday, April 28th.

[4] Joe, welcome to the show.

[5] Hey, thanks for having me. So you are a music reporter.

[6] That's the world you cover.

[7] But this is an episode about artificial intelligence.

[8] And artificial intelligence, AI, is something we've covered a lot on the show, but never when it comes to music.

[9] So tell me how these two things came together for you and your recent reporting.

[10] So there's obviously been a lot of buzz about AI in recent months.

[11] all across all my social media, you know, television, cover of our newspaper, chatbots and image generators.

[12] You know, I'm thinking of chat GPT.

[13] Everyone's saying they're the next big thing.

[14] What do they mean for various industries?

[15] But in my world, you know, the people who make popular music, fans of popular music, and the people who build businesses around pop music, AI hadn't really been a thing.

[16] There had been some experiments with AI on the fringes of the music business over the years, but they've always seemed a little bit academic or niche.

[17] They weren't really touching mainstream culture.

[18] You know, they weren't used by the biggest producers or it didn't involve the biggest artists.

[19] And these are the people that I tend to cover to write about and to listen to.

[20] But then over the last few weeks, there was this massive change.

[21] It started, like most things in pop music, on social media.

[22] I started to see videos on my feeds on TikTok, on Instagram, on Twitter, and it just really felt like a moment.

[23] And what I was seeing mostly were novelty.

[24] Somebody once told me the world is go to roll me. There are memes.

[25] They are covers, essentially, of familiar music, familiar songs and artists, very popular ones, but computer -generated.

[26] You know, a Beyonce song, Cuff It, as sung by AI Rihanna.

[27] Or Kanye West Side, what's it like in New York City, I'm a thousand miles away.

[28] Or Kanye West doing Hey There Delilah by the Plain White Tees.

[29] Hey there, Delilah, what's it like in New York City?

[30] I'm a thousand miles away because tonight you look so pretty.

[31] Yes, you do.

[32] These really strange combinations, but ones that people were obviously taking notice of, because who wouldn't want to hear their favorite rapper, rap a song by their other favorite rapper.

[33] It's just, you know, it's a strange mashup.

[34] So kind of like a deep fake buffer music.

[35] Exactly.

[36] These AI vocal technologies have gotten better and better at generating human voices, especially voices we're familiar with, like Donald Trump and Joe Biden.

[37] Hey, guys, can we talk for a minute?

[38] I'm feeling a bit overwhelmed with everything going on.

[39] Of course, Joe.

[40] What's on your mind?

[41] It's just that the weight of the presidency can be a lot to bear.

[42] The decisions I make impact the lives of millions of people, and I feel like I'm constantly under a microscope.

[43] I know exactly how you feel, Joe.

[44] There were times when I felt like the media was always looking for a way to bring me down.

[45] Because these vocal models are trained from actual recordings, they're extremely accurate when it comes to people who we hear from a lot.

[46] Because there's so many recordings.

[47] Exactly.

[48] And it sort of drew people's attention to the possibilities that were starting to become available from this AI technology.

[49] And people could do this at home on their own laptops in only a few minutes.

[50] Right.

[51] So as people like having fun, just kind of like spoofing or kind of covering, using their computers, their favorite songs, using voices from their favorite artists from other places?

[52] Yeah, like what would it be like if Nikki Minaj did a Frank Sinatra song?

[53] What was it like?

[54] I haven't heard that one, thankfully.

[55] Okay, so it's fun, but it's kind of a novelty.

[56] Yeah, you would see these things pop up on Instagram, on TikTok, on Twitter, and they were lightweight, they were harmless, but they sort of foretold all of this potential for how this technology could be used in the music industry.

[57] And then a couple of weeks ago, in the middle of April, I'm lying in bed on a third.

[58] Saturday morning, scrolling Twitter, and there was one song that just wouldn't go away.

[59] And this song, called Heart on My Sleeve, claimed to use AI Drake and AI The Weekend to create an entirely new track.

[60] And over and over again, I'm seeing, oh, my God, this AI Drake song is so good, it's basically a Drake song.

[61] You know, I like Drake as much as the next guy.

[62] And so I click play And You know It has all the touchstones there Of what a Drake track sounds like It has a little Piano beat That's not great But sounds maybe like Drake would rap on it The lyrics right away Are sort of using Subject matter That would be be in a Drake song that's on the radio.

[63] You know, he's referencing.

[64] Famous friends of his and the weekends.

[65] He's copying Drake's cadence, his flows, the way he delivers lyrics.

[66] And the song is called Heart on My Sleeve, which is a sort of nod to what most people know about Drake, which is that he's maybe more emotional than your average rapper.

