AI and the Fight for Music Rights

I personally believe that it would be tragedy to think that we can have a sea of unlimited generative music free of copyright that was made on human creation because I think it will savage artistry, human artistry. I really do hope that we end up getting major jurisdictions decide that you do need to license to train. I guess then we get into a world of like, should you license if the outputs are free of copyright and hard to monetize? Licensing for music has been one of the most hot topics for 20 years. But something's happening right now and AI is coming in and you know most of the time it's not licensing. Together with Becky I talk about the whole journey of licensing in the past and what's going to happen in the future. What do we see? Hey guys and welcome back to the Sound Connections podcast. I have a guest in the studio that actually was one of the first people I asked to be on the podcast. I guess Becky Brooke, welcome. Thank you. You and I both live busy lives and I can't remember why but we didn't get about to do it before now. But welcome

Becky. We're going to speak about something really interesting. We're chatting back and forth. We're going to talk about licensing music in the AI space as well and within ethics. Licensing, I'll let you speak about it in two seconds. Licensing is a thing I know a lot about from a theoretical perspective, read articles, but it is a particular thing. But before we get into that right now, Becky, who are you and what do you do? So I'm a music and technology advisor. So I work with predominantly startups who often have tech backgrounds and want to work with music industry, whether that is to license music rights and release a consumer product or potentially like sell services to creators or the music industry more broadly.

Occasionally, I sometimes work with very big companies who have music or tech and want to do something with them in that space. So predominantly startup advisory, but occasionally I have, you know, big corporates as well. Interesting. It's a space that both you and I love and there's a few of us out there in Europe and you guys are very much associated to Music Tech Europe, which I know you're also a mentor in. So and that just launched. So this is not a partnership episode. We'll have more of those. But again, go apply for that. That's going to be amazing. But Becky, just explain to me very basically, what is licensing?

So I've been in this world's best part of 20 years. And when I started, I've always focused on what we would have called digital products. So I've never worked in the physical world, really. Physical product, like CDs and vinyl, I've always focused on digital products, whether streaming services or download services. But obviously, those things have got inherently more complex. And so when we started, what we really cared about were recorded music rights and musical work rights. So that is to say each piece, each song we listen to is two individual pieces of copyright.

And so those are the two things we thought about. We thought about music licensing, musical work licensing, otherwise known as the publishing industry and recorded music rights. We're going to talk about AI later. And that really starts to make it much more complex because digital music licensing, when you start to think about AI, starts to look at a lot more rights types, a lot more types of rights beyond copyright. Okay, this is interesting. But before I go too deep into the topic, I need to understand your journey into music.

Because again, what I do is a very particular thing. You do something that's sort of comparable to some degree. You've come much further in your career than I have. So I'm very lucky to talk to you. But how did you get into this? That's a good question. So I was always a geek and I love this space. You know, as a student, I was a very active music consumer. Actually, so much so that I was subscribing to a thing called OD2 back in, back 21 years ago. And old people like me know if they worked in music industry, OD2 ran a very early subscription streaming service.

But it was download. You've got download credits on subscription, time limited download credits on subscription. So I was, as a poor student, I was paying 15 bucks a month to get something like 100 time limited downloads. It was crazy. So I've always loved music and tech. My undergraduate degree was completely unrelated. But I somehow managed to find every opportunity to write about this space in my undergraduate degree. So like one of my last pieces was about the fact that, you know, we had a technological convergence and we were going to have a thing called the iPhone and it was going to be seismic.

I should say I wasn't psychic. It was, you know, at the point that Apple had published the pattern. So we knew it was coming. So I wasn't, you know, I wasn't, you know, I wasn't psychic in any sense of the word. So I said, gee, I graduated. And without a job, it just so happened that I lived in a neighborhood where the major labels were nearby. And as a jobless new graduate, I had quite good data skills. I'd even been asked to run my course's data class. Like I'd been asked to show my peers how to gather data and to manage data.

So I was obviously good with data as a brand new graduate landing onto the job market when almost 20 years ago. And so I didn't have a job. I got a random call from a recruiter asking if I wanted to go in as a temp to work at Warner Music Group. And they were offering a derisory salary. So I said, you know, Warner sounds great because I love music, but I'm not working for that salary. It's insulting. It was basically what I was earning before I'd gone to university. They came back with a slightly improved offer and I went to work at Warner as a temp.

And so my first job was essentially working in what was called then digital reporting. So it was like trying to make sense of what was then particularly bad digital reporting. Because you think sort of 20 odd years ago, so 18 to 20 years ago, there was the big platforms in Europe for digital were mostly mobile platforms. So like Vodafone, T-Mobile, et cetera, had ringtone stores, et cetera. And so, and the reporting was abysmal because those stores predated any digital supply chains.

So, you know, the reporting was just junk. It was like translated titles, no ISRCs, no grids or other identifiers. And so my first job was just trying to make some sense of just trash data, basically. And so that's how I got started. That is, that's such a nerdy way to answer music. Did I not say? I'm definitely a nerd. No, I'm a nerd as well. I appreciate it. Actually, the funny thing is, like a slight kind of tangent, but I grew up in a neighbourhood that is very musical.

