18: What a year...!
Is it still funny in 2,024 to let an artificial intelligence like ChatGPT write a text for you? An introduction like this?
AI Voice:Hello, followers. It is December 2024. The year will soon be over, and we look back. It was the most exciting year for music and AI so far. SUNO and Uryo arrived.
AI Voice:Labels have sued tech companies for infringing the copyrights of their artists. New start ups emerged and disappeared. More songs than ever were published. It is an exciting time to be part of the big artificial intelligence sector, especially for this podcast. There is so much to talk about. Thank you very much, ChatGPT for giving me a brief introduction for this end of the year episode. I asked the model to, quote, write in a few and easy sentences an introduction speech for a podcast that talks about music and AI. Don't get into detail. Just mention a few developments of 24. And what you just heard is the result.
Dennis Kastrup:Really? Did I really do that, or did I just write the text and let an AI voice pronounce it? Who knows in these days what is real and what is not? Welcome to a new episode.
AI Jingle:This is the ILLIAC suite, a podcast on AI driven music. Join me as we dive into the ever evolving world of AI and music where algorithms become the composers and machines become the virtuosos. Yes. This music and text was written by a computer, and I am not real. But
Dennis Kastrup:I am, and my name is Dennis Kasthorp. Hello, humans. For this last episode of the year 2024, I have decided to look back a bit and this with a friend of mine with whom I always enjoy having conversations about AI. Matthias Struble, Matthias and I met some years ago. Although, you will hear in the next 50 minutes at some point that we do not agree on the date when that happened.
Dennis Kastrup:Matthias will introduce himself right now and you can listen to our conversation. I hope you will enjoy it.
Matthias Strobel:I'm running MusicTech Germany, which is the federal association for music technology. We basically take care of, all the people and companies and the research institutions who work on new solutions for the music ecosystem that are technology driven. So everyone who is a software developer, hardware manufacturer, basically the people who build the tools for musicians to be creative or to help the music industry to generate an economic income through techno lot technology driven tools. Okay. And what And AI is a big one, obviously.
Dennis Kastrup:Of course. And AI when did AI, in the field of music kind of enter in your field and maybe what did you think? I mean, it slowly did it slowly sneak in or was it like a sudden, oops, here I am. How did it enter your field and what did you think and how did you approach it?
Matthias Strobel:I think personally, my first contact with AI was the first recommendations in the early days of the streaming services. So that was my personal first contact with, AI in that sense. And, on the professional level with Music Tech Germany, I think in 2017. So I'm also running an event series here in Berlin called the Berlin Music Tech Meetups, which is a irregularly happening, meeting of all the people who are interested in music technology here in Berlin. And I started in 2017.
Matthias Strobel:And I think the first event, the first meetup that I ever did was about AI and music, which was quite a different talk back in the days. But that was the first time that I actually had experts and professional people, coming to the event and talking from their perspective about what they expect from it.
Dennis Kastrup:Well, I imagine back then AI in music was more into, like, yeah, algorithms and predictions, what to listen to, where to go to. It was not really going to the direction of generative AI. I mean, 2017. It was maybe on the horizon, but it was not really focused on that, I imagine.
Matthias Strobel:No. Like, the whole generative AI field back in that days were probably only explored by either researchers or artists who really used it in a in a quite nerdy way. I can't really remember what the first, AI people were, but I think, yeah. It was probably not about generated fire. I just think about, like, I think I don't know if, like, I invited CJ from Databots, once or twice, and I don't know if he was part of the first meetup or one that happened a couple of years later.
Matthias Strobel:But I don't think that generative AI was such a big topic, and nobody could have seen probably at least in my circle, where we are today.
Dennis Kastrup:Yes, CJ. CJ is always CJ is always a Dave. I think I already mentioned him here on the podcast also a couple of times. Our friend CJ, we both know him really well, and you know him longer than me. Yeah.
Dennis Kastrup:For me, I stay with my opinion. For me, CJ is, the the mastermind of generative AI and craziness. All the craziness that surrounds him in the sense that he's a real artist. He goes on stage. He's, like, moving like crazy, and feeling the music.