[67] And then an AI version of The Weekend starts singing in a replica of his trademark falsetto.

[68] And the collaboration is also excited because Drake and the weekend used to make a lot of music together and they no longer do.

[69] So the fans, this was like wish fulfillment in some way for the fans.

[70] And all told, when you put it all together, It's not so different from songs that Drake and the Weekend have actually made together.

[71] And who made it?

[72] Who created this thing?

[73] So it seems to have been seated across social media and streaming services like YouTube and Spotify by someone calling themselves Ghost Writer.

[74] Ghostwriter is a term in the music industry for someone who works in the shadows in studios, helping big artists come up with their tracks and then not really getting credit for it.

[75] So this is interesting.

[76] But it's so, so this is different from the Kanye stuff, which was essentially, you know, regular people having fun on the internet with a bit of AI doing spoofs and cover songs.

[77] This is actually someone who knows what they're doing, actually composing real music to sound using AI, like one of the world's most famous artists.

[78] Yeah, what seems to be happening here is that there's a songwriter with enough skills in the studio to write.

[79] a Drake song that is then using AI technology to make it sound as if Drake is the one performing it.

[80] So do people know it's not Drake?

[81] Most people know that it's not Drake.

[82] And part of the novelty factor is that it's being presented as AI.

[83] But I think there's enough gray area that people want to hear it for themselves.

[84] And it's a good enough copy that the song starts to snowball and becomes really popular.

[85] By the end of the weekend, it's gone totally viral.

[86] and millions of people have heard it across TikTok, Twitter, Instagram, and even Spotify and YouTube, where official streams are counted.

[87] If it would have continued on this pace, it could have very easily made the Billboard Hot 100 right next to a real Drake song.

[88] Wow.

[89] Huh.

[90] Interesting.

[91] So really went viral and has this huge pickup.

[92] Yeah, I think there's a combination of factors.

[93] It's people who think it's a good song.

[94] And we'll take a new Drake song, whether or not he actually made it, and also people who want to hear just how far this AI technology has come.

[95] So then what happens?

[96] All of this plays out over a weekend, which really gives the song space to grow and for its virality to spread far and wide.

[97] But by Monday morning, the corporations involved have seemingly had enough.

[98] So Universal Music Group, the biggest of the three major labels, and the one that represents, both Drake and The Weekend, gets this song pulled down from all the official streaming services like Spotify and YouTube.

[99] Huh.

[100] Okay.

[101] How do they do that?

[102] So it's interesting because the legal questions around AI music are still very much unsettled.

[103] It's a real gray area, whether or not impersonating someone's voice is enough to get a song taken down.

[104] In the past, there have been some court cases that say artists have a right to their likeness to the sound of their own voice, especially if a copy is being passed off as legitimate.

[105] But we haven't really seen this legal precedent applied to AI.

[106] So in this case, it seems like they were able to get the song removed, not because the song is impersonating Drake, but because the producer, Ghostwriter, included a tiny clip of another song that was actually already copyrighted.

[107] So that allowed the companies to use their normal process for when they tried to get a song taken down.

[108] Interesting.

[109] So, in fact, it had nothing to do with AI.

[110] their ability to take it down.

[111] That's right.

[112] It seems like this was just a bit of an unforced error by ghostwriter to include something that was copyrighted instead of something that was just in this gray area.

[113] But it sounds like in any case, the party was over.

[114] Yeah, for this particular song, at least.

[115] But I think overall, it was clear that a line had been crossed.

[116] And the industry, which had never really had to contend with the idea that AI was a threat to their business, had to pay attention because they knew more and more songs like this would keep popping up.

[117] So while heart on my sleeve may have been gone, the legal and creative questions that it raised were obviously here to stay.

[118] We'll be right back.

[119] So Joe, Drake's label, Universal, gets Spotify to take the song down.

[120] How are they thinking about this stuff?

[121] Universal didn't address this song specifically.

[122] They never mentioned Drake or The Weekend, and they certainly didn't want to give any attention to Ghostwriter.

[123] But they do release this statement the day the song gets taken down, and it's this big, sweepy, corporate thing that tries to frame this fight as, you know, a cosmic battle between good and evil.

[124] The statement basically asks people in the music industry and who might use AI which side of history they want to be on.

[125] The side of artists, fans, and human creative expression are on the side of deepfakes, fraud, and denying artists their due compensation.

[126] Okay.

[127] So when you put it like that, that seems pretty straightforward, right?

[128] I want to be on the side of artists and human expression.

[129] Yeah, I think Universal sees this from a couple of different angles.

[130] They don't want random music producers using Drake's sound and his celebrity to make money.

[131] They want to be making money off of Drake.