Like I live around the corner from XL Records and Rough Trade is around the corner and, you know, Warner is around the corner and so on and so forth. And I knew people that worked in the music industry as a kid because of where I lived. I always assumed there would never be a job for me in music, despite that I loved music, because I wasn't musically talented and I was a geek. And all my sort of experiences talking to music industry people were, they were really cool. They were A&R or they were like managers. And I was a geek who liked technology. And so I never thought I would work in an industry despite my passion for music because I thought I'm not cool like that.

I was very lucky that I graduated at a time where the music industry was starting to, I wouldn't say embrace technology, but we're starting to work with, having no other choice but to work at sensory technology. So it was perfect for me because I arrived as a jobless graduate into a market that I needed. It's like me. No, it's amazing. You know, a side comment, just for people out there listening to podcasts, it can be weird being sort of in the music industry environment because there's like these artist-centric sides of the music industry that is very much about the cool factor.

I've always felt outside of that. But in reality, it's just, it's a very weird excluding environment. So never mind. That was a sidetrack, but I definitely have some experience with that. But I talk with a lot of startups and a lot of them are building platforms that wants to integrate music that needs to have some licensing. They want to build everything from streaming platforms to engagement platforms, whatever. How would you, in the most practical way, not talking about AI, but like practically go about licensing music for whatever you do?

Can you walk me through those scenarios, those processes? How do you go about it? So for traditional music, online services, let's call them, you need to understand where they're going to launch. And that's because different markets have different laws. And then how the service works. So what kind of functionality the service will have? Those two questions are decisive in the actions you take. I think it's probably worth me explaining at a very high level.

If you launch in one country, in some instances, your licensing can be very easy, particularly in markets like the UK. But in some cases, also in the US or one European, other European country. But once you're a multi-territory service, your licensing becomes inherently much more complex. And then the question of functionality, that plays into a big factor because at some ends of the market, so services, online services look like a radio service, non-interactive, you know, the users, the linear listening experience.

At that end of the market, sometimes, not always, you can tap into collective licenses or statutory licenses in the US. And so the two most important questions is where and the functionality, where and what the service looks like. And so until I know those questions, the answer to those questions, I can't really help people. But once I understand those things, I can, sometimes I can sway them to reduce the number of countries they launch in because it makes their licensing infinitely easier. Or sometimes I'll say sort of trim back your functionality a little bit.

It's so close to letting you use a collective license. But those things aren't really necessarily in my interest because using a collective license is mostly quite simple. If you aren't collectively licensing and you're licensing directly from rights holders in a multiple territory situation, licensing becomes very, very complex straight away. And that complexity is different market to market. And different from recorded rights to publishing rights, unfortunately, as well. And so I help them navigate that.

So I have a questionnaire that I run through with possible people I'm going to work with. And if they're a single country service and they're just doing radio-like experiences, then that licensing can be really simple. Within the UK, for example, you can get two, a publishing license, publishing blanket license and a recorded blanket license for radio-like services from PRS and PPL, respectively. And you can be up and licensed in a few weeks without a lot of my time. At the other end of the market, your multi-territory services, particularly where there's something like a derivative work kind of experience.

We can talk about that more in a second. You know, services where the user gets to interact with the music in a way that changes it or feels like a synchronization. At that end, particularly on a multi-territory basis, we could be talking about years of licensing time to get it finished. Oh, that's interesting. And obviously, we don't have time to cover all of them. But just to sort of understand the basics. So you have this simple example of radio-like licensing in UK that's pretty simple. What would be sort of the range of price points it would cost for startups to do everything from that to the most complex things?

What kind of numbers, bullock, are we talking about? So you mean in terms of cash or in terms of revenue shares? Well, if there's a model there to be explained with revenue share, please also explain what the scenarios are. So it varies a great deal from market to market. So the UK, actually, I was saying to a friend last night who works for ICE that the UK in particular is quite cheap to launch in because to give credit to PRS and PPL, which is not something I do very commonly. But, you know, to give them credit, they have made their blanket licenses accessible to small companies.

And so PRS, for example, their minimum fees for a year start at something in the region of a few hundred pounds. Okay, wow. And PPL is similar. So, you know, it's maybe a little bit more, but, you know, you could probably get those licenses yourself directly in some cases for under a thousand pounds or dollars even maybe in some cases. If you follow, if you strictly fit within the remit of those licenses, any divergences and you have a problem. But, you know, in some cases, like in case of the PRS license, you can take it out online, you can pay with your credit card, it's a few hundred pounds, you're done.

In those scenarios, sometimes they don't even need my help. I don't believe in forcing people to pay you for things they can do for themselves easily. So I do encounter startups where I do myself out of work and say to them, look, you can do this for yourself. It's really easy. And just give them a link to the forms. In those scenarios, if they tap into those collective licenses, and there's lots of caveats, but in the UK, for example, you would be looking at a royalty rate that's effective at about 40% combined between publishing and recorded rights.