Dennis Kastrup:That's that's what I always like when I see artists, when they feel the music. It's a little bit hardcore, heavy stuff, and he's generating some I would say people who are not really, well, into that field would say, unrhythmic music. But on the other sense, it is really well thought and generated really in a sense that I think it's amazingly avant garde. And, so yeah. If you don't know CJ and the data bot, check him and them out because they're they're a team and they do really amazing stuff.
Dennis Kastrup:Since then Yeah. A lot of stuff happened. And, we we we know each other, also since that time, I would say, since 2016, 17 maybe?
Matthias Strobel:No. We know we know each other way earlier. We know each other Really?
Dennis Kastrup:A lot more? I would say 16.
Matthias Strobel:Haven't you not done an interview with me with Nagel Sounds back in the days?
Dennis Kastrup:Nagel Sounds as what? Explain to the listeners what it was.
Matthias Strobel:Like the first startup that I had, like, where, you could basically we developed a technology where you could translate data streams into musical structures in real time. So we used the Kinect camera and translated that data stream into music, and you could basically turn your body into an an orchestra as we as we said. And I believe that I did an interview with you when you had your days in at Radia Einci in Berlin.
Dennis Kastrup:Yes.
Matthias Strobel:That was already 2012.
Dennis Kastrup:No. No. No. No. No.
Dennis Kastrup:But Oh, yeah. Yeah.
Matthias Strobel:We closed the company in 2015, so it can't be that, it was later than that.
Dennis Kastrup:Okay. We are we're having a fight here, but I I cannot be. I didn't have my show in 2012. I didn't have my show in 2013. Maybe then it was 2,015.
Dennis Kastrup:But I don't but but but I think I started my show in 2,017. And so here here is some time mixing up, but, we'll we'll, we'll figure that out after the interview. But it was Let's do that. I I it was definitely not 2,012. Well, we know each other since a long time, and we talk about AI a lot.
Dennis Kastrup:We discuss about AI a lot. We have, same opinions. We have different opinions. And that's what I like about it because, because, we don't have fights. But, sometimes Matthias is totally wrong and sometimes I am totally wrong.
Dennis Kastrup:And, that's what I like about it. But tell me a little bit, I I invited you to the to the podcast, at the end of the year because 2024 was an amazing or scary or, astonishing year in AI, and music in general. How do you look back in the last, 12 months
Matthias Strobel:to the last 5 months? I would say, like, this year was probably the fastest year in, in serious technological development when it comes to, like I mean, AI is basically happening all over the place, like, in every industry, in all of our lives. It's it's part of the discussion where we go and whatever we do. But, just like, focusing on music, I think, there has not been another technology that has gone through so many iterations in 1 year than AI did in this year. And, when we just look back at, like, how how how how bad it sounded at the beginning or, like, compared to what we have today, when it comes to generative AI and where we are today and, how how little the discussion was about, like, generative AI and how can we license music and what do we do with it and how can we make sure nobody gets a use or how can we make sure that everybody uses it in an ethical way.
Matthias Strobel:And I think, the obviously, the discussion heated up. The the the to expect that the the law cases that are that, yeah, we could expect them are all in place now. And, I think the next year is gonna be even faster when it comes to to AI and technological developments in in all different, aspects of AI, not just generative AI. I think we will see a lot more, applications that are driven by AI in some sort of way, and we will also see a lot more, kinda like, yeah, access to it, like easier access. And that's something that, that kinda like it's not surprising to me, but at the beginning of the year, there was probably a handful or 2 handful of of start ups that were calling themselves a generative AI start up for music or company.
Matthias Strobel:And I can't really count the the startups that are now promising to be, like, the one generative AI startup that came that all popped up throughout the year. And, obviously, with all of the technologies, there's a lot of, scammers out there that promise stuff that they can't keep up with. And, but there's also really, really, good and, amazing examples of how this technology can be used in a really creative way and how it can be used as a as a tool for for musicians in the future.
Dennis Kastrup:What surprised me, because I work, on reports, like, in the last days of the of a year always who look back at the last 12 months. And what really surprised me, I really had to look up again if Sunu and Uryo came out this year. I I remember I may remember then, but it was so much was happening that I was not sure again if it was this year or even the year before because it speeded up so fast after January when June came out last December. Sunu came out last December, and then audio followed, I think, in February. So this year, as you already mentioned, it's it's the speed is is is amazing.
