Michael: Yeah. All right, I'm excited to be here today with my new friend, Matthew Adell. So, Matthew co-founded a company called Musical AI, and they help to solve the issue of attribution and licensing in generative music, which, holy cow, what a world to be involved in, Matthew. Like, this is a major disruptive movement happening right now.

He's previously led or built major music platforms like Beatport and MetaPop, and they are focused on scaling AI music tools and helping to bring this technology to a wider market. So, Matthew, thank you so much for taking the time to be on the podcast today.

Matthew: It's my pleasure, Michael. It's good to meet you.

Michael: You too. All right, so, gosh, where should we start? So, I would love to hear a little bit of your journey and background for working with platforms like Beatport and now, as you find your way to Musical AI. Could you share just briefly about your journey and kind of what led you to wanting to solve this particular problem?

Matthew: Sure. I mean, my journey is a long one. 'Cause I'm pretty old. I got my start working in independent music and labels in Chicago in the early eighties, and after a successful run at Wax Trax! Records in Chicago, where we signed the likes of the KLF and KMFDM and My Life with the Thrill Kill Kult.

I founded a house music label in Chicago in the very early nineties. That was my first founding of my own company. And after doing that for quite some time with some success, the internet happened. I went to work for Motorola, building really early music streaming and consumption technology platforms.

A lot of patent kind of work. I was the young music guy then, and got my start in product management as well then, and eventually found my way to MusicNow, the very first music subscription service based outta Chicago, and built the first multimillion-track library for subscription consumer use.

We had an exit there. Found my way. I'll skip a few things, but found my way to being head of music and product at Napster when it was legal, when we were paying people. Did that for a long time. Really enjoyed it. We sold that company as well. And after that, I returned to my first love, dance music, and was lucky enough to join Beatport as their Chief Operating Officer.

The founders had built a really impressive company that I was anxious to join. I became the CEO after a while, ran it for five years, helped the board and founders sell it to SFX. And then I founded a company called MetaPop, which is a community site for aspirational music creators. I sold that to Native Instruments, the music software, music creation software company.

And so throughout my entire journey, I've sort of been at the tip of the spear on trying to figure out how major digital transformations in the music ecosystem would work, in a way that allows the music ecosystem to thrive. And the first thing required to allow the music ecosystem to thrive are great songwriters, great musicians, and making sure they get paid.

And that's been my focus throughout my career. And about two and a half years ago, I met my co-founders from Musical AI and saw what they were working on and was lucky enough to have them ask me to join them. And we've been working on Musical AI for about two and a half years now.

Michael: Hmm. Wow. Even busy.

That's a, I have been [an] impressive track record. Okay, cool. So, Musical AI. Could you share a quick overview for anyone that's listening, this is their first time connecting with it, of the platform, and what would you say is like the main value proposition or problem that you set out to solve?

Matthew: Sure. Well, the problem I set out, we set out to solve was that generative music was gonna be probably the biggest disruption to music creation, music consumption, and making sure artists and songwriters get paid. And we set out to make sure the ecosystem can work in a way that allows the ecosystem to continue to thrive.

And my colleague Nico, our CTO, has invented a generative AI attribution technology, which is sort of the core of what we do, and what we do is we sit in between AI companies and music rights holders to make, to enable them to do business together in a way that's transparent and trustworthy and compensates the underlying works that AI companies train on in a way that isn't just based on volume.

I've always believed that the music that creates the most value for the consumer or end user deserves to make the most money. Just like in streaming services, the stuff that gets listened to more gets, makes more money than the stuff that doesn't get listened to more. That seems like a fair way for things to work.

And so our technology allows us to monitor the training data or inputs. And then monitor the outputs from generative AI system and attribute value to each and every output based on the inputs. So for every single output that comes from one of our AI company partners, we can say, well, this song in the training data had a 21% influence on the output.

This composition in the training data had an 8% influence on the output, and so forth and so on. And basically what you have there is a transparent and consistent way to figure out how to share the dollars that come back out of AI companies. So AI companies work with us. We help them conduct business with rights holders to get training data that's licensed and available to them.