That is to say a PPL license for recorded rights for the UK only is likely to be about 25% of revenue and your PRS monde should equate to about 15% with minimum fees that are recoupable. That's a highly simplified explanation. In practice, you're paying a minimum fee, you're buying a, in case of PRS, you're buying a number of plays or a number of downloads. But it should equate if things go well to about 40% of your revenue.

That the UK is probably one of the cheapest and simplest markets to launch in for those things. Lots of other, particularly Anglo-American or continental European markets have something similar, but I would say that the minimum is a much higher. So, sort of places like Italy, Switzerland, Belgium, Holland, sorry, correctly, the Netherlands. I'm pretty British there. I don't know why. I do normally use the Netherlands, anyhow. Some of those markets, their minimum thresholds are quite a lot higher.

You know, sometimes thousands a year. I think Switzerland is equivalent to $1,000 roughly per radio station per year. So, quite quickly, the minimums are like $4,000, $5,000 or $10,000. And that's just for the publishing rights. So, that's the easy end. Starting as little as maybe $1,000 for everything and paying 40% of your revenue. And the other extreme end, well, I mean, depends who you are.

The minimums are, there are some absolute minimums, and then the minimums also scale as your company scales. So, you know, when I've represented Samsung or Sony or Intel or who else have I represented as big? They're pretty big. They're pretty good. We get the picture. When I represent big corporations and they want to do something with music, those minimums aren't still the same minimums that I would get from my little baby startups.

They are different minimums. And depending on who you're talking to, they scale differently. There's not a linear scale. It's a very, like, haphazard scale. But, you know, depending on what you want to do, if you're trying to do something very ambitious, your advances might be more substantial again. If you're doing something that's very vanilla, hopefully you can keep the advances down. But, you know, in some cases, you know, I have negotiated advances for services. So, like, let's say, imagine it's starting and it's the first year. The first year advances, in some cases, have been tens and millions of dollars.

Oish. Now, I should say, in those scenarios, you're talking about directly licensing and you're talking about a revenue share that equates to between 65% and 70% if you're successful. I say if you're successful because if you're paying $40 million in advances, the only way you're paying a 70% royalty share is if you recoup all of those advances. And in the reality, as even companies of Spotify's size has seen, it's actually really hard to recoup tens of different advances completely.

So, in practice, even if things go well, you know, in early years, Spotify's royalty rate was effective at about 73% because despite going amazingly well in terms of growth, paying lots of people little advances, it becomes very hard to recoup them properly because you maybe overpay one person and underpay someone, but you still have to top the other person up. So, your royalty rate, historically, when you're a smaller service, your actual effective royalty rate is way over 100% in a lot of cases because you're not acquiring customers as quickly as your advances suggest you are.

But even when you've got good scale, advances are challenging to fully recoup. So, I'd say if you're very successful and you're like Spotify, you might get a royalty rate down to about 67%. But that's like if things have gone really well and you've got excellent licenses and you haven't overpaid at all, you might get all-in licenses at 67%. That sounds like a nightmare, to be honest. I mean, you and Gareth were talking the other week about challenges of raising money for music tech broadly.

This is a huge challenge for companies raising funding for streaming or other digital consumption services. The idea of going to investors and telling them, we would like you to give us $5 million and we're going to hand over three of it on day one to the music industry. You can see why investors don't have a lot of appetite for that, unfortunately. Yeah, and there's no value creation of that $3 million. That's just, you know, the premise for even doing anything. I should say it is possible to do it for under $3 million.

Yeah, of course. You interviewed Jonas from Pacemaker, or historically from Pacemaker. So I was a chair of his board. I became that chair because I did his licensing. That's how we got to work together. You know, we managed to get Pacemaker licensed for a number that's substantially lower than $3 million. Well done. You know, volumes below. But it was still a decent amount of cash. And so I've debated this with my peers a lot. I think if you're licensing an interactive, interesting service that needs direct licensing, the best you could conceivably do anything for is probably a few hundred thousand dollars.

And at that rate, you're not going to get a perfect service and you're not going to get everything perfectly done. But you will have licensed as much as you can, and you'll have some major label content, and you'll have multi-territory licenses, and there'll be a whole bunch of holes and a bunch of risk in your licenses. And so I should say it's not that you can't do anything with at least $3 million to $5 million, but in practice, you can't do anything other than radio-like services for under a few hundred thousand dollars. This might be a controversial question, but how much does relationships affect the pricing?

It's in my interest to say massively that everyone has to hire me because I've got all these relationships. You'll save money using me. Yes. I don't think it... So the royalty rates are largely set by market precedents. Yeah. And even if you're friends with Lucian Grange, the reality is he's probably... It wouldn't be him giving you the royalty rate anyway. It would be one of his lieutenants, Michael Nashall, one of Michael Nashall's team at Universal. In practice, you're not...

