Front Burner - The Beatles and the future of AI music

Episode Date: June 16, 2023

Before his death, John Lennon recorded a demo of a new song, "Now and Then" on a cassette. His Beatles bandmates later tried to repurpose it for release, but abandoned the project in part because of the poor voice quality. This week, Paul McCartney revealed that, 43 years after Lennon's death, the song will drop – thanks to AI technology. It's just the latest example of artificial intelligence's increasing presence in the music industry. Fake Drake songs, AI-generated Kanye covers and posthumous Biggie collabs have raised alarm about copyright, and existential questions about songwriting and creativity. Today, Saroja Coelho speaks with the host of Vulture's Switched on Pop podcast, Charlie Harding, about what the technology means for the music industry and art itself. For transcripts of this series, please visit: https://www.cbc.ca/radio/frontburner/transcripts

In the Dragon's Den, a simple pitch can lead to a life-changing connection. Watch new episodes of Dragon's Den free on CBC Gem. Brought to you in part by National Angel Capital Organization, empowering Canada's entrepreneurs through angel investment and industry connections. This is a CBC Podcast. Hi, I'm Saroja Coelho. Behind all the hiss and hum and crackling, what you're listening to is a demo John Lennon recorded before his death. It's from a cassette he left behind that was labeled For Paul.

His bandmates from the Beatles tried to make it into a song in the 90s, but had to abandon it in part because of the poor voice quality. And I don't want to lose you But this week, Paul McCartney revealed that they've finished a, quote, final Beatles song, with Lennon's voice extracted from that cassette. And they've been able to do it thanks to AI technology. We were able to take John's voice and get it pure through this AI, so that then we could mix the record as you would normally do. So there's a good side to it and then a scary side. Artificial intelligence has caused something of a panic in the music industry this year. There was the fake Drake and the Weeknd

song that became a hit all on its own. Studios have grappled with what that means for copyright. Spotify has been flooded with AI-generated mood music, and some artists have disavowed what AI creates as inhuman. But the final Beatles song, that's an example of how AI in music doesn't always mean robots are playing songwriter and how it can actually be used to enable creativity. So today on FrontBurner, I'm talking to someone who really knows the art of pop. Charlie Harding is co-host of the Vulture podcast Switch on Pop, which is all about the making and meaning of popular music. Hey, Charlie.

Hi, thank you for having me. Okay, before we get into all of this, we got to pull some of this apart. I think there's a little confusion around AI being used in music because not all of it is that chat GPT where you type in a few words and it spits out a song. So I want to break down some of the different ways that it's used, starting with that final Beatles tune. How does it use AI in a way that isn't actually all that new? You're right. The Beatles have previously recorded music with John Lennon in 1995. They had released Real Love. Free as a Bird.

From old tapes of Lennon, and they all got together and made a song. And this is kind of in the same vein. In this case, there is a recording that isn't very good quality, and AI is being used to separate all of the bad artifacts and background instruments to just highlight John Lennon's voice so that they can mix it up and sound more professional. This is technology which has existed for a while, and that the Beatles previously used on their film Get Back. Something in the way, sir. What attracted me at all? Just say whatever comes into your head each time.

Attracts me like a cauliflower until you get the word. The documentary highlighting the making of their final recordings. And they used AI to improve the really poor audio and video quality that had been taken to make it HD for today's audiences. What is it exactly that's happening there? Why weren't they able to do that with all the musical technology and editing suites available to them? Why was it AI that got them to that nice, clean, isolated voice? Well, with all the great trickery that post-production folks have had forever, there is always the challenge that it's hard to fix things in post. If there's too much crackling or wind noise or

someone's playing bass in the background, oftentimes those frequencies overlap and you just can't remove them with existing technology. AI trained through a neural net can say, I know what part of the frequency spectrum is just the voice and go in and peel it apart from the recording and separate all of the various elements from each other. You can get the voice, the bass, the piano, all on separate tracks again. It's not perfect, but it's pretty good. You know, the moment for me that kind of made my brain do a small explosion was when I heard the fake Drake and the Weeknd song. That was just in April. Wow. I mean, it's almost difficult to believe it's not them. How do you think they went around making it sound like Drake? You know, I think there has been some hyperbole in some of the

