Untitled episode

Music and technology have been codependent partners for decades. And excitement about creative forms of artificial intelligence raises questions about how that relationship will evolve. We've already heard songs performed by deceased singers after fans trained systems on the voices of their idols, and then set software to work on synthesizing fake performances. But living musicians are getting involved too. Grimes said she'd split royalties with people who make

music with her AI powered voice, and closer to my heart, the heavy metal band, TesseracT, used generative AI to inspire concepts and themes for their most recent album and music videos. So on this episode of "AI IRL," we're going to speak to experts in the music industry to find out whether the working relationship between music and technology will be as sublime as that of me and

Nate's. Yeah, or less friendly like Liam and Noel Gallagher. Ryan Groves, great to have you here. So you're the co-founder and CTO of a startup called Infinite Album. Tell us a little bit about what you do there. Yeah, absolutely. Thanks for having me. So what we do is we use AI to make infinite soundtracks for video games, essentially. And we do this for gamers and game

devs, game studios as well. And what this allows is for the soundtrack to be relevant to the game scenario, in whatever scenario it is. How do you use AI to make some of those tracks? Yeah, it's quite a process. We sort of break it down. So we sort of break down the music composition process, music composition, production performance process, and kind of create different AI models

that can do very specific things. 'Cause in general, AI, we're not really at artificial general intelligence yet. And AI is really good at doing very specific things. So we sort of hot swap these AI modules based on whether you need like a sad ambient music, or you need like a upbeat rock soundtrack, or something like that. So let's give a hypothetical example. What game do you

do you... Do you game? I can't remember. I don't really game, but let's say I'm doing, I don't know, Angry Birds. All right let's say Angry Birds. So you're lining up your angry bird, and you've got some optimistic, hopeful music, and then you miss, or it hits a building and dies. And then you are generating presumably sad music. Exactly. It's along those lines, okay. Because I

just, I sort of feel like you can do that without AI. So what is it that AI is bringing to that? Why is that different to just having a roster of tracks and just picking- Yeah. The appropriate one for a fight scene or something? Yeah, it's a very good point. The challenge with doing it without AI, is that you have to prepare all of these essentially

loops that can fluidly transition from one to another. So when you're a game composer, you have to create, you know, all of those sad tracks for that particular level, if it has a unique song per level, say. And so what we've done is sort of abstracted that, so that it's just every song that you play in our system can move from happy to sad, to angry,

to tense, excited, all these different emotions. And it can do it in real time. So aside from Infinite Album, AI is catching wind from Hollywood, to every corner of the music industry. What are some of the applications in AI and music that you're most excited about? Yeah, well, I'm a little biased personally, that I feel the gaming opportunity is incredible. 'Cause music has been sort of

part of that. Partially because of rights, but also partially 'cause of this format that's adaptive and hard to build if you're just a regular, you know, music producer. So we see that as a huge opportunity. It sounds like you're on the more optimistic end of the spectrum, but curious what your take is on what AI could mean for some specific jobs in the industry. Yeah, absolutely.

I think like any revolutionary technology, there's gonna be jobs that are lost to AI, for example, I think probably production music libraries are gonna be pretty hard... It's gonna be pretty hard to compete there. So I'm gonna speak as a musician very briefly. He's being very modest. Well, I've played the drums for 30-year-old years. Yeah. But, and my point is that a computer can take my

drum kit over my dead body. And there are people like Sting and Nick Cave, who have spoken out against AI, and I'm all for AI in general. Mm hm. But I always find it very annoying when I hear people saying, oh, isn't it great how you can just sequence this drum track? You don't need a drummer anymore. Like in the sea, gone. Yeah. So I'm just

curious from that perspective, like, what are those people slash me getting wrong about the potential for AI in music creation? I think that it boils down to more about musical taste of the person using those AI tools, rather than just being able to generate it. I don't think it's gonna be very special if everybody can just generate a drum track, it's not gonna be special and

it's not gonna have resonance with people, unless you put your own unique spin on that drum track. Because AI also does sort of generalize things, so it sort of makes it a little bit more bland. So, I don't worry too much, I still think there's... My hope is that it brings people to be inspired to learn how to play the drums. I think, I mean, to