Even if you're best friends with the CEO of a major label, you probably aren't going to get dramatically different royalty rates or even dramatically different advance requests. Because in a market where Spotify, Apple, Amazon, Google are licensing music, they would be nervous to set a precedent in market that's dramatically different to what they consider is what they hold out as a fair market price for their music. And so I think what relationships get you is not so much a big discount, but you're probably more likely to get a faster response because, you know, in people like myself, we have personal relationships with people.

So we can... You know, it's often very hard to get rights holders to engage enough. It takes a very, very long time to get complex music licenses. And so the relationships help you get people to engage more. But actually, I'd say the bigger value, even more than that, is actually helping you understand the right space. I described it earlier. It's like people like me can help you understand that if you make this small tweak to your service, you can license it much quicker and much cheaper. Or if you remove this bit of functionality, then your service would be much easier to license because it's much less sensitive.

Or, you know, explaining that the service is going to be really complex to license because it looks like a derivative service and therefore you're going to impinge on moral rights of writers and performers. And so I guess what I'm saying is actually I think our advice, people like me, our advice is less about the relationships, though those are valuable, but actually about how we help you work out what to do strategically for your licenses and your product to get the best outcome. But also, like, making sure that if you spent $300,000, you have something to show for it at the end.

Because there are far too many startups who run out of money before they get any licenses. And I think, sadly, I try and make sure that I understand up front what my clients have as one way and what their ambitions are and then work with them to build a plan that's sustainable. I'm going to ask a last controversial question, then we actually go into AI, which is what I'm going to be the big nuances to licensing. You know, I can imagine that, you know, a lot of these startups, scale-ups, companies from other industries, blah, that does licenses, pays a lot of advances, but a lot of these advances might never be recouped.

Who sits with that pool of money, that advance that's not allocated to any access to rights holder? It's a controversial question. So, those advances are always, pretty much always sitting with a rights holder of some kind. Because, I mean, I'm trying to think of an example. I don't think I can think of a single example where the advance is stacked with the platform, because it doesn't make any sense the whole point of an advance is you're paying it over.

So, what happens to it? There's an interesting question. There is a moral answer, and then there's a reality, a commercial reality, which is getting closer to the moral answer, but isn't always aligned perfectly. So, let's assume that, you know, imagine a story where you've paid over 10 million advance, and you've only basically redeemed, recouped half of it. You've basically earned 5 million in royalties, or spent 5 million in royalties, and now you've got 5 million unrecouped that you've paid over.

You're not getting that money back, really, in any circumstances. I have, on very rare occasions, and it is very, very rare, seen you being given a little bit more time to try and recoup it. Like, you know, carried over for a few months, maybe. Or, you know, I think one or two PROs on occasion have given us another 6 months or something like that. But, generally, that's lost to you as a platform. That money is gone, and you're not getting it back. It doesn't matter, you didn't recoup it, tough luck. How it's then reported and or paid to the rights holders, either customers or members or artists, varies from company and country to country.

I would say, historically, when I first started, there was probably far too much of it not being reported or paid out at all, and just being additional profits to the bottom line of these companies. Less, I was about to say less PROs, but that's not fair. There's been some pretty notorious sort of things like arrests into PROs. You know, PRO staff have been arrested for some of their conduct around this episode. Oh. I would say it used to be far too much of it was just, like, profit or, like, sort of absorbed into other business operational areas of these companies.

In fact, I don't think that's really the case anymore. Like, you know, if you look at royalty statements from PROs, you will see that the big PROs for sure are paying a portion of that unrecouped money out as essentially like a bonus, essentially, based on previous usage in some form. They call it proxy. So they basically, you know, pay out some of the royalties based on previous. So, you know, if you're a songwriter, you'll get an additional bonus 5% based on what you earned previously from that platform.

And I think in a lot of cases, that's also true on the recorded side. Now, you know, people like Martin Mills from Beggars have been quite aggressive at pushing the narrative, particularly in the British market, for, like, making sure the industry does do right by its creators. And so, I say creators because I mean that as, like, all people who create music in some form. So, you know, things like, you know, all of the major labels and, well, I say all the major labels got Spotify equity, for example, as part of their license.

And most of them have paid it out in some form to their, to their writers now. Also, in that case, sorry, a recording artist. Sorry, mixing up my stories. That's good to hear. But you talked about Jonas, a common friend, pacemaker, brilliant guy. He was running a company that also had something to do with AI. So, can you, you know, people want to go back and listen to that episode, you should.

It's an amazing story. It's actually a really interesting flow and the conversation would sort of end at the end, which is kind of cool. But tell me a bit about Jonas' company. How did you guys go about thinking licensing with an AI startup? And then also, walk me a bit through what is it differently in general about this new technology and how does it affect licensing? So, I should say, I only joined pacemaker in 2019. So, there were years before me and a bunch of people, including Jakob Key and Jonas, had crafted a bunch of this stuff ahead of my joining.