headlines around AI makes music because we don't know exactly how this song was made. It was written by someone who is going as a ghostwriter. But I think it's very fair to speculate that the music was probably made by a person. And when I say music, I mean the underlying beat. The lyrics probably were written by a person. They sort of mimic how Drake might write some lyrics. And in fact, someone probably rapped those lyrics as well. The part where AI comes in is in vocal resynthesis. Basically, taking your voice and superimposing the qualities of somebody else's voice onto your voice. I've done this myself. There's open and available software on the internet that will allow you to re-synthesize your voice in the likeness of

other people. So I made my own Drake soundalike track. I'm not a good rapper, but when you superimpose his voice onto mine, it doesn't sound so bad. So I think the Drake song is probably made mostly by humans, but uses AI in one portion of the creation of the song, not unlike Paul McCartney is going to be doing with John Lennon's voice. The song that you're talking about there is the Laserbong song. Is that right? It's a laserbong. It's a bong. Oh, yeah, I can't believe you found that. Oh, my God. Yeah, I can't believe you found that. So this is so much fun because here's what's kind of interesting for me.

You go into a studio. You had a really short period of time to play around with voices. You capture Drake, and it sounds just, you know, just the right kind of energy, the right amount of sad boy, the right amount of emotion. I mean, it sounds like him. It really sounds like him. But I wonder how important the human story is for it. You know, it's not, does the song, do you think these songs stand on their own? Or do we need to know this is connected to Drake in some way? Or this is connected to an artist or a human being that has a story? I think that's a really great point because all of popular music, I think, is predicated on this veil of authenticity, that it's coming from a person, that it's speaking from their real life,

or using metaphor to communicate some essential human message. And so much of popular music is based on the relationship between the artist and the listener. And so while a handful of these AI tracks, I think, are grabbing headlines at the moment because it's novel, I think in the long run, we're still going to want to listen to music by people, whether or not they use AI in the creation of that music. We are social beings, and we want to connect with other people. And I think that music is going to remain a human art form, even when AI is involved in the process. You know, the one that really made me think, okay, the human story is still important here was listening to Kanye West, an AI version of Hey There, Delilah. Have you heard that?

I have heard this recording. It's quite fun. Hey there, Delilah, what's it like in New York City? I'm a thousand miles away. I mean, that is a song that doesn't act. The original didn't speak to me as much as listening to Kanye do it, in part because it feels, after all the turbulence and the hideous headlines, it felt like a musical mea culpa. And it did.

And it feels, I mean, the AI has recreated him so well. You can feel the kind of leaning into the mic for meaning and the breaths that come in between the words. It isn't just about resonance or tone. There's something emotional there that is recreated surprisingly well. from the original performance that when I performed as Drake, it was my voice and my performance underneath. It was all of the inflections. It was the breath. It was the energy. But then it melds that with the timbres and the styling of the re-synthesized rapper on the other side. So it is a blend of a real human performance with an artificial intelligence. There's one that was dropped by producer Timbaland. That one was also,

he was very excited about it. I got to share something I've been working on because I always wanted to do this and I never got a chance to. I always wanted to work with Big and I never got a chance to. Until today,

it came out right. Play. With this one, he is bringing Biggie back from the dead. Can you tell us about that one? Oh, this is an interesting experience where we have artists sort of ventriloquizing the dead. And Timbaland claims that he always wanted to produce and work with Biggie. And of course, you know, Biggie's not been on this earth for a long time now. And so he's made a track that features, air quotes, Biggie's voice. This is a very different thing than what the Beatles are doing, which is recovering a recording that had been previously unrecoverable from existing technology. Now we have voices generating new musical material using AI. And, you know, again, I think the first couple of these is going to be

interesting. They're novel. But fundamentally, there is something creepy about it that we are bringing folks back from the dead. And I think in some cases, it may work and fans may be excited about it. But again, we're all about the human connection with artists. And we grow up with them. We co-live our experiences with them. And I think that I can't speculate too far into the future, but it seems like a novel way of capturing some headlines and a sort of strange way of connecting with culture. But it is still kind of this amazing communing with the dead.

