AI is on a collision course with music — Reservoir’s Golnar Khosrowshahi thinks there’s a way through

Photo illustration by Alex Parkin / The Verge The publisher behind the songwriting copyrights to some of the most popular music ever recorded doesn’t think AI spells doom for the industry.
Today, I’m talking with Golnar Khosrowshahi, the founder and CEO of Reservoir Media, a newer record label that I think looks a lot like the future of the music industry.
You might not have heard of Reservoir, but you’ve definitely heard of the artists it works with. Everyone from A-ha and John Denver to Evanescence to Joni Mitchell and even legendary film composer Hans Zimmer. What makes Reservoir different is that Golnar built the company through acquisitions.
You might think of a traditional record label as sending people out into dive bars to find new talent and then break new artists. But Reservoir doesn’t do any of that. It buys catalogs of existing hit songs from established artists. Reservoir owns the songwriting copyrights to about 150,000 songs and another 30,000 copyrights in master recordings.
As Golnar explains, Reservoir thinks of those individual songs as assets, and after acquiring them, the company sets about monetizing those assets in various ways. This is a copyright-based business in an age where copyright is under a lot of pressure. There are the familiar issues on social platforms like TikTok and YouTube. And now, there are new existential issues created by generative AI tools. You’ve all heard of fake Drake. Fake Drake is going to upend copyright law one way or another, and it’s the music industry that traditionally leads the charge in pushing back.
Golnar founded Reservoir in 2007. She’s been in this game for a long time, and she’s seen a lot of tech-related changes come and go. But now, there are huge companies and private equity firms pouring tons of money into the same catalog-based business model, sometimes with disastrous results.
You’ll hear Golnar say all of her competitors seem to have a billion dollars to spend on songs. And you’ll hear us talk about a company called Hipgnosis, which spent a ton of money building a giant catalog but now has very unhappy shareholders after that catalog wasn’t worth what they were promised.
Of course, Golnar and I also talked about AI and training AI systems. A lot of record labels are very unhappy that their copyrighted work is being used to train generative AI tools. And a lot of artists are even more unhappy that their voices are being used by AI systems. Golnar has some surprising thoughts here and one very instructive example she brought in to share. It’ll definitely make you think about who should be making money on what and when.
If you’re a Decoder listener, you know that I love thinking about the music industry. Whatever technology does to music, it does to everything else five years later. So paying attention to music is the best way I know to get ahead of the curve. I also just love music. Golnar is herself a musician. She obviously cares about music a lot, and she’s clearly given a lot of thought to what happens next. So this was a great conversation.
Okay. Golnar Khosrowshahi, CEO of Reservoir Media. Here we go.
Golnar Khosrowshahi, you are the CEO and founder of Reservoir Media. Welcome to Decoder.
Thank you so much for having me.
I am really excited to talk to you. Reservoir Media is kind of like a new-look music label. It’s newer than sort of the big giants that are floating around. You have a different approach to the business model, and as I was just saying to you, I really think that whatever happens to the music industry happens to every other kind of creative industry five years later. I’ve been saying this since we started The Verge.
So I’m always trying to figure out what is going on in music right now because then I seem like a genius when it happens to TV or Hollywood five years from now. And there’s quite a lot going on in the music industry right now. The whole industry is restructuring itself. There’s AI to talk about. Distribution on the internet seems to be changing. Marketing seems to be changing. It’s a lot, but let’s start at the very beginning. What is Reservoir Media, and why’d you start it?
You’re characterizing us as the new look, but here we are, a precocious teenager at 16 years old. We started as a publisher, and we built the business from there, acquiring rights. It really wasn’t until a few years in that we bought a very small recorded music catalog, Philly Groove, and then it really wasn’t until we acquired Chrysalis Records in 2019 that we got into the label business. So our business really is focused on publishing, and we are looking at sort of revenues that are 70 / 30 publishing versus recorded.
And just for the listener, publishing is songwriting.
Songwriting, we own copyrights, we own the intellectual property. We own the words and the music, and that’s a really, really important distinction versus the recorded music where you own that actual sound recording, that specific recording. When you own the publishing to a song like “Take Me Home, Country Roads,” there are 500 covers of that song on streaming platforms. So you are monetizing that every time that copyright — those words, those notes — are used, regardless of whether it’s me or you or John Denver. Master recording is only… we don’t own the master recording rights to that song particularly, but it’s that particular execution. So that’s the distinction there.
And then when you bought Chrysalis, you own the masters from Chrysalis, and you own Tommy Boy Records as well—
Correct.
… then those masters. So you have a pretty broad base of copyrights.
We do, and it’s been interesting growing that business, and we’ve had incredible opportunities come before us. The Tommy Boy opportunity was just amazing, and then what we did with it afterward and with De La Soul, and the release of all that music was really a pivotal point career-wise, personally and professionally.
When I say it’s a new-look kind of company, I recognize it’s been around for a while. What interests me the most is you have built the company through acquisition of really successful existing catalogs.