I think Jonas said in his podcast that he was originally using the Spotify API. So, licensing thinking until that point was sort of, I wouldn't say simplistic, that would be unfair. It was removed from the coalface of licensing. It was just about high-level principles and working with Spotify. So, I came on board because Spotify no longer wanted to support that. And so, they had to get direct licenses. What was different with pacemaker is we were undergoing those licensing discussions late 2019, early 2020, just as COVID hit.

I mean, that didn't help, but ignore that. AI wasn't yet a hot topic. And AI is heavily used in that technology in a lot of really, really interesting ways, including some ways that you would actually now describe as generative. Yeah. But we were licensing ultimately a streaming service, a streaming service that enabled users to mix their music. What was kind of exciting in the AI use case, and I think Jonas, I should say, I don't think he gave himself and the pacemaker team enough credit.

He's very humble. He's super humble. One of the things they were doing, which was really fascinating in my mind, which I would describe as generative, was it's a streaming-only service. There's no download element to it. And so, when you're mixing, there can be issues with buffering. And obviously, you run a podcast, so you understand what I mean, right? So, what they'd actually built was this amazing technology, proprietary technology, which would enable the mix to be extended in the scenario that there's a problem calling the next song.

Oh, wow. They had, like, essentially, they were using generative and machine learning AI to extend songs for maybe five or ten seconds in a kind of natural-sounding fashion to fill the gap until the next song was finally loaded enough to play it. Like, because you can't stop a mix midway through and be like, hang on a second, we're just loading the next song. No. I should say, you know, in practice, people like Spotify have managed to get their buffering to the extent that you don't really get a meaningful silence between songs anyway.

But, you know, they had some really cool AI technology. But in practice, in those days, we just licensed the streaming service. We licensed the streaming service with descriptions of what we were doing, and nobody asked us how we were doing it. So, we never actually said in much detail AI. We just said, we are doing these things. And we described accurately what we were doing. But nobody really asked us, like, what the technology was underlying in it that enabled us to mix dynamically on the fly, to create DJ mixes.

and experiences and enable people to share DJ mixes with their friends with full commercial catalogs. So, Jonas and Pacemaker was such a, sort of, they were so aggressively innovating in the space, but get very little credit for, like, the innovations they brought about. It's amazing. But then, and I agree, I absolutely adore Jonas. I'm very lucky to jump what I call with him every three weeks. So, he's very generous with his time. He is.

But, AI now and AI licensing is fundamentally different. Or at least, what's happening in this space right now? Or what's not happening in this space right now, which might also be the question. So, over the last three to four years, we've gone from being, sort of, largely a seep of the wheel on AI to aggressively lent in. So, I think it really blew up in the last, particularly last two years. So, I'm just trying to think, I'm trying to explain it.

It's like, you would, historically, if you were licensing a service, you'd just license a streaming service. I think that's still kind of true. AI, if it's a part of a streaming service, you just license a streaming service and the AI is kind of irrelevant. But, what's different now is we're moving into a world where we have to envisage licensing generative AI. And that's where it gets very, very different. And I think we should, we're going to kind of ignore the fact that in most major jurisdictions, the laws aren't entirely, the laws and the legal precedent aren't entirely settled on whether you need to license generative models that train on copyright.

And lots of very smart lawyers I know, even they don't seem to agree on whether Anthropik will win against the music publishers or whether Getty in the UK, for example, will lose or win against Stability. And I'm not going to comment on that because I'm not a lawyer. But, if we envisage a world where, in major market, it is decided you do need to license large language models when they're training on copyrighted material, then it's a fascinating question.

Because I talked about quite how tough it is to license streaming services. Well, you copy all of that complexity and then magnify it for a number of reasons. Because in an AI generative world, there is, you're creating new works, let's assume, from the old works. There's a lot of debate about this, but I'm trying to simplify it. If we assume that these outputs are derivatives of the input commercial music.

Now, that's not a settled matter, by the way, that is an open question. But if we just jump forward to the scenario, wherever that is, that is the scenario, then you've got to license those inputs. And licensing those inputs is more complex than streaming services because it has all of the same rights, plus more. Because the outputs could be very different. Those derivatives are, I mean, they could frankly be inflammatory or racist or they could be anything, right?

And so inherently, they are going to be at least controversial from a creator perspective. So you've got people like the CMM in the UK. They put out like an AI manifesto where essentially they're saying, you know, they agree with the rights holders that AI large language models using copyrighted material needs to be licensed. Everyone's agreed on that, on at least the creator and the rights holder side.

But where the difference comes is, you know, if you're a council of music makers, you want to represent the rights of your members, you want your individual members to have a say on whether your music is used in these generative models. Because the outputs are, in theory, at least in this argument, derivatives of your music. And you might not like how they've used your music. Now, AI companies will argue that outputs are so far removed from the inputs that they could not be derivative works.

That's unsettled legally. I think it's, I think morally, they are related. Whether you can identify the output as being a derivative of the input. Most cases, if the AI company is smart, you can't determine that the output came from the input. Because it's being diffused through a model. And often it's, the output is the output of hundreds or thousands of inputs. But I guess the, I wouldn't say the moral, but I should realize the moral means a couple of different things in this discussion.