a certain extent for the last few years, technology has massively formed the bedrock of most recording studios, right? Yeah. I mean, if you look at something like Pro Tools and plug-ins, and triggers on drum kits, personal bugbear of mine for some reason, even pitch correction and auto-tune- And all that kind of stuff- Yeah. Like manipulating sound has in some way formed a core part of everything

that we appreciate about modern music. Even if singers can sing, for instance. I don't get why AI is so controversially different to some of those tools, depending on how it's used. Yeah, I think the big difference is that a lot of AI models were trained on thousands to tens and thousands of artist content. So there's- So it's the rights. It's the rights I think, is what

people are really concerned about. People right now especially assume that any big AI model is essentially infringing on copyright. And there's probably good arguments to say that some of 'em are. So- Do you think that? Yes, I think a good example of this recently was Stability, who worked with a production library specifically for their Stable Audio text to music solution. So they actually worked with a

production library, they got 800,000 tracks, they licensed it from them, so that at least there was a clear line of a contract between the people who their music was trained on, and the output, the company. So other... I think that's a recent thing. And I think, you know, people have even been talking about having like a universal license for training data, the same way that we

have a license for playing music on the radio. So I think that's sort of the emerging ideas that are happening, but I think everybody agrees that, you know, you shouldn't have... You should have control over the music that you create. Are you finding that music executives from the industry, like Sony for example, are receptive to the idea of AI generated music? Or have they not really

taken it seriously just yet? It has changed quite a lot this year actually. So this TikTok, or Ghostwriter, came out with this fake Drake song, where he used voice transformation to imitate Drake and the Weeknd. And he published it, he even submitted to the Grammys, I don't know if you saw that. Yeah. He went viral. He did. Yeah, exactly. So that actually has changed the perception

of AI quite a lot, negatively, I would say. Because people thought first of all that it was a fully generated track, and it wasn't fully generated at all. He composed the whole thing and then just for the voice, as far as I know, that was the whole thing. For the voice, he actually has to sing that first. So he probably did like an imitation of Drake

and an Imitation of the Weeknd, and then he used an AI model to transform the sound of the voice. So really it was just sort of that last step, but obviously there's a lot of issue. That's a very obvious issue with consent- Consenting. Yeah, exactly. And we've seen things since then of performances or cover versions performed by dead singers. Yep. Over, you know, contemporary or even

older songs. Whoa, oh, sweet child of mine Very obviously without their permission. And a lot of the time it's just done for fun, or it's a proof of concept. And a lot of the stuff that we've talked about, it does always come across as it's kind of a, "Hey look what we can do." Yeah. Not necessarily does anyone want us to, or, should we? Yeah. It's

more like, this is possible, shall we explore it? And it seems that right now, that's kind of where music creation broadly is, except for in the world of maybe things like video games, or little bits and pieces on the edge, it's, I dunno- Yeah. It's a roundabout way of saying, I think drummers are safe, right? I think you still are gonna have to develop a really

strong sense of musical tastes, to be able to really leverage these AI models. And the other thing is that, you know, when people follow a particular artist, they don't only like just the sounds that they produce, right. It's a lot more than that. It's the personality of the artist. Yeah. It's, you know, their values, it's their aesthetic. So there's quite a lot more than that. And

that's not something that AI is going to easily reproduce. So there is a sort of humanistic connection. Some critics would say that AI generated music will lead to a world in which music is standardized and formulaic and kind of divorced from the emotion that you just mentioned. Mm hm. And some of the best songs in history were written and inspired by feelings of love, or loss,

grief and heartbreak. Is it possible to have AI generated music that evokes these feelings, even if a machine didn't feel them? I think that it's plausible. I think that an AI could do it, but it might be something that it's sort of stumbled into. Because a lot of the times with AI you're generating like a lot of content. And then the typical process is that humans

go and curate that content. So you can generate a thousand songs, nobody cares about a thousand songs, and nobody's gonna listen to a thousand songs. So there's quite a lot of effort in sort of the typical way that people interact with AI, where they're essentially digging through, you know, a pile of generated material. Lemme ask you though, Ryan, I mean, do you think the hype at

the moment is kind of ahead of the reality of what is possible, or desirable musically right now? Yes, absolutely. I think it was so easy to make a headline, for the last three, four years, I would say. Where someone would come out with a song, and say, this was generated with AI. And everybody just assumes that it was completely generated with AI. And there's a huge

range between what the human contributed, and between what the AI contributed. And that's really sort of what we're trying to be transparent about. You know, there's this huge range people can... There is creativity on what AI models you choose. There's like artistic vision with what ai AI models you choose. There's like hundreds of different ways you can approach it. You know, people even sort of subvert