Yes.
When I think of a record label, I think of I’m going to go discover an artist, we’re going to pair them with a songwriter, we’re going to put them in a studio, we’re going to go market the hell out of some hit singles. Oh, look, it’s Justin Timberlake. We’ve done a thing.
That was the ’90s approach. That was basically the approach that I grew up in understanding the music industry. We discover talent, we market the talent, we create new talent, we burn them out, we find some new teenagers, and off we go again. That was the music industry. You’re really focused on, “Okay, there’s a bunch of catalogs that are really valuable, and we’re going to acquire them and then we’re going to get value out of them.” Is that a fair characterization that you’re kind of on the other end of it?
I think so. We never really got into the business with the intention of developing a frontline label that was going to develop artists like Justin Timberlake. That’s an expensive proposition… we’re slightly more risk-averse than that.
We look at buying these existing assets and what we can do from a value enhancement standpoint and what that translates to in the form of organic growth. That’s what’s really important to us. That’s how we believe we’re building long-term value for ourselves, for the artists, for the songwriters, regardless of who that other rights holder partner is that we have in the equation. And every deal is different. That continued value enhancement is what’s really important to us. And yes, you’re right: we’re not out there finding the next thing. That’s a different business, and I’m not sure how effectively a company our size, with our risk profile or appetite for risk, can compete with others in that area.
This is a really fascinating fracturing of the industry in my mind: on the one hand, the discovery marketing of new artists, the social platforms are just completely encroaching on that. And I hear from young artists, “The labels won’t even talk to me until I have so many followers.” I have a person I went to college with [who] literally got a record deal by buying followers on Twitter, and then that got him some attention. This is crazy. This is a thing that has happened in the past. You’re saying we’re going to leave that aside. That seems like the big labels’ business. We’re going to go find successful artists and help them get more from their existing catalogs, help them gracefully retire. What’s the pitch there?
Every pitch is different, and it really depends on where you are on your creative journey and where you are in your life and are you 30 or are you 60? Are you wrapping things up, or are you thinking about estate planning and legacy, or are you looking at what your next portfolio looks like?
So, every deal is different, but I think the pitch from a bird’s-eye standpoint is we want to be your creative partner in your musical journey, whatever that means. That could be a “go forward” deal, that could be, “We’re going to buy the back catalog, and we’re going to partner with you and enhance the value of this catalog.” It really depends on who the client is because our goal is to be in business with people long term, and I think that if you cater to and satisfy people’s needs and their objectives, then you can have a pretty good shot at staying in business with people.
You have quite a long list here. Joni Mitchell, John Denver, De La Soul, Evanescence, Hans Zimmer… It’s a pretty intense range. Some of these artists are pretty current and relevant, some are older artists, quite frankly. Do you have different strategies for each of these catalogs? You say, “Okay, we have the publishing right to a Cardi B song — our songwriters wrote the Cardi B song. We can go out and license those for television or advertising. But then you’ve got Louis Prima, and I’m assuming that’s a different strategy.
It is a different strategy. I would rather say that the strategy is different for each song versus the writer.
How many songs do you have in the catalog?
150,000 on the publishing side, just over 30,000 on the recorded side.
Okay.
So this is a little bit of a bigger theme in that we do everything on a song-by-song basis. We actually assess value on a song-by-song basis. A lot of people assess value on an income-type basis, but if you look at somebody like Hans Zimmer, you’re looking at a portfolio that dates back to Driving Miss Daisy up until films from a few years ago. Very, very different films. You can’t really look at these assets. You can’t look at Driving Miss Daisy and Kung Fu Panda as the same thing. They have a different audience, different music, different genre. It’s all very different.
So I think it’s really important to look at song by song, especially because you’re dealing with songwriters who have this range and who could do all kinds of work. And so the pitch is different. The pitch is different for every song, but the pitch is focused on film placements, and it’s focused on advertising placements. Digital licensing is super important for us. We were the first independent to actually do a direct deal with YouTube. That was back in 2014, I want to say. So we do things at a very granular level, but we can because we only represent 150,000 songs.
Yeah.
It’s the benefit of being nimble and independent and small.
So there’s a huge surge of money into companies that look like yours—
Everyone has a billion dollars.
… like KKR and BlackRock and Blackstone — not the grill company, the huge hedge fund. We’ve had the CEO of Blackstone, the grill company; we’re working on the hedge fund. But these are huge private equity funds just dumping money into these catalogs. They have a pretty shaky investment thesis. I can never quite figure out why they want to own Pink Floyd’s catalog or whatever. I can kind of understand your investment thesis. It seems like on a song-by-song level, you can evaluate, “Okay, here’s the earning potential [of] this song.” But in general, is there a model that you use that says, “Okay, we own 150,000 songs, we paid X dollars, we can expect Y return.”