But I personally believe that it would be a tragedy, it's a big word, isn't it? Tragedy, to think that we can have a sea of unlimited generative music, free of copyright, that was made on human creation. Because I think it will savage artistry, human artistry, going forward. And so, you know, in that scenario, let's assume it has to be licensed.

But now, it's not just about licensing it from the people who hold it right, the labels and the publishers. But it's also about, in theory, getting permissions from the people who wrote the music or even people who performed the music. And that's, frankly, terrifying as a concept, practically, how do we operationalize that? Because anyone who's licensed a music service, a digital streaming user service will know that actually a lot of our licenses are quite vague about what rights they have, particularly on the publishing side.

And so, you know, this AI, so building a world where we are licensing AI models feels like something we have to do. But it feels like we're operationally very, very far from it. There's a number of companies out there who are starting to offer sort of, essentially, AI data and training solutions and data sets. But really, what you'll see right now is they're largely based on production music sets. Yeah.

Because that's simpler. Saying, yeah. I have some thoughts in this space that is a bit controversial. And it comes from, so the context is, when I went into music, I stumbled across music. And I'm very good at music. I have a master's in music production. But I've said it a few times in the podcast, but I don't have a particular love for music compared to other creative things. You know, movies, I would say I love movies more than I love music.

Oh, interesting. Yeah. But for me, music has- Surprising as well. I'm fascinated and I love the music industry because of the complexities, because of all the narratives, because of what it creates. But I'm very much in love with the creative storytelling. And oftentimes, I feel that more masterfully done in films at scale. Like, there's not as many films released. And it's easier for me to be catered, high-quality master of storytelling that is for music, where it's more of a jungle.

This is sort of how I see it. Which also means that I have a part of me that's also just a consumer. So, you know, so when I look at what's happening in the AI space, I have two sides of me. I have the emotional and intellectual understanding that this is not right from a moral standpoint. But then I also have a consumer side of me that are like, this is amazing. Like, what's happening, technology progression. And for me, I truly believe that if everything was rightfully licensed, which it should be, I understand, what is happening right now would not be possible.

Like, people are stealing, 100%. I do not condone that. But I also recognize from a consumer mindset that it's only possible because they are stealing. And I think that's an interesting dynamic to live in as both a consumer, as a music industry person, and living in that space. I don't really have a point with it, but I think it's not spoken about that much, though. I speak about this kind of complexity a lot because, actually, if we look at where we are in the streaming space, a number of our largest streaming platforms are companies who started out unlicensed, like YouTube, because they were able to hide behind things like the DMCA protections for user-generated content or, you know, mere conduit protections in other markets.

And so, actually, licensing first is a barrier to innovation. It's basically, you know, what you'll describe. Absolutely, it's a barrier to innovation. You know, I specialize in, you know, advising people about licensing. And I do sometimes have to say to them, like, I don't think, I don't know how they'll ever do this if you license first. But, you know, in the era of sort of early YouTube, et cetera, those platforms were relying on protections that actually really weren't developed for them originally, right?

Like, a lot of them, a lot of, like, DMCA protections were centered around, like, early days piracy and the fact that digital service, sorry, you know, ISPs, so internet service providers, were trying to ensure that they weren't sued for the fact that their end users were pirating music. Now, obviously, we'll ignore the fact that's evolved extensively in the last 20 years. But if you think of it as the world, companies like YouTube managed to rely on protections that were designed for a slightly different purpose.

And they got aggressive scale almost 20 years ago now, right? Because they could use protections that were built for another use case, essentially, because they looked closely, they looked similar enough from an legal perspective. And there is no such protections, at least, obviously, for AI companies. So, like, in the UK, there is, like, certain protections that, but essentially, they've agreed that those, well, not they've agreed, it has been made clear that the things you could argue might protect you for training are not going to cover you.

And so that there is still an open question. If we move to a license first scenario, is it going to stifle innovation? For these companies in AI? Yes, for sure. And it will be a really fine balancing act. And I worry about it greatly. To be clear, I worry greatly about the supply of streaming services. Well, I say streaming services, but I mean as a non-generative consumer consumption services.

We have a very, very small pool of new services coming through because it's very complex, as I put earlier, to license streaming service. And as you and Dareth said, it's very, very hard to raise capital at the moment. So there's already, like, a really poor funnel of new innovations in core streaming and or digital consumption. It would be naive of us to think that going to a licensing first approach isn't going to massively stifle generative innovation.

It is. And I worry about it. And I spend a lot of my time, at least my emotional energy, sort of lobbying to make it better, the current world. Because if the current world is better, it will help innovation more broadly. I should say, to some extent, though, while we're still in this area of legal uncertainty around AI training, if I was a rights holder, I also wouldn't be licensing generative models in any meaningful fashion.

Because the outputs in most jurisdictions are deemed, if they're fully machine-made outputs, then they don't benefit from copyright. And so then you have no control over them and no strong basis to monetize them from. So if you're a major label, you license your whole catalogue to a large language model now, and you might get, I don't know, 50 or 150 or 250 or even, like, a billion dollars, right?