You know, usually when an artist is gone, they are gone. But last night I discovered a Nirvana cover. Well, it was never actually made, but an AI-generated Nirvana cover of Creep. An album of creep. I mean, you can just imagine all of the possibilities here. An album from Bowie, an album from Amy Winehouse, Tupac Shakur. I mean, the list of artists who we felt we weren't finished with yet being suddenly available to us. We may want it, and I'm sure that there are states that will manage this well.

The challenge is still how do you make great work with it? How do you curate things that feel like they would be in the voice of that artist? At the moment, AI cannot generate whole songs that sound mix mastered and incredibly well recorded with arrangements and lyrics that fit the styling of an artist. But AI can be used in many of those steps to help aid in the creative process at this moment. And so estates that are really good at managing the afterlife of an artist

may be able to create some interesting material. But we've also seen estates that have gone against the wishes of the artists. We've seen estates that have published music on Spotify that had been kept off of Spotify. I'm thinking of the Prince Estate, going against the wishes of the former creator. So I think it's going to be on a very case-by-case basis. There are really wide reactions. I mean, Drake, well, Drake was not happy about the fake version of him doing a collaboration with, I mean, daddy was mad. But there are artists who have fully embraced this. Grimes comes to mind, Canadian artist. herself open source and launched a service to replicate her own vocals, even offering a 50%

share with anyone who produces a song that she thinks has really good value. Why would she go in this completely other direction? Well, Grimes is somebody who has openly embraced cutting edge technology, whether that has been in the crypto space or elsewhere. And so I think this is consistent with her identity as an artist. I feel like the artist and the engineer should like remain in a dance. And like, the engineers are doing something crazy right now. And like, I just like want to be part of that dance.

The idea that she's open source, I think we should probably not give full credit to the open sourceness of her voice. I mean, she's creating this elf tech technology where you can use your voice, do the resynthesis of Grimes' voices onto your voice. It's technology created by another company and licensed. And then if she approves your song, then there's a 50-50 royalty. It kind of seems more akin to remixing in the past, where maybe you got an acapella and you remix the song, and maybe there would have been a royalty split. Oftentimes, remixers don't get great royalty splits. But it feels like an

update to a pre-existing way of making music. And again, I think she's doing so in a moment where AI is very much in the news. It's a great way of capturing some attention. And on the artistic side, she has said that she's been able to hear herself in productions that would never have come from her own creation, and that there's something artistically validating about hearing her voice in new contexts. I'm going to go. through Angel Investment and Industry Connections. income. That's not a typo. 50%. That's because money is confusing. In my new book and podcast, Money for Couples, I help you and your partner create a financial vision together. To listen to this podcast, just search for Money for Couples. The ownership piece becomes really interesting here. I mean, copyright conversations in music are so tangled and weedy and they just go on forever. It's like a spiral that never ends. But where you are in the United States, do these pieces of music go against copyright when you're training the AI to use the voice of someone? But here we're creating music without their input.

Personality rights are not covered by copyright. So your likeness, the sound of your voice, anything that could identify you and be used for a commercial product, that's not part of the U.S. copyright code. It's an area of law that's usually managed at the state level. So if you have a complaint about a song that is using your likeness, it's much more complicated to litigate. You're going to have to go to every state which has different statutes, and you're going to have to figure out how to claim rights to your music. So there are a lot of cases ongoing at the moment that are going to determine the future of how this sort of likeness is going to be granted to folks,

but it is not a part of U.S. copyright law. And it's not as simple as, hey, you have stolen a sample from their recording and used it without permission, which would fall within copyright. Lawyers have said that things are pretty murky here in Canada, too, and that it's going to take some work on our Copyright Act to figure out who owns AI creations and what infringement looks like. So we step away from that stuff, the things that are trained on someone else's voice and what a copy is, and just look at what the uses might be in pop production. In what way could you see voice tech being practical or useful for the way that artists put songs together? There are so many ways that we can use AI creatively and practically. One that comes to mind is that oftentimes when you have a recording session, you might mix the song and think,