[132] They also don't want these AI tools to be able to learn from the music that Drake and the weekend and their ilk have already put out.

[133] That's copyrighted material.

[134] But can they even control that?

[135] I mean, tell, you know, the computers of the world that they can't scrape the internet to listen to Drake songs because the Drake songs are copyrighted?

[136] I mean, they're just out there, right?

[137] That's true, but they can try.

[138] A lot of people have compared the rise in AI music to file sharing, to Napster.

[139] And these were issues for the labels back in the day where they got caught a little bit flat -footed.

[140] Every time there's a rise in new technology, the labels want to find a way to get in on it.

[141] But in the meantime, they have to sort of play whack -a -mole with people doing it on their own.

[142] So that means getting their partners like Spotify and YouTube and Apple Music to take down songs that come from copyrighted material.

[143] But at the same time, they're trying to work with these AI companies behind the scenes and say, your machines can learn from our catalog, you just have to pay us.

[144] And the other thing that labels like Universal are doing is trying to invest in their own AI innovation.

[145] So they're okay with using AI as long as it's authorized.

[146] Huh, interesting.

[147] So they're not just going around trying to get stuff taken down.

[148] They're also trying to harness this power.

[149] That's exactly right.

[150] So tell me about that, Joe.

[151] Like, how is AI being used to make music more broadly?

[152] There's a few different categories of what we're calling generative AI music.

[153] What we don't have just yet is an AI bot that can create a song, from a simple text prompt where you would say, make me a song by Drake in The Weekend, and it would spit the whole thing back out at you wholesale.

[154] Hmm, okay.

[155] What we do have are tools that professional musicians and even amateur musicians in their bedrooms can use in bits and pieces.

[156] Songwriters could use a chat bot like ChatGBTGPT to come up with lyrics in the style of another artist.

[157] Or a producer could use AI technology, to create little loops of a guitar riff, for instance, or a drumbeat, or a horn sound.

[158] One example that gained some popularity even before the Drake track, This is the future rave sound.

[159] I'm getting lost in an underground.

[160] This is the future rave sound.

[161] I'm getting lost in an underground.

[162] Was one by the French DJ David Getta, who had used AI to make an M &M song.

[163] There's something that I made as a joke, and it was.

[164] He worked so good, I could not believe it.

[165] I discovered those websites that are about AI.

[166] He used a chapbot to come up with lyrics in the style of the rapper Eminem.

[167] And I went to another AI website that can recreate the voice.

[168] And then he has a David Getta track that sounds as if it features Eminem.

[169] And I played the record and people went nuts.

[170] This is the future rave sound.

[171] I'm getting lost in an underground.

[172] So David Gedda, this French DJ, wants Eminem to wrap on one of his songs, but he doesn't need him because he can just replace him with AI, which kind of sounds like exactly what Ghost Writer did with Drake, right?

[173] That's right, but the key distinction here is that David Gedda wasn't trying to sell his song.

[174] He sees this as an artistic experiment, but he's not putting it on Spotify where it could make actual money.

[175] And what do artists think about this, like Eminem and others who were being used in this way?

[176] or Drake, for that matter.

[177] Drake hasn't commented specifically on ghostwriters, hard on my sleeve.

[178] The Drake's and the weekends and the M &Ms of the world know how big their fan base is and how intense that connection can be between listeners and their favorite artists who have invested in them as people over the years and really want to hear what they have to say, what they're feeling.

[179] So I think they're not sweating just yet.

[180] At the same time, there are other artists.

[181] who see this as a creative possibility.

[182] They think it's cool for people to try their voice on as another instrument in their toolkit.

[183] So you have independent singers like Grimes or this artist I spoke to for my piece, Holly Herndon, who are saying, go ahead, use this tool, use my voice, however you want.

[184] You have my permission.

[185] But they were okay with that?

[186] Just like giving away their voice like that?

[187] I think they realize that this is happening whether they like it or not, and they'd want to see what people can pull off.

[188] It reminds me of the beginnings of sampling in hip -hop, where disco records all of a sudden become the basis for a completely new genre.

[189] Right, and at the time, that felt pretty radical.

[190] Yeah, and innovation often comes out of people breaking the rules or doing things that make people or corporations feel uncomfortable.

[191] But Grimes is saying that if people do make money off of her voice, she should be paid a cut.

[192] So she's trying to get ahead of the business problems.

[193] I think the risks come when you realize that with Grimes' voice as a tool, you could make it sound like she's saying or feeling anything you want.

[194] This raises all sorts of questions about identity and creativity and cultural appropriation.

[195] A white artist could record a song and then make it sound as if Kanye West is rapping their words.

[196] Right.

[197] In its most extreme form, there are critics who see these vocal fakes as yet another form of digital blackface.