They train on that data, and then once the AI company is out in the market serving their subscribers, we collect a piece of their revenue. Just like a distributor might collect a piece of the revenue from a DSP, and we use our attribution technology to determine how much of that revenue goes back to each and every rights holder based on the value their music and composition has created for the AI company and their subscribers.

Michael: Hmm. Wow, that is wild. I mean, I've been thinking about this quite a bit as it relates to generative AI and the outputs and just try to wrap my head around neural nets in general. And I'm curious to hear your perspective as someone who probably understands this stuff a lot better than me.

The mechanics of how, like, when I think of this challenge right now. I would love to hear if you have a good read, a pulse on the heartbeat right now of where things stand as it relates to publicly trained data. What's fair use? Are these still ongoing conversations in music like that? We're still defining, like, what is okay for AI to train based off of? I'm, as I'm assuming, based on this platform, you'd lean in the direction of, well, you know, you should be required to pay a license to be able to access and train on this data.

Is that a view that's reflected by the wider market right now, and what's your perspective overall on, yeah, like how these AI models are being trained?

Matthew: Sure. Well, first and foremost, not a lawyer. Second, importantly, not a software developer. But I know just enough to have some opinions.

Obviously our business is based on the assumption that nobody should scrape copyrighted data off the internet and build a business off of that, period. Certainly in the case of music, no business should be in the business of taking copyrighted music without permission and creating a billion-dollar business out of it without licensing that music and, through the process of a license, coming to some agreement about how the owners of that music and those compositions should be paid.

Michael: Mm-hmm.

Matthew: We have customers in the market, so there are AI companies who want to do this in a way that de-risks them from getting sued. They also wanna do this in a way that creates support amongst the music community. And there are some AI companies out there who have scraped the entire internet and are trying to build huge businesses without compensating the people who make the music.

And without all that music, these AI tools can't do anything.

Michael: Hmm.

Matthew: There are lawsuits that I'm not a part of between major rights holders and some of the large AI companies that have chosen to do this without a license. Eventually that will be settled in some way, either through congressional intervention.

You know, as we know, mechanical royalties are set by Congress effectively, so there's government intervention that may play a role in it. There's lawsuits that get settled before they actually have a judge decide how it's supposed to work. And there are some lawsuits that go all the way to an adjudication that say this is, you know, having a court say, this is how this should work.

This is not unfamiliar to me. This is what we've been through since the first era of file sharing. And in fact, you know, way before my time and your time, someone had to figure out how is radio supposed to work? How do people get compensated when songs get played on the radio? And even before that, people had to figure out how does it work when a big band uses someone else's composition on sheet music and performs that song in a nightclub, you know, 150 years ago?

Michael: Mm-hmm.

Matthew: So these processes, you know, these arguments have happened before. I happen to have opinions about where I think all of this will land. It seems like this is all gonna land where there's some agreement among copyright holders and AI companies about how this is supposed to work, and our business is to enable the way they all agree it should work.

To make it transparent, consistent, and as fair as possible for everyone involved.

Michael: Mm-hmm.

Matthew: That's so—

Michael: Good. I mean, it's super interesting too, the point that you made around how this isn't like our first time needing to solve the tricky problem of attribution on, you know, how do you compensate, you know, people who own intellectual property for its use in different mediums.

I can also see how this might be a bit of a different territory from a standpoint of copyright in the definition of copyright, since these are like quote-unquote original works that are different than the originals. And I mean, the way that you just presented it, I mean, it feels like such a fair perspective that, like, of course we should. Like, it just feels almost immoral or wrong or unethical to scrape all of this data from existing songs and then just use it to profit, make billions of dollars without any sort of acknowledgement or compensation of the initial source.

So I can absolutely understand that perspective. And I mean, to steelman the opposite point of view as well, it'd be curious to, yeah, if I'm understanding this right from, like, steelmanning the opposite point of view, is it that if we go that route, we could risk overregulating the way that AI as a tool is used for the benefit of everyone, that we might lose out to China, because China doesn't have the same regulations as we have.

So then they could win the AI race, which of course we don't wanna have happen over here. And that it would just be more challenging to regulate. Is that like the strongest steelman case, or are there other points that someone on the opposition would make?