Sure, we can get to that top-line number, but the engine behind that is the song-by-song information, and that song-by-song information can be organized by our top earners. So we can say, “Look, this many songs contribute to 80 percent of our revenue.” And that’s going to give us, “We paid X. This is now worth Y. Here’s our return.” But it’s still the engine behind it is song by song because you’ve got portfolios of songs where you have significantly more valuable songs in completely different genres.
I think the investment thesis for a lot of private equity is that you are looking at uncorrelated assets that behave very much like annuities, that have dependable cash flows in an industry where the data around growth and monetization is becoming more and more compelling every single day. That’s in Western markets, coupled with compelling data in the emerging markets. So that’s a pretty good investment thesis. And then it becomes a pretty great investment thesis when you’re operating in troubled macroeconomic times. I think that’s the motivation there. It’s also… it’s fun.
Well, there’s that part of it.
It’s music.
A bunch of rich guys want to buy some Boomer music catalog. I get that part of it. The thing you’re describing, we’re going to describe massively important cultural works as annuities, and those are going to be safe investments in troubled… Isn’t that a little bit heartbreaking? To me, as somebody who is very much a creative, and you yourself are a musician, there’s something heartbreaking about that, right?
There is, and there isn’t. There’s something heartbreaking in that something that someone considers their child, which is really how these bodies of work are thought of by their creators — is thought of as an asset, has that label “asset.” So that’s heartbreaking.
And commodified—
And commodified, 100 percent.
… inside of that language switch is a commodification.
Exactly. It has economic meaning around it. At the same time, it’s really also nice to see creators be recognized for their work in the long term. That’s not heartbreaking. That’s great to see that their art, their child, has that kind of longevity.
Yeah, I think this is the central tension of the entire music industry. In particular, I feel like Hollywood people are like, “We want to be rich.” And many musicians are like, “We would prefer not to think about it.”
Right, and so when are you hitting that edge of selling out versus—
Do you think the artists care about selling out anymore? So I’m a huge Clash fan, and there are stories like when they bought their first cars, they were in a moral crisis, and I just don’t think the artists do that anymore.
I don’t think so. I think that you have enough people in the music industry who are revered for their art doing these kinds of deals such that the stigma associated with doing the deals and, hence, selling out is… Maybe it hasn’t disappeared, but it certainly… it has been reduced.
So the value of the catalog over time, you described it — the private equity guys think of it as an annuity. I know that you care about them as deeply cherished pieces of art — but if you describe a catalog as an annuity, we’re going to buy all of John Denver’s catalog.
Okay.
This is now going to generate returns at some rate for some amount of time. At some point, the fan base is getting older, maybe spending less money, maybe harder to advertise to. And you’ve got younger fans who need to be initiated into John Denver’s music, but they’re on TikTok, and they’re flooded with a bunch of new music. Is there an expiration date on these annuities?
I think about that a lot because I have two 21-year-olds. So, are they going to be listening? Do they even know what this song is? And where I have come to is, a great song is a great song, and great lyrics are great lyrics. So [the] next generation may not know John Denver, may not know his story, may not know this song, but if Google is taking this song and putting it in a commercial and launching Google Home during the Super Bowl with this song — an instrumental version, no lyrics — I feel that it’s going to continue to have staying power and relevance. There were several different iterations of this song in The King’s Man — I think maybe five different iterations.
Wow.
That’s a film that was catering to, probably, an 18 to 54 demographic — a wide demographic. So, again, people might be listening to that music and not knowing the history behind it, but it’s a great song with a great melody and great lyrics. I think that relevance will continue.
And do you chart that on a curve? The relevance will continue, but the instant hit of nostalgia that makes you buy a Google Home will decline for the younger audience. So the value of the song by song will decline.
So we don’t really look at it that way. We look at a rate at which cash flows will be generated over time. We have certain periods where there’s a plateau. You see spikes every so often as it relates to something like The King’s Man event or some kind of big film sync event. And then we sort of smooth that out over time. So that’s really how we look at it. And something like that is really in a mature state, and so it has plateaued.
You have some competitors in this world. Everyone has a billion dollars, as you said. The big one was called Hipgnosis, which, just this week, is in a moment of bizarre controversy.
Yes.
And because it is a British company, the British press is covering it. They’re having the time of their lives covering this controversy. We’ll link to some of the stories, but basically, they wanted to sell some of their catalog to another part of the fund. They had to take some offers — nobody bit. So they made the sale, and their shareholders were like, “No, this is a disaster.” The founder might leave. You could take that as evidence of owning these catalogs does not deliver the returns that it costs to build them. What do you see this as evidence of?
I can’t comment on somebody else’s business.
Fair.
What I can say is that we take a conservative approach as far as how we value these catalogs. We look for compelling data around growth. Some things may be subject to growth, and some things may not. And we tailor that to the assets that we’re looking to buy. We take into account our cost of capital, which, keep in mind, we did business for 13, 14 years with zero cost of capital to today’s cost of capital. That’s a very different environment to operate in. We had no cost of capital for that long. In fact, we had only known that environment.