Dennis Kastrup:I I cannot even I think I cannot even no. I think I know I cannot sum up what happened in the last forever. What came up? What new companies came up? Which companies disappeared again?
Dennis Kastrup:What discussions we had on so many levels? I remember I did write a a big article in March, for an organization in Germany, and I had to sum up the situation of generative AI and or AI and music in general. And I finished the article in in April, and it was published in May. And I thought it was so old. I mean, it was published the beginning of May and I thought, what did I write?
Dennis Kastrup:Yeah. Yeah.
Matthias Strobel:So much
Dennis Kastrup:so many things already changed in in 2 months that it was so hard to keep up. It was also so hard for me to keep up the speed, of reporting about it because once something opened, another door opened. And but on the other side, this also showed me one thing. What I when I look back at 2,024, for me, it's clear. This is the year where I stopped saying, wow.
Dennis Kastrup:Because yeah. Yeah. No. Because this everything is possible now. Yes.
Dennis Kastrup:It it started to speed up in the way, wow, it is possible to change your voice. Wow, it is possible to generate this. Wow, it is possible to generate a whole song whole song. It's over now. For me, everything is possible.
Dennis Kastrup:What we can imagine of in AI all over the world, it is possible also like also in image generating or that's another subject. There's no wow anymore. So right now for me 24 is the point. Okay. It's there and now we have to find solutions to live with it.
Dennis Kastrup:Would you agree there or would you say there's still some stuff on the horizon we don't know?
Matthias Strobel:I mean there's always stuff on the horizon that we don't know, I guess. So we are not like the the super brains out there that have like, some sort of, like, knowledge that nobody else has. So I'm pretty sure that there will be lots of stuff that we wouldn't expect to have happened. Like, I just learned about, like, the NVIDIA, generative AI model for music that they just released. I think it's called Fugaru or something that I wouldn't have expected this company to come up with a with a product like that where you can do, like, really silly stuff that, in the first like, when you think about it, it doesn't make any sense at the beginning.
Matthias Strobel:So you can use horns and turn them into barking of of dogs and all the kind of crazy stuff. So I think, the the potential of using AI in ways outside of the box, are still unexplored. And I think we're still at the really, really early days, when it comes to what's possible with AI and what how it can be used, for music. So I think the you're right. There's no wow effect anymore because, like, okay.
Matthias Strobel:Yeah. Wow. The this is something that, I wouldn't have expected to happen that fast, but it's clear that it's gonna happen. I'm just still impressed by the speed and by, how small these LLM models are are are becoming to be and, how fast the whole open source community develop stuff. And so, I think that there's a lot of things that, we haven't even well, a lot of people haven't thought about, and there will be a lot of stuff happening in the next couple of years.
Matthias Strobel:And I think the relevance of all these companies that build these more general generative AI models like UNO and Studios gonna become less. There will be more niche specific companies coming up. I guess there will be probably more people using or building their own, LLMs. And so there's a lot like this. This is the the the exciting thing with AI is that it impacts so many different layers of music and the music industry and the creative process and the legal stuff and the access to music and so on and so on and so on that there's, this is that's not gonna be over in the next, 2 or 3 years.
Matthias Strobel:It's gonna go on for for for a long time. So I think, there's a lot of stuff that we haven't thought about yet, and which will still surprise us. But, yeah, the wow effect when you see, like, the first generated picture or the first generated sound with AI, that's that's gone.
Dennis Kastrup:It's interesting because, I could hear by what you just said how mixed up you are also because you said, uno in studio. But it was kind of a nice mixture of it. I wanted to I I know I know your idea what you wanted to say. It it happens to me also sometimes. Uno SORA, Sudio, Sudio, WO audio.
Dennis Kastrup:It's it's it's all a big mixture. But, yeah, you're talking about this also this legal stuff. One thing that stays in my mind, really present for the for the last year is, yeah. Of course, the discussion about the data sets, about the copyrights, about how do we deal with in the future. Maybe we'll talk about this a little bit later.