Matthew: I mean, that's, that's, that is, you know, the case a lot of folks make, but frankly, it's the exact same case illegal file-sharing networks made, you know, 20-some years ago. Of course, if I wanted to start a car dealership in which I just stole all the cars and then sold them, I could make the case that to tell me not to do that would be overregulation and that that might stop me from competing with other people who are stealing cars.

And wouldn't it be great for consumers if I stole cars and sold them, you know, gave them away for free to consumers?

Michael: Mm-hmm. Marsh? All the blueprints. The blueprints for like creating cars. Because if it was, sure, yeah. It's like, sure, like everyone could use it for free, but yeah, like it, there's, that's not necessarily moral to allow people to do that.

Matthew: I don't think it's even really about morality. We have copyright law, and it's actually not the case that China is out there doing this completely differently. We have Chinese rights holders in our system that are using our system. We're working with future Chinese AI companies that want to license data.

And again, what the people made the case: well, if we're not the kings of illegal file sharing, China will be. That didn't happen.

Michael: Interesting. Yeah, so I can see what you're saying. Yeah. Using the same case for other types of, you know, illegal use of copyrighted material.

The, the what clicks for me for sure is the, just like copyrights as a whole, like, would it be easier for us to create, to be more creative and create, you know, more products if we didn't have to worry about copyrights and we just were able to use anything that we wanted to? Sure we would. And if there's another country that's not acting in good faith and has no copyright and just like, steals everything and does stuff like, do they have an advantage because of that? You could steelman the case for that. Does that mean that we shouldn't have copyright? Probably not. It's probably a good thing that we can protect our copyright and our rights, unless we're going into like an even more, like, you know, far-out world where we're talking about copyright and ownership and possession in general.

But, um, okay. I mean, I—

Matthew: I believe in copyright. I think it allows people who are creative to apply their trade, to sometimes make a living, sometimes get rich, sometimes just get by. But it allows them to continue to be creative. And so I would posit that without copyright, creativity crumbles.

Michael: Mm.

Matthew: It is a creativity enabler.

And in fact, that's been the position of the United States ever since the Constitution was written.

Michael: Mm-hmm. Absolutely. Yeah. Just having the safety to be able to share things. I know for sure with having lots of conversations with the artists, how many are afraid of having their music stolen or their songs stolen, you know, even if it's like early on, it's all—

Matthew: Yeah, it's already been stolen.

Michael: Mm-hmm. Yeah. It speaks every day being—

Matthew: Wow. But you know, there was a time in the music business where we, the music business, were worried that we would not be able to put illegal file sharing back in the bottom, and that there would never be a commercial music business again. And in less than eight years, which is the blink of an eye in human history, that was resolved.

Of course, there is still illegal file sharing, but it's minuscule compared to the volume of consumption and joy that people get from subscribing to music or purchasing their favorite records. You know, I was just in a store that was filled with CDs the other day. People were buying CDs.

And so, you know, we got past that and we got to a place where not only could musicians make a living, but really for the first time in the world, any musician can distribute their works globally. Which I think is also an exciting aspect of the digital transformation. I mean, before digital, the average home music maker, which I am one of, couldn't get their records into Tower Records.

Right? And now they can get their music on Spotify and Apple and YouTube, and SoundCloud, and virtually anywhere they want in a manner that makes sure they're protected and they can get paid for the value they create for consumers.

Michael: Yeah, that's a great point. It definitely is the golden age in terms of being able to create music without necessarily requiring the infrastructure or a gatekeeper to distribute it.

Mm-hmm. Um, probably even more so the case with these generative AI tools, anyone can make music. That sounds amazing. Yeah. But also, yeah, I mean—

Matthew: I'm not sure, I'm not sure the user of generative music is making music. They're, you know, they're generating music, which is still a very joyful process. You know, I wanna be clear, I believe that inherent to being human is the desire to communicate and create.

Michael: Hmm.

Matthew: And if the way someone wants to do that and can do that is through generative AI, I believe that can be really, really valuable and create joy for the creator and the listener of that stuff. One of the new challenges we have in this world of generative AI is that, well, I'll tell a little story.