Just to unpack that for the audience, you’re talking about the zero interest rate environment—
Exactly.
… and now it’s very different.
We only operated in a zero interest rate environment, and now we’re not. So we, too, take that into account, and it’s important for us to create long-term value and return for our shareholders, and we’re conscious of that. So we’re never going to do a deal that doesn’t have the numbers that qualify for us making that investment.
We’re really passionate about music. We are all musicians, somewhere along the way, we are trained. We love music. We’re really dispassionate about our investments, and that’s important. I know that that’s how we have run the business thus far. That’s how we hope to continue to run the business. I think people have gotten very, very excited about music, about music assets, about subscriber growth. There’s just been a lot of press, a lot of education, and sometimes you might, what’s the saying? Get over your skis. So perhaps that’s the situation.
That was very polite. That was a very polite way of describing what’s happening to your competitor right now—
I’m a very polite person, and I would say that bad news in music is bad news in music. There was a lot of traffic last week around accruals, for example, and how people were accounting for accruals over the course of the past year with the new royalty rates. That’s bad news for music. Now, there is no issue around that, but do you see what I’m saying? We want generally… Investor positivity around music is important.
Yeah.
I want to ask a couple of the Decoder questions, and I want to talk about AI, and I want to talk about where copyright law is going because those things seem to be crashing into each other, and you are very much at the center of those two ideas. But real quickly, tell me how Reservoir is structured. How have you organized this business?
We are a very lateral organization, and we have been since day one. We continue to grow that way. I think that at some point, you get to a point of tension where it’s not possible to continue with that flat hierarchy, but we haven’t hit that yet, and it hasn’t been a problem for us. But that’s a lot to do with our culture. We’ve had zero senior-level management turnover since inception. That has a lot to do with our culture. It’s a very open environment with people who have grown in lockstep with the company, and I think that’s just been really important. And I’m both proud of and indebted to my team. That’s a little bit about how we’re structured.
But get into it. So the function of the company: you go, you make acquisitions, and then you license them out.
Yeah, it’s that easy.
Right, yeah, but I’m just saying mechanically—
And then I go home. [Laughs]
… you buy some stuff. You do a bunch of licensing deals. You must account for where the revenue goes. How is that structured?
So we buy some stuff.
[Laughs]
When we buy the stuff—
See, on Decoder, everything’s very simple.
That’s fine. I’m going to decode it for you. When we buy the stuff, there is a team of six people who are analyzing stuff. Those six people are really working during that deal process. Once we complete the transaction and we finish all of our underwriting and we finance the deal, we then have an ingestion process. That ingestion process is kicked off with a meeting that involves all the different functional areas so that people know, “Hey, we just bought this asset. Here are the rights that we’re getting with this asset. Here’s what we need to do with ingestion. Here’s what we need to do with administration. Here are the rights that are not covered. Here are the territories that are not covered.” Every deal is specific. “Here’s how we’re going to ingest it into our licensing system, etc.”
From there, digital licensing sync, etc., has a lot to do with actually taking the metadata associated with the music and then starting to pitch it out and register it and claim it and do all of the things that we do on the digital side. We also have marketing involved the entire time along the way because they will announce the deal. We will think about when we’re announcing the deal. There’s obviously some relationship building that needs to happen if it’s a new client, not somebody that we’ve been in business before somehow or another. And there’s a lot of creative discussion around what we can do with the catalog, especially if it’s a relationship where it’s catalog and an ongoing relationship with an artist. So that’s something that we’re focused on with our creative team. And then anything else that’s specific to that deal.
At the same time, everything’s being loaded up and ingested by our royalties team to prep for next royalty runs and how these people are getting paid based on what rights, etc. It’s a lot of very, very detailed work that goes into buying the stuff, getting the stuff, selling the stuff.
Right. You’ve had to operationalize quite a lot of licensing deals and royalty deals.
Yes.
That’s up against the platforms, right? In large part, you’ve got big digital music distributors: Spotify, Apple, TikTok, YouTube. Do you have to build technology or have relationships to reconcile how those platforms think about royalties?
We haven’t had to build technology, and we’re not software developers, and I don’t see a world where we do that. I see a world where we are always going out and buying the best technology that we can to actually accelerate our systems. So, for instance, we use DISCO to manage a lot of the music and how we license the music, etc. Relationships are very important with our DSPs [digital streaming platforms], and we are always looking to cultivate and evolve the relationships that we have there and continue to just improve and make the licensing more efficient and have relationships that go across marketing and playlisting and all kinds of things like that that can be beneficial to our catalog.
Because you have catalog, when you talk about playlisting and marketing, you’ve got catalog — you don’t have new stuff. A lot of the pressure on those services is promote new stuff, break new artists. Do you have people who are saying, “Okay, we got to get some TikTok dances going on John Denver?”
We do.
You do? How does that work?
I mean, it works exactly that way. We have people in marketing whose expertise is looking at how you can… I don’t know if it’s John Denver, but—
I just like picking on John Denver. I imagine Cardi B is a little bit easier than John Denver.