Dennis Kastrup:But one thing I also saw happening on the other side of the spectrum that people are, trying to find ways not to fight. Yeah. Fight also, but also find ways to raise a voice to say, hey. This is not going in a good direction. I don't like it.
Dennis Kastrup:And also, generative AI models who try to integrate the artists in saying, okay. We are using artists who gave their consent to train our data, to train our models. And so there was some stuff happening on that side, which I thought that was good that we also have a, a community which says, okay, this far is good, but we still also want to go in a different direction. I I remember, at Newton Rex, of course, who who did start this organization called Fairly Trained. What do you think about this?
Dennis Kastrup:I know you you had some opinions about this too, which I saw online. He also, yeah, released this petition where you can sign and support at Knewton Rex for his idea that artists should be paid who are in datasets. What is your opinion about this?
Matthias Strobel:That's a difficult one because I think obviously, everybody should be paid for the work that they're doing. Right? So, that that there's no question about that. It's like it's a if you're not out of your mind, then clear that, if you do work, you should be paid for that. So if you've done creative work, you should be paid for that if somebody else uses it to train their model.
Matthias Strobel:So there's that's absolutely clear. So I'm in favor of, like, having the people who build the found or who's who supply the foundation for for generative AI models, in form of datasets that these people should be, paid royalties in some sort of way. The problem here is that, there is no general system in place right now. So, if I wanna start if I wanna train a model right now, I don't know where to go and where I can actually, license music in order to train my model. So I have 2 options.
Matthias Strobel:Either I train it with, with, like, accessible music that's, that's out in the Internet, and it's, it's royalty free music, and it's allowed to train a model with that. And it's basically a public domain music. So I can train a model with that kind of music. It's some often shitty quality. It's it's not the music that I wanna train my model with.
Matthias Strobel:Or I can go to Pirate Bay or any other place and illegally download that music and train my model with that. It it's also probably not the best data yeah. And it's probably also oh, you scrape it exactly. But it's not the the the quality of the data is probably not the best that you can get. So it's, so that that sucks.
Matthias Strobel:So, the the thing that I see here is that, it's easy to complain. Like, like it's it's it's cool to have, like, the Fairly Trained thing going and have, like, other campaigns going. And it's the Human Artistry campaign that started already. I don't know. It's probably a year ago that they they put out this website and letter.
Matthias Strobel:Then there's probably another 10 campaigns where, like, the biggest artists in the world and the smallest artists in the world and whoever signed in favor for, like, okay, we should pay the artist. You should not use music illegally to train a model because that's that's unethical and so on and so on. And I think everybody would agree. But it's so easy to complain stuff and to say, like, you shouldn't do it if you don't, provide a solution. So the thing here is that, the music industry, as they always, are lacking on incentivizing people using their creative assets by providing an infrastructure in order to license the music and paying out the actual rights holders in a fair and transparent way.
Matthias Strobel:And this has never been the case with all the other inventions from from MP 3 streaming. I don't wanna talk about all that stuff. But it's it's just we're in the same situation right now. And I think that's kinda like, gives it a bit of, like, a a faulty taste if, like, if there's, like, everybody's complaining, but nobody's providing a solution. And, I think that's a big problem, and I think that needs to be solved.
Matthias Strobel:And I fear that, a lot of companies think, okay. If if audio audio and Suno can train their models on illegal stuff and then just have to go to have to wait until they they are sued, let's just do the same. So everybody thinks, like, it's okay to illegally train their models, and then they'll wait for somebody suing them. And, that sucks. And that also means that a lot of these generative AI companies who have trained their models with, illegal music are gonna shut down in the next 12 to 24 months when Udo and Sudio Udo and Suno are coming to an agreement with, like, the major labels which they will eventually come to, and then, the major labels will say that they're so happy to, like, work with such a, such a expert, tech company that has, like, the best knowledge in AI.
Matthias Strobel:And, like, we know all that. That's a game that the the major label is always playing. They're never acting. They're always reacting. And that's gonna be the case here as well.