When I owned a house music label, the big dance music shop in Chicago had about a hundred new releases show up a week. When I got to Beatport, we were at about 3,000 new releases a week, and when I left it was 27,000 new releases a week.

Michael: Hmm. As—

Matthew: We all know, a lot of the DSPs are receiving between a hundred thousand and 150,000 new tracks every week. And that number's growing exponentially right now because of AI. I mean, it's really growing exponentially. And one of the challenges we have is human, you know, it requires time to listen to music. Mm-hmm. And human beings only have 24 hours in a day, and most people don't spend 24 hours a day listening to music.

Michael: Mm.

Matthew: I spend five or six hours. I bet the average person spends an hour or less. And so the more music there is in the marketplace, the less time someone is likely to spend listening to your music or my music. And in the way the DSPs work, because they share revenue based on the amount of time effectively people spend with each individual track.

The volume of AI-generated content in those systems can push aside human creativity and the compensation human creativity gets. I don't know the answer to that. And in fact, thank goodness I don't have to define the answer to that. I do believe, however, that the industry and technologists and consumers who play a huge role in this, and musicians and songwriters who play the biggest role in all this, are gonna come to find a way to make that work.

I don't know if you saw that Deezer said that they may be getting as much as, I think, a quarter to a third of all music sent to them every day as AI-generated. Spotify just announced that they removed 75 million spammy AI tracks.

Can you imagine if they identified 75 million as spammy? How many more are still in that system at this rate? By the end of next year, half of all the music on the DSPs, the streaming services, will be 100% AI-generated, and at this rate, four or five years from now, most of it will be AI-generated. Other than, I believe strongly there's gonna be some resolution about how to manage all this.

I don't believe music listeners want to go through a record store that has a billion songs in it to find the 10 songs they wanna listen to right now. Mm-hmm. Ultimately there's gonna be some figuring out of, most importantly, what's the best way to help listeners find what they wanna listen to and make sure that people get paid properly when it's listened to.

Michael: Hmm. That's so fascinating. So what I'm hearing you say is that, you think that these tools are amazing, you know, creative tools and they're definitely increasing the throughput by, like, how many songs being released every day. We have a limited amount of time and attention. We still have 24 hours a day.

There's, and so proportionally we just have less to be able to actually listen to all of the new music that's coming out. So both, like, there's a need for a curation, like, who can most effectively curate, you know, the actual songs that you're going to hear. And also, you know, in terms of music generation, like, more and more of these songs are being generated.

Using AI and answering the question is a bit tricky of at what point do we sort of say like, this is, yeah, this was just completely generated and there's no human, like, creativity here. Mm-hmm. Or how much, yeah. When do we kind of draw that line versus spammy, spammy AI versus actual, like, you know, music.

Yeah. What a crazy world that we're in right now. We, I mean, I know we're probably both gonna look like fools if, when we have this conversation, but I'd love to hear where do you think that things are headed? 10 years from now, 20 years from now, when it comes to this type of creativity? Like, I think we're should be in a world where, like, everything is generated on the fly and it's all personalized.

Or like, where does this, where's this going?

Matthew: That's a great question. You're right. We would both look like fools if we could try to say with some certainty where this is going. But I do agree with you that, you know, long term a lot of the media we consume will be generated on the fly. If you think about the evolution of music recommendations.

You know, which are highly automated, right? You know, now basically machine learning is what gives us all the recommendations we get in systems today, that rather than recommending the next piece of content, music, television, film, it's just gonna create the next piece for you. And it's gonna create that next piece based on the context of where you are, what you're doing, who you are.

Scarily enough, your credit report, your health insurance information, it's gonna digest all that information and push material to you.

Michael: Mm-hmm.

Matthew: The extent to which people want that, I don't know yet, because I do believe that we largely consume music, at least, if not other forms of media, for human connection.

Michael: Hmm.

Matthew: Or at a bare minimum to know that someone else feels the same way we do about something.

Michael: Hmm.

Matthew: And can AI do that effectively? Maybe. Will consumers of media feel the same way if a generated piece of content feels the same way they do? I don't know. Some people will, I'm sure. But, you know, there's gonna come a day really, really soon where you can watch Die Hard on Netflix, and they replace Bruce Willis with you.