I don't know the exact number because I've done some of these models, but we've never got to those kind of check sizes. But, okay, great, so you've got a check for hundreds and hundreds of millions or billions even. But now the generative outputs, you know, that's probably a fixed fee. You don't get any recurring royalties from the outputs because they're free of copyright, or at least you don't get any substantial royalties. You might get a bit, you might get some royalties for new generative pieces of work, but the outputs are no longer connected to your inputs.

So you're not going to generate, because there's no copyright on them, and so all you're getting at best is royalties from the model on an ongoing basis to people's use of the model. You know, if people pay to build, to generate content, you might get more royalties from that. I don't think it is wise, actually, if you're a major label or major publisher, or frankly, anyone of any commercial success. That's really interesting. To put your music into these models right now, while we're in this sort of situation where those generative outputs are free of copyright, but there's nothing stopping them competing with you for consumers.

Because when you look at the power structures in the music industry right now and the recorded music industry, it is catalogs. It is, you know, what has come before and what it brings to the table now. So how do you want to remove that model that's kept you strong and, you know, help financing all new ventures? Yeah. Also paid, you know, like it's, you know, I've got, I said I grew up in an area, it's just full of like musicians and recording studios and executives and stuff. I've got family friends who've had commercial success musically, you know, if suddenly there's no monetization of their music anymore, because let's imagine that lots of consumers probably will end up consuming generative works that are free of copyright.

So therefore you're going to get very, very limited royalties going forward. Of course, that's a massively simplistic, the actual scenario for consumers is more nuanced than that. But in that scenario, it's deadly for future of human creation because creators need to pay their bills too, right? So I guess that point about, you know, creators moving from listening to humans versus listening to machine generated outputs. I definitely know lots of people, my peers, who would argue that'll never happen.

I disagree. I think some people, like music engine executives generally, they care about the album and the artist and definitely some consumers do. I remember as EMI, we segmented our consumer market and I think we said something like 5% of overall consumers are super fans. Those people really care about making sure that their money gets to the person that they're listening to. And they care about the identity of the person they're listening to. And they care about, you know, they align themselves. You've got artists, you've got bands like Radiohead, where like their fans will be fans until they die.

But is the average kid out there listening to TikTok caring that there's a real human behind the song that's on that video of that girl dancing? I don't think so. And if you look at late last year, not early this year, there was an announcement that, I think it was Stim, one of Stim's writers. He had, I think he was like number three Stim writer in terms of income. And he writes for Epidemic Sound. So it's production music. So he's writing under pseudonyms and patronyms, et cetera, et cetera.

So he has no artist profile. He is a writer that's essentially writing production music. That is comparable in a end user's mind to generative outputs, right? It is not a real identity. It's, you know, the artist's name is something that's made up. He's number three writer. Yeah, I think it's 15 billion streams. I think that was the last number. Yeah. But also things like functional music. Like, you know, when I talk about people not caring about artists, of course, if you're very young, you care even less generally.

Because you haven't, you didn't grow up with CDs and stuff like that. And maybe you're not even old enough to go to gigs yet. So you probably care less about the artists. But actually things like functional music are also massive with older generations, right? Like, you know, I don't think, I think it's very naive to think that consumers won't listen to generative music in large volumes. Particularly if it's completely free. Yeah. And our time is finite. Well, our time here is finite. But end users, there's only 24 hours a day.

Oh, are we stopping now? Okay, sure. I was actually having an interview with Tom Fritz on the podcast a few weeks ago. And he sort of works with neuroscience research in music and music and health. And one of the conclusions from Autoxx is, according to their research, these are the numbers that I put myself, is 50% of people's appreciation of music is purely acoustic. And if they don't have a cultural, contextual, or, what was the word he used, like interaction with music, that is, our perception of music is truly acoustic.

Which means there is, from a scientific level, going to be, assuming that quality of AI music is going to be comparable to whatever it is and what context it's played in playlists, there's going to be no differentiating factor at all. And I would even argue that some of the AI models are getting really close now. Like, it's, like, 10% off. Yeah. We've leapt ahead, for sure. Like, I remember early days, like, when I was still on the phone, so this was almost 10 years ago, we had a company-wide demo weekend.

Like, just hackathon, basically, internally. And they used some open generative models then, this 2015, I think. And the stuff that came out of it was abysmal. Really, really terrible. Like, completely unlistenable to. And to come forward now, what you listen to on Suno and Udeo, I don't really, or Udeo, rather. I don't really like any of it. But is it possible that it would be good stuff behind a TikTok video, for example? Absolutely. And you can't also ignore that much of the funding for these companies, whether it be Anthropik, Suno, Udeo, even OpenAI, is coming from the biggest companies of the world.

Yeah, it's coming. But, you know, I've talked openly about this on the podcast. I do a lot of, I call it research. It's the easiest way for me to understand it, but I use a lot of these platforms. So I think in Suno, I've done like 7,000 tracks. Wow. So, and I do, I do, I try to do steering, which is basically trying to figure out what prompt triggers a output that's very close to an input quality. And I can do that now. So I have, I have made, I would say, close to 50 tracks that are in this thingable school from your music.