But we have a team that is pitching this all the time. Again, the record side of our business is not the largest part of our business, but it’s… As you said earlier, part of your value is determined by your numbers on social media, whether you’re a new artist or an existing artist, a new catalog, existing catalog, those views, those eyeballs that you have are contributing to value.
Do you look at something like… the joke I keep making is that I’ve been trying to get away from Fleetwood Mac since I was in high school and—
It’s not working.
… I can’t escape this band. They just keep coming back. Do you look at something like the resurgence of Fleetwood Mac?
Absolutely.
And say, “Okay, this is an opportunity for us. We have to operationalize that and figure out how to get value out of our catalog.”
Absolutely.
What does that look like? What does that marketing meeting look like?
I don’t really participate in that marketing meeting, but you look at music… Cigarettes After Sex, that’s music that has had a viral moment that certainly was not their audience. And yet when you attend a show that they had, say, in the last month, you’ve got a completely different demographic than who was listening to their music because of the viral moment. So those are the types of events that you can’t predict them. You can’t plan them down to a very granular level, but you can certainly start developing opportunities around them. Not everyone’s going to work, but that’s what that team is focused on.
Alright, last Decoder question. Maybe the most important one. How do you make decisions? Do you have a framework for making decisions? What does that look like for you?
I mean, coin-flipping?
Yeah, that’s mine. I feel like every CEO on the show wants to say, “I don’t know, man. I just make it up.” And then they tell me about Amazon or something. But what’s yours?
I like to sit on things. I think it gives you perspective. I definitely see myself changing my mind. So when I have time, I really take the benefit of that time. I’m super conscious about not focusing on the wrong thing and not focusing on the noise. When I’m making decisions, I feel like I’m looking at a bucket of stuff, and I need to take everything that’s irrelevant out that’s really clouding my judgment. And once I do that, I do certainly have a lot of clarity. I walk through a lot of consequences with decisions because I think that real-life walkthrough helps how you think about the outcome. So that kind of sums it up.
An enormous set of decisions you’re going to have to make in the future is around AI, is around fighting through the courts to change copyright law, is about deciding whether to sue some of the generative AI companies around training. I want to come to that because I think that’s a huge, thorny problem that you are well suited to speak to. But I first just want to get a sense of how you think it’s going in the industry right now.
So we just had Larry Lessig on the show. I know you listened to that episode. His view is that fair use in the music industry is the worst of it all: that the things that I can do in text by quoting The New York Times are just not allowed by the norms of the music industry, and there are infinite stories about publishing rights getting taken away from songwriters because they quoted a song. There’s the Ed Sheeran lawsuit, you name it. It’s bad. And we’ve come to the place where major record labels are setting up songwriters with catalogs to do interpolations in new songs. So all new songs are now built on old songs because everyone’s terrified of copyright lawsuits. Is that workable? Is that a steady state? Does that need to get fixed? Because I feel like if we don’t talk about that part first, the AI conversation just begins on really shaky ground.
I don’t think any of us are changing copyright law anytime soon.
Yeah.
The people who would be leading that charge from my conversation seem to be hesitant about going down that road because of the unexpected results that could come about in other changes in that exploration. So it’s almost the devil you know is better than… What’s the saying? I’m very bad with my idiomatic phrases.
The devil you know is better than the devil you don’t. I think it’s that simple. I might get it wrong, but I think—
That one.
… that’s the idea.
What you know is better than what you don’t know. So that’s a little bit of the conversations that I’m getting. I think you’re always going to have these types of infringement lawsuits come out of the woodwork. I do think technology is going to help predict what’s happening there. I think that we might live in a time where songs get delivered and they actually go through some kind of assessment of likelihood.
You have to tell the computer how to measure that likelihood.
Right.
And that feels to me like right now, the likelihood is a bunch of copyright owners like yourself deciding that the value of the catalog is suing Ed Sheeran for a new song that has nothing to do with the catalog but might sound alike or share a chord progression. And that feels like it’s on the shakiest ground right now.
That’s very much on the shakiest ground. That’s not enough. I mean, there are a finite number of notes, chord progressions, combinations. Every creator today, consciously or subconsciously, has been inspired by something in the past. The extent to which a work is a derivative of work is what is that gray area and where there needs to be a decision. I think we’re always going to have these types of infringement cases. The degree to which they have legs to stand on today has certainly been compromised, I would say.
Yeah, it feels like fair use in music is more of a coin flip than ever before, and the risk of going to court, maybe you’ll get the “Blurred Lines” verdict, or maybe you’ll get the Ed Sheeran verdict.
Right.
And that is a total coin flip. I couldn’t tell you why Marvin Gaye’s estate—
Except—
… got the money in “Blurred Lines” and didn’t in Ed Sheeran—
Because Ed Sheeran was in the courtroom.