You can watch yourself in the movie Die Hard. Mm-hmm. Is that something people really want? Hmm. Time will tell.

Michael: Hmm. Wow. Yeah. And like, change the details of it. So it's more relatable to you, like, it's like your kids.

Matthew: Sure, yeah. Change the politics of something. Change the level of sexual explicitness around something.

It'll be very interesting. You know, any parameter that somebody might judge a piece of media on could be changed on the fly. Is that what people are gonna enjoy? Well, I'm a big proponent of if it brings someone joy, it has value. Period. You know, I grew up working in record stores and I was punk rock and I had my taste in music, but I also learned growing up in record stores that most people have limited time.

They've gotta drive their kids to soccer practice. And if they wanna listen to that record I don't like, but it brings them joy, then that's a great record. Hmm. And I, you know, I feel that way about all media as long as it's not deceptive. We're not talking about deep fakes here. To me that's a completely other issue, and that's, you know, fraudulent media in my mind.

Michael: Hmm.

Matthew: Makes sense.

Michael: Yeah. That's such a fascinating idea, right? Like having personal, like, where it puts you as the character, the main character in a movie. And I guess if we're in a world where that's possible, then maybe the direction is more so like completely immersive or interactive. So it's like, don't just watch yourself in a movie, like be yourself, like acting out this character in this, you know, immersive experience or game and, mm-hmm.

And, yeah, that is, that is a very interesting world. I mean, I, at risk of like totally going down, derailing this conversation. It's interesting from a landscape of, like, what, like simulation theory. I don't know if you're, I've thought about that. Educate me conversations. So I'm not the world's expert in simulation theory, so this is also probably gonna make me look like a fool.

But from what I understand, the base premise is that based on how quickly these technologies are emerging, that there's a very high likelihood that we're not existing in quote-unquote base reality, that we're in some form of a generative simulation, but not a simulation from a standpoint of like, this is like fake, but from just like, the nature of reality as we experience it is in some form, like being dreamt up or simulated or created.

There's a creative force to that. And of course this line of thinking kind of leads to bigger questions of purpose and fulfillment in God and, certainly as it relates to technology and intelligence and simulation and generative stuff, the things that we're doing now, like, compared to a hundred years ago, it's not a far leap to think that we could have that experience in the next 10 to 20 years where we are like in a lifelike experience where we can, you know, design it, in a way that feels very close to like base reality and is in like high fidelity, which is kinda crazy to think about.

Matthew: Yeah, my instinct is technologically that is much farther away than the people invested in that technology financially would have us believe. Additionally, you know, one of the reasons music in particular is such a potent form of media is because most people can consume music while they're doing something else.

Michael: Hmm.

Matthew: We're all gonna have a very limited number of time we can sit in a bowl of gelatinous goo, and with goggles on and do nothing. But, you know, this virtual experience, I am a gamer. Or I won't say it that way. I play video games. Hmm. And I enjoy them. But that's, that takes me away from everything else in my life, right?

I can listen to music while I'm driving, while I'm in the car, while I'm walking the dog. It's gonna be pretty difficult for me to be in a gelatinous bowl of goo having a fully synthetic experience and walk the dog at the same time. And frankly, my dog wouldn't like that very much. Your virtual dog would love it though.

I'm just kidding. Yes. And meanwhile, my own dog would be dead on the floor. Right?

Michael: It's like that meme with like, the, hey, have you seen that one with the, it's like a mother with her, one of her kids in the pool. And the one of the kids is like drowning next to him. And then like it shows, the next scene is like the kid has a skeleton at the bottom of the pool.

I haven't really—

Matthew: I can picture a future like that. You know, reality. And if we can go, you know, you brought up this is, has spiritual implications and psychedelic implications. The fact of the matter is there are facts in this world. What we generally think of as reality is deeply personal.

None of us have the same brains. None of us have experienced the same lives. I think it's a mistake to believe that any human being experiences reality exactly the same way as someone else. I'm not really sure every human being sees blue the same way, much less anything else. I mean, how can we know we can't be inside someone else's?