Matthias Strobel:And, so but, that still means that, there will be, backdoor deals that nobody knows about. Nobody will see the contracts. The actual rights holders will not get the money they are supposed to get. The labels will basically ripe all the fruits from these Chen AI models and then leave their their artists, which are so important to them, with, like, a couple of pennies like they did with streaming. And, that's that's a threat for artists.
Matthias Strobel:That's a threat for rights holders, and it's an opportunity for major labels to open up another revenue stream and, fucking up the artists, in in new and different ways that they couldn't do before generative AI.
Dennis Kastrup:Yeah. They are they are already positioning themselves again as the fighter for the for the artist, But in the end, they're fighting for their money and not really for the artist money.
Matthias Strobel:Never. Never. Like, all the companies who are on the stock market and they're, like, all the public companies, they they fight for their shareholders and never for the artist. And the artist is replaceable. Like, the money they put into Taylor Swift must be probably 100 of 1,000,000 because they knew she's or because she's successful.
Matthias Strobel:If she fucks something up, they just, like, let her go and then take another artist. Like, it's you know, like, artists are replaceable for for major music companies, and, they don't really care about the human being behind that. They mind that. But that's not that's not a, like, a secret anymore. That's, like, it's clear.
Dennis Kastrup:I have I have talked about this subject also in many reports I did for radio stations and it's a never ending story. I can I think I can each month, present a new report about what is the discussion, what is going on, where do we go, who's suing, who who's planning to have a a dataset which is ethically fine, who's who's saying, okay, it is fair use? Who's saying, I don't give a fuck at all. I just do myself. I do for for myself, for my company what I want.
Dennis Kastrup:What, how do you think, we can, we as as as let's call us customer. We as, customers who love the musicians, who love the music, who love the artists, and, then on the other side, use Odoo I said also Odoo. Odoo and Sunu all the time to play around, to have fun because, they say, okay, I, we we are a big company. We take, we take I think it's in in Great Britain. They said, oh, it's it's, it's it's fair use.
Dennis Kastrup:We are we should be allowed to do this. I think what we have to do right now in this moment, and which I always try to also tell people is, do the the thing, not just re just react and sit there and be amazed of, oh, what's going on? And let them, have the power to decide on on top of our heads. Okay. This is the way it goes because that's the same thing that happened with Spotify in the end.
Dennis Kastrup:We loved it. And it's a good way to pay artists, you have to say. But in the end, the model is not really working for the artist. The idea was good. So it is time for us to take actions now for us as customers, I say.
Dennis Kastrup:And what do you think we can do? I mean, I'm always saying, like, it is an an option would be to choose, which, generative AI model you wanna take. But on the other hand, then you miss out the models you really like. You are doing really good stuff, and you sometimes really don't know what is behind the model. Also, they say it's ethically trained.
Dennis Kastrup:So what do you think we as a customer, as as users, what what can we do? Is is there something we can do?
Matthias Strobel:It honors you that you're trying to be on the good side and you wanna fight for the artist and you wanna, only use AI in an ethical way. Do you
Dennis Kastrup:use Spotify?
Matthias Strobel:No. No. No. No streaming service.
Dennis Kastrup:I use Deezer.
Matthias Strobel:Oh, you use Deezer. Okay. At least that that that's a that's a that's a try. But, if you wanna be, like, on on the artist side, you should only listen to music on Bandcamp, buy music, and then not go into the streaming thing at all. This is the same
Dennis Kastrup:like Wait, one thing. I use a generative AI model, which, has I use Soundful because,
Matthias Strobel:That's that's that's good on you because you are a music journalist and you, you know, you know you should. But, there's, like the and you're not a threat to the artist. Like the general public, let's talk about, like, the 16 year old until the the I don't know, all the way up, who, who receive their news and their access to music by their peers or by common media, whatever it is. They don't care about that. They go the easiest and fastest and most accessible way, and that's never gonna change.
Matthias Strobel:And, so I think that, it's it's cool that you're trying to be like that, but, it doesn't matter.
Dennis Kastrup:Like, people out there. But, you know, but I I I totally agree. I'm I'm not naive. I I agree. That's the way it goes.