I have sent, like, I've sent Shangri experts certain tracks in their style. And that's like, cool, who's this? Like, there's no initial perception that this is AI generated. So I think the people trying to hack the system, it's not even hacking, try to specialize in understanding how the prompts work, are already now 90% there. And I, you know, I come from music production. I know what I'm talking about. And some of this is way better than any of the writers I had signed at one point in some of my publishing companies, which is scary.

Well, I guess we have, what, 30 million music creators? And I say music creators instead of not professional musicians, but people making music and putting it out there in some fashion. That's roughly the number, I think. Yeah, it's natural that a lot of them aren't going to be musically talented necessarily. And maybe if we imagine even one million of them are musically talented, I think most of them won't get any commercial success. There is a lot of music out there that isn't great. Well, it's a statistic something like 45% of music on the biggest streaming platforms has never been listened to.

Yeah. You play it out, but you don't even listen to it yourself? That's kind of wild. Like, you know, but I'm with you. I think outputs are increasing in their quality aggressively. And I think we are very much likely to end up in a world where soon enough they'll be... I mean, you're saying they are already indistinguishable. I guess I'm finding it hard to say that because they don't sound like anything I would ever listen to. But do they sound like they could have been created by a human? Yeah, I think so.

Particularly the low of five stuff and the stuff that sounds more like production music anywhere and stylistically, massively. It's a worrying direction. Did you see the work? I mean, you haven't had Ed Newton Rex on the podcast, have you? No, not yet. You know who I mean. Obviously, Ed did some really quite damning analysis of Suno and UDO's outputs and how they related to what he suggested were probably some of the inputs. And it was...

I mean, some of the outputs were so, so close to the inputs. Not all of them. But, you know, I think... It was a pretty quick muddle update after that article, though. Yeah. Yes, well, so I've worked with Ed when we were doing some work at Stability. On Stable Audio last year, which was a licensed product, which was using production music. And he's very... He's super smart. To me, it felt like both Suno and UDO came out, I guess, less...

Both of them came out within the last year or so. It felt like they were very naive in their approach. Absolutely. Because they came out and didn't pretend they hadn't used copyrighted music. They were intentionally quite vague. Or, in one case, I think one of them actually said, yes, of course, we've used it. Or rather, something to the effect of, if you want the best output, she put in the best inputs, I think was actually the quote. But they could have been much, much smarter in the way they limited both inputs and outputs. You can put in, you can block certain inputs, you can block certain outputs.

You know, the engineering of inputs and outputs, if you're smart, would mean that you wouldn't have given someone like Ed the opportunity to prove your inputs. I would say it wasn't, he would admit himself, it wasn't absolute proven, but yeah, it's a fascinating area. And, you know, there's a discussion about the fact that, you know, this idea that the outputs are not derivative works because it's a transformative use. And they are so different to the inputs that you can't possibly suggest they're derivatives.

And we're not keeping a copy, so it doesn't matter anyway. It's basically the AI defense large league. But then you look at the analysis that Ed did, I think it was for MBW, on both Cineo and Udeo, and well, that's just not true. Like, the outputs were clearly derived from, in some form, the inputs. And so, you know, I think it's going to be a fascinating time. And I really do hope that we end up getting to a world where major jurisdictions decide that you do need to be licensed to train.

I guess then we get into a world of, like, how on earth do you license and should you license if you're a successful label or writer or anything? And should you license if the outputs are free of copyright and hard to monetize? Wow. Yeah. It's such an interesting point. And I think we sort of close the conversation here because we've covered so much area in this.

And we should have spoken a couple more hours. And I think we'll just have you back in a couple of months. And when we have some more clarity in this point, I normally don't slug what I'm going to do in the future. But I just want to do it here in the podcast because next week I'm releasing an article about, it's very arrogant, but it's my prediction for the future of creators in music. What's the environment going to look like? And it's based on more than 20 conversations here in the podcast, a lot of off-the-mic conversations. And sort of gathering what I believe to be some of the foremost thinkers in this space and making, like, a prediction.

So it's a nine-piece article with prediction that I'm going to plug. I'm going to also do a podcast introduction of it with my AI co-host, AI Jake, that's been on the podcast a few times before. It's going to be a live interview where he interviews me about my findings. So I'm just going to do this plug now because I think it's very relevant. And I think, Becky, let's do a small experiment. I'm going to do this article. I'm going to have AI Jake dissect it. And let's meet in three months. And let's see if we can see if there's some things happening within licensing that's going to support or not support some of my claims.

I think that could be really fun. Sounds like a plan. Great, Becky. Well, thank you for your time. You're amazing. And the people around you love you. And I think it's so fun that here in the music industry, like, these strong personalities that give so much to the world around them, they stick. And you do that. So I'm very lucky to have you on. And we'll talk soon. Thanks so much for having me. Thanks so much for having me.