oh, we're missing this essential little harmony note. We didn't capture it. Oh, no, the artist is on a world tour. What are we going to do? We can't get them back in the studio. It'd be very convenient to have someone else come and sing that line and then use vocal resynthesis to make it sound like the artist, if that's desired. I do fear the moment when a particularly lazy artist just says, hey, ghostwriter, go make me this song. Let's

resynthesize it with my voice. I don't even have time to get in the studio. The technology isn't, from my ears, not quite good enough to do that. I mean, it's getting very, very close, but I can still hear some little digital artifacts and sort of strangeness when these technologies are applied, but I'm assuming those will get much, much better. But there are countless ways in which AI is being used in the studio, from helping in the mixing and mastering stages to using unique AI-based instruments to help create your arrangement. So I think AI is not only at the moment a part of the creative process, but it will be increasingly so. I imagine that large language models like ChatGPT can be used for some parts of the songwriting process, like finding better rhymes, for example.

And when it comes to composing music itself, there is some really amazing technology already out there. Google released a tool last month, Meta put one out. So I'm wondering, as you put your ears on that, in general, when you listen to that AI-generated music, what's your feeling? Well, AI can generate music in many different ways, but two primarily come to mind. One is when you actually get like a finished whole recording, a waveform that you can hit play on your computer from maybe a text prompt. And you might say, yeah, write me a Kurt Cobain song, but make it country. And it spits out the whole final recording. Those recordings sound hallucinatory. They kind of sound like if you're driving between cities and you're listening to AM radio and you're

in between two stations. But it's getting there sound like, yeah, that kind of sounds like a Kurt Cobain country thing in a hallucinatory AM radio kind of way. I actually think that that music is interesting in itself because it's so strange and I would love to sample it and use it as part of the creative process. Now, the other way that we can use AI to make compositions is to say, make the notes and the rhythms for me. Spit out MIDI data, data that I can use to import into my audio software and assign more beautiful sounding instruments where I can then go in and properly mix and master the song. this case, when you ask a lot of these, especially the large language models to write a melody, it's not making hooks yet. I'm sure people will make much better tools, but right now, again, they just sort of sound amateurish or strange. So these are the two ways that I think about AI writing whole songs, giving you a finished waveform or writing the notes for you to produce out yourself. And both of them are, at this point, fun, creative tools to help you

hear things in new ways, break through creative blocks. But the finished music is kind of meh. Well, one place that we've been seeing just a ton of AI-generated music is on streaming services. I mean, streaming services are almost tearing out their hair over this one. And you've got startups like Boomi. They are cracking out, I don't know, millions of tracks at this point, I think. And they've been accused of using bots to fake the amount of listeners. But still, why would AI tracks find a home on something like Spotify?

That's because a lot of these streaming services use functional-based playlists as a way of engaging listeners. And when I say functional, I'm talking about mood-based playlists. I want a chill playlist. I want the thing that I can study to. I want a cooking playlist. I want a workout playlist. And the job of that music is to absolutely just fill the background to another primary activity. And in certain cases, I think that there is good enough music sounding like AI written material that can be the background thing to help you fall asleep because all it's doing is just like wishy-washy synthesizers that don't have to go in any particular direction. And so listening to functional-based music is a big part of the

streaming business model. I'm not surprised that people are trying to capture some of that with music that they can make on the assembly line floor instantaneously. You know, so much of our conversation has been about how much faith do you have in this? And are you afraid of this? And you've got a, you know, clearly very arms wide open feeling about all of this. But there have been so many moments in history where new technology came along and we were all very frightened that it would kill something in the art, that it would kill music, that something would be changed irreversibly, altered forever. And we don't even have to go that far back in history. If you think about samples in hip hop or even the synthesizer, I mean, people thought that was going to be the music. So if you think about looking back at this moment in time, you know, as we grow with this technology, what do those moments tell us about how we should approach AI? I think I have faith in musicians using technology to find new modes of human expression. There are certainly ways in which this is a different kind of technology than the synthesizer, which,