But that’s what I mean by the coin flip. If you can only win if you provide Ed Sheeran and not on a legal basis, to me, that’s weird. [As] an ex-copyright lawyer… and what I see is the music industry is like, “Screw that. We’re going to build our own private copyright law system where we’re going to do all these side deals, and people will never really know.” And you see young artists have to give their publishing to older artists just to avoid the controversy.
Well, I mean, whether that has to do with what’s happening today for a young breaking songwriter, that’s often been the cost of entry. So I don’t think that’s necessarily unique.
But for you as the catalog holder, do you see that as, “Okay, this is potentially an avenue for us. We can go assert our copyrights against a new artist because they sound like one of the songs in our catalog.” Is that a line of business for you?
No. I mean, we may have revenue that comes from that line of business, but our business is creating new, to an extent possible, unique music and cultivating the careers of artists and songwriters. That may become a revenue line item for us, but that’s not one that we’re going to build a team around to go and chase.
You have competitors who are, right?
I know.
The big sort of faceless private equity companies are like, “These are my chords.”
Okay, and say there’s, I mean, you pick the number, there are 100,000 tracks being added to the streaming platforms a day, 150,000. Whatever that number is, it doesn’t matter because it’s just so many that nobody has the capacity to listen to that, but the consumer is not clamoring for that nameless, faceless content. So until there’s market demand for something like that, I’m not really sure why we would go and develop it. Other people can develop it. If we have IP that has an interest in that new work, great. But as I said before, we’re not in the software development, technology development business. We’re in the music creation business. So we’ll continue down that road with people and artists and creators because we believe that people’s connection to music is a human experience, and that still has value. So that’s where we are.
Well, that is a refreshing values-based approach to this. I appreciate that. I look at the industry, music industry, as famously not a values-based situation, but that brings us right to AI. Where you completely divorce people from it to even make these kinds of value judgments. You have big companies ingesting tons and tons of material to train various kinds of models, to do various kinds of things, and then you have a series of escalating problems. Do you think training the models is fair use?
So I loved what was said in the last podcast, which was training is free.
That’s what Professor Lessig believes.
I know, and I kind of agree with that personally. Now, I may be saying something completely wrong from a business standpoint, something I shouldn’t be saying. [Laughs] But how is that any different than reading a history book, than reading a science book? It’s training. You’re learning, you’re educating. If that training results in a product that is then commercialized and monetized, that’s a different story. Then that IP is standing on the shoulders of IP before it, and so that creator should have recognition.
So that’s a derivative work, in your opinion.
A hundred percent.
So the problem is, I understand it, and I’m very curious for your understanding, is that if you ingest the whole of recorded music in history and then generate something — a song, a voice, a vocal melody, whatever it is — there’s no way to trace back the works that that was derived from, right? You have the massive ingestion—
I think there is a way—
You think there is a way.
… I think there is a way. I think that you need to essentially hit an inflection point where a work is enough, like another work where there can be some IP, some ownership attributed to that other work. I don’t know how you determine that threshold, whether it’s core progressions, notes, etc., but there has to be some way to give credit. I generally agree with what you’re saying in that all of these ideas are based on a past idea. All of this new music is based on something you’ve heard before, but you’ve created something new. But this is a little bit different. Can I tell you my test that I ran this morning?
Yes. I’m very excited about this.
Okay.
I want to tell the audience: Golnar walked in here, just glowing with excitement about this test. Go ahead.
I asked Claude—
That’s the Anthropic AI?
Correct. I asked Claude to write me a song about country roads leading me home—
[Laughs]
… and Claude wrote me a song that has a very nice first verse. The chorus of the song that Claude wrote for me goes like this (maybe it sounds familiar): “Country roads, take me home, to the place I belong. West Virginia, mountain mama”—
No way.
“… country roads, lead me home.”
Well, Claude. Where do you think Claude derived that from?
I don’t know, but it might have something to do with—
There’s an Anthropic lawyer right now who’s just freaking out in their car listening to this. I just want you to know that. That really happened?
It’s right here. I can show you. I printed it out.
Let me see this. Yeah, that’s just “Country Roads.”
I also printed out the real lyrics so that we could—
Oh, very good.
… compare, compare and contrast. It sounds like a high school assignment.
So, in your mind, that’s just completely unfair, right? Claude should not do that. And you own—
Claude can do that, and that’s fine, but if this is recorded and released and monetized and commercialized and used in the next Google Home product launch—
But let me ask you this: I don’t think Claude right now is a subscription product, right?
It’s not.
When they start charging 20 bucks a month to use Claude like OpenAI does with ChatGPT, should they pay you then?
Yes.
How much should they pay you for that?
I don’t know.
I mean, but this seems like the problem, right?
Right. I really don’t know.
I pay Spotify an ever-increasing amount of money a month at this point, and I can calculate to some pennies based on how much I stream from various artists how much might go somewhere.
Right.
It seems impossible to do that with an AI.