AI models, where they like use a text model for generating notes, or, you know, vice versa. So yeah, I think the hype, there is a bit more hype around AI, and a little bit more fear, because of those sort of attention grabbing headlines, and the ability to do that. So- Like you and your drums. Again, over my dead body, they're staying with me. Yeah. Brian Groves,

thank you so much for joining us. Yeah, thank you. Tom Kiehl, great to have you here. Now you are the interim CEO of UK Music, which is a British music representative body. Tell us how you work with artists and labels and the wider industry at the moment, and how that's changing with AI. Sure. Well, we're a representative body for the music industry. So we bring together

all the various component parts, that might be the record companies, the publishing organizations, the creative organizations, all those kind of trade association and campaigning organizations that have an interest within the music industry. We bring them together under UK Music, and we talk to parliament, and government in the UK about shared priorities and an issue like AI has been something which has really brought the sector very

much together in terms of our shared response. Yeah, I bet. So what have the nature of the conversations been this year on the topic of AI specifically? Well, sure. Well the UK government last year proposed like changing copyright law, to basically make it easier to use AI technology without permission and licenses. And actually that would create a lot of concern amongst the music industry. And so

we campaigned against that proposal and the UK government ended up sort of essentially reversing that and decided not to proceed with that exemption. But it's raised a lot of issues about the impact of AI on our sector, and I think we are really trying to get to grips with that. We've produced a position paper for the sector, which we've worked towards, and that sets out a

number of areas where we think that the UK government actually, globally, we can maybe take a leadership role, in actually making sure that industries like music are protected whilst also working together with AI firms to make sure we grow in tandem. And you also very recently met with Prime Minister Rishi Sunak, to discuss the impact AI is having on the industry. What specifically came out of

those conversations? Well, we basically met with some of his advisors and his team about AI and technology, within our position paper we made a number of recommendations, particularly around maybe like how the training process with AI technology goes, how that works in practice to make sure that consent and creators consent is part of that. So we talk very much deeply involved in that. Are there any

points of disagreement between those two parties? In which two parties? Between you and the government? I think we want them to grow in tandem. I think we want them to basically say that the law gives protection to creators and to kind of put that firmly in place. So, I mean there is an idea of there being some kind of code of conduct, which the intellectual property

office in the UK is developing. That's around trying to bring the AI companies and some of the rights holders and creators organizations that we represent, trying to get them to agree something. It's probably worth us taking a nice comfortable step back and saying what was being proposed as an exemption to copyright as a result of AI. Like, what was being proposed as a good idea? Sure.

Well, the government was proposing last year to basically enable people to use copyright protected content for AI purposes, without the permission for licenses. So, it intrigues me, because if you listen to enough music, you are being, let's say you, a person, Jackie, is being inspired by like that artist's work. And so when Jackie picks up a guitar, which I've seen her do, sort of, and starts

writing songs in the style of whoever. Taylor Swift. Taylor Swift. You know, she's inspired by her, and like how is that different from just feeding that music into a system? That's a very good point. I mean, we would say that that inspiration, there's always something behind that. Whether you've heard a song on a radio, or you've heard it on a streaming service, or you've bought the

kind of sheet music to learn that song. There's always some kind of like backend transaction between whoever created the content and the rights holders, and then who's then putting it out to the public. And so people have bought that, essentially bought a license to use it in some way, can be inspired in that way. What's not happening, or potentially isn't happening within the AI sphere, and

what the proposal would've done, was to actually kind of cut that process out, and actually say, well, you can train AI systems, these large language models they can be trained on music and copyrighted protected materials, but without that permission being needed at all. And so that compensation. So it's another form of that debate essentially. But it's financial. I mean, what if a theoretical company, that wanted