[198] You know, you can put the words of a white creator into the mouth of a black artist.

[199] Say more about that.

[200] I mean, the history of American popular music is full of appropriations.

[201] specifically white artists who take, borrow, and steal from black artists.

[202] I mean, rock and roll, Elvis, the Rolling Stones, success of certain white rappers.

[203] That's always been there.

[204] So the uncomfortable thing about this technology is that it could allow for a much easier way of impersonating people.

[205] One thing an artist brings to the table is their identity, their culture, their biography, and to just be able to try on that voice like a costume, you know, raises a lot of complicated.

[206] questions.

[207] Right.

[208] Life experience, identity, culture, like, these are some of the things that make music human, right?

[209] And if you take those things out of the music, then what is it?

[210] I think that's ultimately what worries and interests me the most about these AI technologies in general.

[211] Right now, maybe it's just a vocal impersonation, but what about when AI is writing the lyrics and the melody and impersonating a human wholesale.

[212] For me, I'm a music reporter and a fan.

[213] I'm obsessed with the people who are creating this stuff.

[214] What I'm listening for is the human being behind the music if they sound angry or sad or confident or in love.

[215] As a listener, you assume that that's real on some level.

[216] There's this connection between the listener and the human being performing and writing these songs.

[217] But the question AI raises is, do we even need that connection?

[218] Do we just want something that sounds pleasant enough in the background?

[219] Oh, that sort of sounds like Drake or that sort of sounds like Bob Dylan?

[220] Or do we need to know that this is coming from the depths of their soul and from their lungs and their heart?

[221] Maybe we do or maybe we don't, but as AI keeps coming in music and in other art forms, we're all going to have to answer that question for ourselves.

[222] Joe, thank you.

[223] Thank you.

[224] We'll be right back.

[225] Here's what else you should know today.

[226] On Thursday, numbers released by the Commerce Department showed that the American economy was growing at a slower pace.

[227] Fresh evidence that the Federal Reserve's efforts to cool it were having an effect.

[228] Gross domestic product rose by just over 1 % last quarter.

[229] That was down from 2 .6 % in the previous quarter.

[230] The factor's slowing growth included a slowdown in the housing market and companies pulling back on hiring.

[231] But consumer spending was still rising at a rapid rate, keeping the U .S. for now, out of a recession.

[232] And talk show host Jerry Springer died Thursday.

[233] He was 79.

[234] Welcome to the show.

[235] My guests today say they've been keeping a secret from a loved one far too long.

[236] Springer's wildly popular program, the Jerry's.

[237] Springer show set a new standard for tawdryness on American television.

[238] He encouraged everyday people to make shocking confessions, engage in high -drama screaming matches, and get in regular fistfights on stage.

[239] The show, which ran for three decades, ushered in a new era of daytime television and made Springer a cultural phenomenon.

[240] Today's episode was produced by Will Reed and Michael Simon Johnson with help from Nina Feldman.

[241] It was edited by Liz O 'Balen with help from Michael Benoit.

[242] Contains original music by Dan Powell and Marion Lazzano.

[243] And was engineered by Chris Wood.

[244] The Daily is made by Rachel Quester, Lindsay Garrison, Claire Tennisketter, Paige Cowett, Michael Simon Johnson, Brad Fisher, Chris Wood, Jessica Chung, Stella Tan, Alexandra Lee Young, Lisa Chow, Eric Kruppke, Mark George, Luke Vanderplug, M. J. Davis Lynn, Dan Powell, Sydney Harper, Michael Benoit, Liz O 'Balen, Asta Chetreveidi, Rochelle Bonja, Diana Wyn, Marianne Lazzano, Corey Shrepple, Anita Badajo, Rob Zipko, Alishabey, Mouj, Patricia Willins, Rowan Nemisto, Jody Becker, Ricky Nevetsky, John Kacham, Nina Feldman, Will Reed, Carlos Prieto, Ben Calhoun, Susan Lee, Lexi Diao, Mary Wilson, Alex Stern, Dan Farrell, Sophia Landman, Shannon Lynn, Diane Wong, and Devon Taylor.

[245] Special thanks to Sam Dolnick, Paula Schumann, Lisa Tobin, Larissa Anderson, Julia Simon, Sophia Milan, Mahima Chablani, Mahima Chablani, Des Ibequa, Elizabeth Davis Moore, Jeffrey Miranda, Renan Borelli, Maddie Messiello, and Isabella Anderson.

[246] Our theme music is by Jim Brunberg and Ben Landsberg of Wonderly.

[247] That's it for the daily.

[248] I'm Sabrina Tavernisi.

[249] See you on Monday.