Yes. Because there’s so much volume of what would be generated and so much unique volume, whereas those songs are you and this many other people are going to listen to X. Yeah, I haven’t thought about what the economic model looks like, but clearly, this lyric is based on something that the Danoffs and John Denver wrote at some point.
When you see that, do you get a call from the John Denver team saying, “What is going on here? They need to pay us for this.”
When I see this, I copy and paste it and put it in the digital licensing Slack channel, and then I see what happens with it.
So, an interesting aspect of this problem, like I said, it’s sort of unsolved. What does the deal look like? It could be a flat fee. It could be, “Hey, just recognize when you’ve obviously derived something from our work and pay us a percentage.” Who knows what that’s going to look like? You might get one deal from Anthropic, you might get another kind of deal from OpenAI, which is backed by Microsoft, and a third deal from Google, which owns YouTube, which is, as you said, a key partner of yours.
How are you managing thinking about all of that? And Anthropic has a bunch of money from… I think they have Amazon money now, right? Also, a key distributor of your work. Is that part of the puzzle here? “Okay, we’re going to go in through the music parts of these organizations are all gigantic companies and tell the music executives to tell the AI executives to play ball.”
This is a dynamic that is ongoing in the music industry where you sue each other during the day, and then you have dinner together at night. So you break bread, and then the next morning somebody files a lawsuit you didn’t know about that was going to happen. It’s common to have to juggle these different relationships and different level of tension in the relationship at any given time. We just have to navigate that. And to some extent, I also recognize that they’re navigating a similar set of tensions on their side.
So, we are all in the business of monetizing content and putting that content in front of the consumer. What I think that value is versus what somebody at DSP thinks that value is obviously differs, and hopefully, we come together and get to some agreed-upon value. This is going to be a little bit of the same thing. And it’s no different than all of the tensions we’ve had along the way with the DSPs and the settlements that we’ve come to. And it’ll be a continued battle because we continue to fight for creators, and they will continue to fight for minimal cost on that side.
The off-ramp for a lot of those battles, like you said, you’d have dinner at night, you see each other in the morning. The off-ramp is the legal system, which at least in the case of music until very recently kind of knew what was going on. I’m trying to imagine just explaining to the average federal judge what happened there with Claude and Anthropic and telling the judge to reach an outcome that is stable or reliable. And I don’t know that that is possible at this point in time.
But the off-ramp isn’t the legal system—
Okay.
… and here’s the history on that. We go to dispute royalty rates every five years. We go through this proceeding with the Department of Justice. We’ve now concluded CRB IV; we’re going to continue. So the legal system is decided on a royalty rate and how that royalty rate is going to change within a five-year period. And when that five-year period is done, we’re going to go back to the table and demand for greater compensation for creators. So, yes, the legal system made a decision, but we’re still going to continue that battle because that decision has a finite term around it. So I don’t know if that’s always… That off-ramp, we’re never getting off because we’re always going to go ask for more.
This is the problem you signed up to solve every day. How do we get more money for the catalog?
Exactly.
But inside of that, just this week… This is a very newsy week to have you on.
It is.
There’s Hipgnosis.
There is.
Spotify is rumored to be changing its royalty rates, so you need to have a minimum number of streams to get paid.
Correct. And this is all in this endeavor to remove a lot of content from the platform.
They want a lot of garbage off, and they know the AI stuff is coming, and there’s white noise chum all over the platform. You have a catalog again of older artists. Do you worry, “Okay, at some point, the numbers song by song, some of these songs are not going to get enough streams to get payouts — we need to negotiate with Spotify for deals to keep the whole catalog up”? How does that work?
So we don’t have those specific deals. We’re a much smaller company, and so we don’t have those specific deals. I do worry about what you’re saying because you have a finite number of listening hours, and you have a growing content pool. And at some point, you have these really, what are standards, getting in front of fewer and fewer and fewer people and ears. I think that one of the ways we look at the industry is that the diversification across platforms has actually been accretive on a net basis. So you have Peloton, you have all of these new platforms that have actually supplemented consumption across the other platforms.
So Peloton plays music in its classes. They obviously pay licensing fees for that reason.
Correct. And I mean Peloton specifically, TikTok specifically, these are also discovery channels. So now people are discovering our music that way because they heard it in a Peloton class, because they are down some TikTok rabbit hole and are seeing a video of something. So that diversification across gaming and other social media platforms has been beneficial for us. It’s always been accretive. We are always seeing this push and pull in our consumption and across these platforms. So we have to supplement with other value-add vehicles and channels to continue to build that value.
We’re coming up on time. I want to end by talking about an op-ed that you just wrote about AI.
It’s obsolete now.
Well, you wrote it. It’s good. You have a lot of ideas in there. The big idea is: “Don’t be so worried. The industry will solve this, like it solves everything else.” You have one idea in there that I just want to ask you about specifically, which is maybe AI will enable micro-licensing, where the AI will not just generate the songs but it will also detect bits and bobs of the song and will pay fractions of pennies to various—
Correct.