to train an AI model on all of, insert major record labels name here, on all of their copyright works, could they not just pay a very fancy subscription type fee and have it fed in? I mean, would that be a solution? Well, I mean the music industry has a number of licensing solutions in place to achieve that. That's all based on negotiation. What the proposal would've

done, would've cut out any kind of kind of negotiation on that. So there isn't necessarily a licensing solution base at the moment. As I said, we're learning all the time about how AI actually works. And one of our key recommendations in our paper was around transparency and actually making sure that the AI firms have audited, or kept records. So we actually get that data, to know

what's actually being fed in. So that the right peoples works can be compensated as well. So I think there's a complicated system, but because things are growing an exponential level, we're all trying to keep pace with it. This isn't a time to really undo that, that legislative framework. On that note, earlier this year Spotify pulled a track that had AI cloned voices of the Weeknd and

Drake. And just very recently the CEO came out, and said that it wouldn't consider a complete ban of AI generated music, which to some might seem controversial given their previous move with the Weeknd and Drake track. What's your stance on what should and shouldn't be banned? Yeah. The Drake/Weeknd example is a very interesting one, because from what I understand, we've heard it was pulled because of

an uncleared sample, rather than kind of like the fact that it was AI generated. One of the things we've been arguing for is about labeling and making sure that there's, from a sectoral point of view, that it is quite clear to consumers whether something's being created with AI purposes. Yeah, I mean that's something that I've looked for in the liner notes of albums. And it doesn't

say, you know, oh, well this was, this song was engineered using these pro tools plugins, and you know, or on a lot of pop songs. And not even just pop songs, but most modern music has some degree of manipulation involved in it. It doesn't like, we're not expected to see how that sausage was made. We just sort of generally accept that some degree of electronic manipulation

was done. So like why should AI be any different? Well I think people, the industry, has been using AI for a number of years, whether it's at an artistic level, and as you say, that innovation has been been happening within the sector. People want to use it, certainly. We think that our issue is more about the generative point. It's about how you generate a new piece

of music from existing other pieces of work, and then the output. 'cause that could have potentially quite big commercial success and then it would transfer value away from the creators in the sector that created it. To the point that you were making earlier, about looking through the notes of how a song is composed, or what tools are used in a particular song. I think that's a

really unique example, because you play the drums, you are a pretty sophisticated listener. Most of us aren't. If I hear a song that I sounds cool, it's great, but I'm not looking up exactly what instruments were used. To your point about music education, and kind of bolstering that framework for future generations of artists. What do you think needs to be done now in an AI context?

I mean, I think you're raising a very good point in particular, if you think about things like film soundtracks and video game soundtracks, these are actually very big parts of the music industry, in terms of like the growth potential. Lots of the people we're concerned about within AI, that is their living, but they may not have the kind of profile of an artist. But how do

you kind of like protect that situation? I think that's a lot of where our concerns come from. I'd also say, another area we're kind of exploring, which kind of moving away from the kind of issue around copyright law, but around more like personality rights as well, for the artists who do have actually have quite significant profile. But more generally it kind of goes into wider areas

of civic society about the idea of deep fakes and political campaigning, and ensuring that... Personality rights are not as strong in the UK as they are in some other parts of the world. So one of the arguments we've been talking about with UK government to see if we can kind of strengthen that particular regime too, to make sure that that is secure. Have you seen any

good examples elsewhere in the world of like unique approaches to, not necessarily regulating AI music, but protecting artists that you think should really be applied worldwide? Sure. Well that's an interesting question. 'Cause in the EU at the moment, there's a proposal regarding an AI act. So it's not law yet, but they're obviously taking steps there. Also, I mean even countries such as China, we actually are

seeing that they are putting some quite robust mechanisms. So from a a UK government point of view, for us to kind of be seen behind China in that way is actually quite problematic really. Talk to us about how artificial intelligence is impacting the economics of the music industry and what that ultimately could mean for consumers. I think it's probably too early to fully say. I mean,

we do annual reports on the music industry. We've got one coming out very soon. In terms of like capturing the sort of economic value, I think it's probably a little bit too early yet. There's potentially huge growth potential in AI for society. And so as a commercially minded music industry, we'd want to be part of that success. But we kind of need to have our underlying

business model protected in that process too. Thanks so much for joining us, Tom. Thank you.