So, you know, I think simulation theory, what little I know about it is fun to think about. There are a lot of fun things to think about. I think it's more philosophical than it really is technically feasible today. Mm-hmm. And I also think that we, as human beings, we are networked creatures.

I think we're a lot more like bees than we realize. And I mean that complimentarily, you know, that we're a hive. I think we have connectivity electromagnetically that we don't understand. And any technology that would, that separates us, as opposed to allows us to be connected, ultimately, I think is a disservice to human beings.

And we see young people right now who are getting dumb smartphones 'cause they want to have more community and intimacy with those around them. I think there's gonna be a real, I wouldn't say backlash, but an underground movement towards human connectivity, at least throughout the rest of my lifetime.

Michael: Mm. Yeah, it certainly feels like that, right? Like, already it feels like there's growing tension right now with, you know, at odds with social media and with technology. There's a bit of, like, a movement of back to our roots, back to nature. What I'm hearing you say is that, you know, technology, when it's in service of us connecting and establishing more authentic relationships, that's technology that brings more joy into the world, whereas I think so.

And technology that does the opposite, you know, over the long term, like, it's going to probably fizzle out because it's not generating real value. Yeah, it is really interesting. So, for context, one, I had a conversation on our podcast a few months ago with a guy named Nolan Arbaugh.

Noland is paraplegic and so he has, you know, paralyzed from the neck down. And he was the first human patient to have a Neuralink installed. And so this Neuralink is a neural interface that allows him to control things telepathically with his thoughts. And we, on the podcast, created a song together, "Telepath," the first song telepathically with his Neuralink.

And certainly in terms of this conversation and the idea of like, the hive mind, that was what comes to mind for me is like, man. If we had the ability to communicate more directly and authentically. Words and language are amazing, but like, pretty easy to misinterpret or misconstrue.

Pretty slow. Pretty slow. Mm-hmm. Oh man. Are we gonna live in a world where we can just think thoughts together and communicate more directly and actually communicate the experience of blue? Like, what would that look like? And it certainly feels philosophical definitely right now because it, it just is not there yet.

But the fact that we can have this conversation even, like, ponder that maybe in the next, you know, 20 years this is a possibility is pretty wild.

Matthew: Yeah, I mean, what a great story. I, you know, I'm unfamiliar with that gentleman and his story, but what a gift that technology, you know, is, presumably for him if he believes that, he does.

Michael: Oh, he's, he's a—

Matthew: Great spokesperson for it as well. And, but I would posit that we all do have the ability to have those kinds of authentic connections and unspoken, undocumented communications. Mm. You know, the best example is love.

Michael: Mm. And music came to mind too. I was like, oh, that's kind of what it's for, isn't it?

Matthew: Yeah.

Michael: Yeah.

Matthew: I think so. I, you know what, that's how it functions for me. I'm not gonna judge what other people use music for.

Michael: Mm. Mm. So good. Matthew, really appreciate you taking the time to be on the podcast today and have a conversation like this. This is, this is a fun one. Oh, Michael—

Matthew: This was a joy.

I have shivers right now. I'm so glad we spoke philosophically and not just about work. This was a delight. I hope your listeners enjoyed it.

Michael: Thank you. I know I absolutely did. And I'm very grateful that we have folks like yourself that have the, both the perspective and experience from your past, you know, experience with these tools and evolution if they happen, but also can help us to figure out, you know, how do we—

You know, use these tools in a way that benefits, you know, the people who deserve to be credited. So thank you for the work that you're doing. And for anyone that is listening or watching this right now, who wants to connect more or learn more about the platform, what's the best place for 'em to go, to take next steps?

Matthew: So our business is directed towards people who own copyrighted assets or a backend system. But you can find us @wearemusical.ai on the web. I'm also pretty easy to find on LinkedIn. There's only two Matthews. One of them is a real estate mogul in New York, and one of them looks like me.

Mm-hmm. And, you know, I love hearing from folks. I love, especially when people can share their joy with me.

Michael: Hmm. Awesome. Well, Matthew, thank you again for being on the podcast today and looking forward to helping create the future together and creating more joy, you know, through music and spreading it with more people.

Matthew: Thank you so much, Michael. You have a great day.

Michael: Yeah.