How do you think that would work in practice? What I always worry about is the kid on YouTube who doesn’t know what they’re doing and they’re just putting their music on YouTube. And then the industry comes and takes all their pennies away. And once you add computers and automated enforcement to art, it seems like it gets really dicey. But I wanted to understand what you meant by micro-licensing.
Micro-licensing, really think of it as outside of the big-ticket sync that we do. Outside of the trailer placements, the film placements, the big commercials, etc., we do a lot of micro-licensing: high-volume, low-cost music licensing. The way I think about how we could do that is that we have technology that actually understands all the characteristics and attributes of our music that we license for that use. So, understands that metadata, classifies that metadata, and is searchable. And even though it’s searchable now, I think it can be a lot more efficient as far as being used to then search, create a license, obtain that license, and you’ve now completely automated a process that is still carried out by humans right now.
I got it. Okay.
So that’s how we think about micro-licensing as much more of the high volume, low price point licensing that we do.
That would be for advertisers or Instagram brands or whatever.
Exactly. And a lot of commodity music. That’s how we think about it.
The other thing that I’m constantly thinking about in the context of AI is we don’t have a great system for fake Drake, right? There is no federal likeness law. There’s no right of publicity law that would cover it.
It’s another big news item this week.
Constantly. You’re in it, perfect week. How did you arrange all of this? Universal is basically solving this problem by telling YouTube to make some stuff up and have some private copyright enforcement system for Drake. YouTube will extract whatever concessions it wants to. That’s not necessarily going to be available to a smaller label. It won’t necessarily be available to an independent artist without a label yet. It’s not auditable. If YouTube makes a mistake in enforcement, there’s no way to stop it. If a kid just wants to sound like Drake and it’s not an AI, should that be allowed? Lots of questions there. How are you thinking about that kind of idea?
So, right now, we have two forms of protection. One form is that… One of the reasons that the DSPs actually took that particular track down was that it was a violation of their agreement between Universal and the DSPs. So there was a breach of contract there. The other is that there’s a lot of attention around and disapproval just around this ability to create something that it sounds exactly like something else. Now, that’s neither here nor there. What happened last week is the presentation — I think it was last week or two weeks ago — was the presentation of the NO FAKES Act, which actually is an acronym that stands for name and likeness and all—
Congress loves this.
… Yes. And that’s being led by [Chuck] Schumer, I believe. And that’s really what’s in play right now. Obviously, it’s not tomorrow, next week, this year thing even. But—
I would say our current Congress is not well suited to passing a complicated AI regulation—
Exactly.
… or doing anything.
Yes, but at least that process has begun, and there will be a tension on that. I don’t know if I have to believe this because of the business I’m in, but I really don’t believe that we are operating in a business where the creative work and the artists and those creators are… The value is just going to be rendered zero.
Yeah.
I don’t, so I’m going to have faith in the system and the people and an appreciation for art.
What do you hear from your artists about it? What are they worried about? What are they asking you to fight for?
They worry about their skill set being rendered obsolete, but at the same time, they are seeing efficiencies created in the studio with things that they can do and things that they can source and how engineering processes are accelerated and things like that. So there are a lot of B2B applications that are just not being talked about that are positive. The focus is on this uncertainty around “the machines are going to take over our lives.” That’s just not what I think is going to happen.
What do you think the near-term future of the music industry looks like? What should people be on the lookout for? We’re in a time of great change. Where are the markers for you?
I think that we’re in a time of what’s going to be a lot of creativity. I think we’re in a time where a lot of people are focused on this glocalization concept and a complete breakdown in barriers and just—
Wait, you have to say what glocalization means. It’s a great term.
… Well, it came from Will Page’s summary that… A white paper that he had written. And you have this world where all of these barriers are down, and people have access to all different kinds of music. So you have local music that’s actually having a global impact. That’s been really interesting to watch because we do have a business in the Middle East, and we represent a lot of artists out there. There’s going to be a lot of momentum in what’s happening on that side of things, and we’re interested in watching that. We’re very focused on how culture moves, how culture moves from East to West and West to East. And that’s something that I’d like to see shift and change.
That’s great. Well, Golnar, I could talk to you for hours and hours about this. I hope you can tell, this has been great. Thank you so much for coming on Decoder.
Thank you very much for having me. It was really an honor.

{Categories} _Category: Takes,*ALL*{/Categories}
{URL}https://www.theverge.com/23938498/golnar-khosrowshahi-reservoir-media-ai-music-copyright-publishing-decoder-interview{/URL}
{Author}Nilay Patel{/Author}
{Image}https://cdn.vox-cdn.com/thumbor/7snkUAL1evBvhk2cQSwlku2loac=/0x0:2048×1365/1310×873/cdn.vox-cdn.com/uploads/chorus_image/image/72813209/236892_Golnar_Khosrowshahi_Reservoir_Founder_CEO_WJoel.0.jpg{/Image}
{Keywords}{/Keywords}
{Source}POV{/Source}
{Thumb}{/Thumb}

Exit mobile version