frankly, is not very good at synthesizing the sound of a violin or a horn. If you want a violin or a horn on your recording, you should go actually record one with a human player. The computer technology so far is not as good as the human performer, but in this case, we do have a technology which seems primed to replace more and more parts of the human process. And I can see why folks are anxious about it. I think that there are jobs that will be lost in music. But I also think that those who find new and creative ways to embrace it will find new kinds of human expression. And I think that's ultimately what music is about. As much as music is a business, it's primarily a mode of communication. And if we can find ways to be

more human and express human creativity, I think that we can do so using these tools. They can expand the way that we see the world, the way that we can express ourselves. What a wonderfully optimistic and open-hearted way to approach something that does feel scary for a lot of people. It's interesting that you bring up that piece about the business and loss of jobs, because that makes me think of the Napster moment, iTunes, streaming. So if we look at industry for a moment, what lessons do you think they can learn about how they could adapt here? Well, I think it's worth pointing out that the music industry is a highly consolidated business like the rest of our economy, unfortunately, and that there's a very few number of people who

rein in a ton of money at the top. And there's a lot of people who simply don't turn it into a meaningful income. It's a very difficult business to work in. I think for those at the very top, those top 100 pop stars, I think in a lot of ways, they're going to be okay. Because, you know, not only do they own all the power in the Capitol at the moment, but they are also the folks that have the strongest human relationship to their fans. And I just don't think that you can easily replace that. If you can, we're talking about a huge paradigm shift in culture where humans stop caring about connecting to other humans and that they're happily willing to just consume computer generated art, which is interesting in and of itself. But I do think that these Hot 100 artists, they're going to be okay.

is interesting in and of itself, but I do think that these Hot 100 artists, they're going to be okay. For those who are trying to make it in the middle class, it's going to be more challenging because it makes me think of, you know, we talked about technological replacement. Certainly, the camera was not great for people that were doing portraiture and oil painting. And there are parts of the music business that sustain a middle class living for some folks, especially film scoring, sync, content music, library music that might be used on a podcast or a radio show like we're talking on. Those kind of composers, it's going to be harder to find those jobs when there is very easy to generate background music. So I don't know if I have a message of optimism for those working in business. I think that for me, I'm mostly excited about these tools as a force of creativity.

And those things are not always going to be, you know, kumbaya. There's a piece here that I want to take where we take a couple of steps forward into the future, because what we've been speaking about mostly is where human beings are a jumping off point or a collaborative partner, somehow still engaged with this in the creative process. But could you imagine a world in which AI takes the initiative itself, creates the music from beginning to end itself, puts it out itself, a music that is entirely created by AI from start to finish. Sure. I think we can speculate on that. I think it's probably going to be possible, at least in terms of the direction of this technology and how fast it's evolving.

I think it will have a place in human listening. And I think humans will still want to connect with other humans. And I think that all of this will be determined by what we choose to value. I also think that when AI is writing music at a really high level, we just have much bigger concerns to worry about what AI is doing in other fields that have more direct material effect on our life in the day to day. more direct material effect on our life in the day-to-day. That's replacing all other kinds of jobs, generating new scientific research that could be both for our benefit, but you replace existing scientists. So there's just so many... Music, so often people can criticize, especially pop music, for all sounding the same. But really great music is often the coming together of tens of thousands of small creative choices to generate something new, especially in the recording, mixing and mastering of a song. There's so many human steps that parts of them will be replicated that when all of those things are replicated, I just think AI is gonna be much more powerful at other things that we

have to be concerned about. It has been such a pleasure to be in this Wild West with you for a few moments. Thank you so much, Charlie. Thank you. It's been a pleasure. That's all for this week. Front Burner was produced by Imogen Burchard, Derek Vanderwyk, Lauren Donnelly, Rafferty Baker, and Jodi Martinson. Our sound design was by Mackenzie Cameron.

Our music is by Joseph Chabison. Our executive producer is Nick McCabe-Locos. I'm Saroja Coelho. Thanks for listening, and we'll see you next week. For more CBC Podcasts, go to cbc.ca slash podcasts.