Dennis Kastrup:I agree. That's capitalism. I it works like this. It's if it's if it's easy, if it's cheap and it's fun, we use it. But on the other hand, I don't wanna say I don't do nothing.
Dennis Kastrup:I mean, that's the thing. If you if you continue to say we don't do nothing, where do we go?
Matthias Strobel:Yeah. But you're a music journalist and you are I mean, this is you are in a really privileged situation. So what you can do is you can try to, like, make this a topic in on all the the stations you work for and all the situations you give interviews. Like, that's that's the only that that's not not that I mean, that's the only, but that's a major power that you have. And that's probably more than other people have who might have the same mindset like you, but not, are in the same situation or in the same, yeah, in the same with the same network and with the same opportunities that you have.
Matthias Strobel:So when it comes to that, like, you can obviously privately, use only ethical trained, AI models, which is fine. And you can always raise your voice and, ask for better, environments, for AI companies to license music. And you can try to put a spotlight on these AI companies that have ethically trained their their their models with, like, licensed or royalty free music. The thing is that, so I did a bit of a kind of, like, probably I wouldn't call it consultancy, but I I worked with a company that is trying to, like, provide a licensing infrastructure for AI companies by, having a system in place where they, collect catalogs from labels that and from artists who own all the rights, like, all the three rights and then, then, provide parts of that catalog that may make the most sense to these AI companies. It's not a problem to get labels and artists on board to actually tell them, look.
Matthias Strobel:If you own all the rights, give us your catalog, and we provide it to, AI companies. That opens up a new revenue stream for you, and we make sure to collect the money and then forward it to you. The problem is that AI companies, are not interested in it right now for the reasons that I just mentioned before that, they don't see any reason to, like, legally license music in at the moment. This will change. I'm pretty sure this will change in the next 12 to to 15 months.
Matthias Strobel:And, then then I mean, then it's it's all about, like, making sure that there's a good usability in place, there's an interface in place that people wanna use, that, the services that, provide all these generative AI music systems have have a catalog that's actually big enough for people to use it. If that's that's the area that you wanna go for, I'm rather believer that we will see more and more niches. People are gonna provide, small AI generative AI models for certain genres or even for certain sub genres or for certain elements in the creation process, be it voice or be drums or be it synths or be it whatever it is. So there will be more and more niche niche focused AI companies rather than having, like, this this huge model from Suno and Odeo which, there's, I mean, there's only I think that the the number of people who are using Sunon Audio, I have no numbers here, but I think that from the beginning to now, it didn't, grow. It rather, went down, if I if I'm correct.
Matthias Strobel:So I think that, yeah, there's a lot of different, ways of, like, supporting this ethical AI movement, and, you personally can raise your voice. That's what you do.
Dennis Kastrup:But I yep. But I do raise my voice because I want others to raise their voices too. Because if I don't raise my voice, nobody would raise their voice because others think others don't raise their voices. So that's why I wanted to talk about Yeah. That's not very sweet.
Dennis Kastrup:But on the but on the other hand, talking about licensing, how how do we do this? I mean, like we license let's just imagine something is possible. Like you have a data set of licensed music and the artists know this and they give it away. So how do we pay? I mean, do we pay a one time, fee?
Dennis Kastrup:Do we pay a monthly fee? Do we pay a yearly fee? And then, yeah. How how how is this how how will this work?
Matthias Strobel:Now that there's there's different ways of dealing with that. There's, ways of, like, dealing with it in, like there's tons of different models to to license music for training an AI model. But, but, like, let's just put 2 out there. Like, one is that, the one I think that's also, Stable Audio is using is, like, if you give them your your music, or if you license music for them, then, they give you a share of all the pro licenses that they sell through their service. So let's say they they sell, like, a 100 and pro licenses for €10, then they make a $1,000, and then they share it with you as a as a licensing company, all the the the the share.
Matthias Strobel:You get the the share. And then you as a licensing company, you can decide how much you you forward to to the people that you license the music from. So you make a deal with them upfront and tell them, like, okay. If we get in, like, I don't know, €500, we give, like, then it's the next thing. You give it further to the to the label.
Matthias Strobel:The label then decides how much of the the catalog that the label provided, will go to the different artists from all of the catalogs. There's a there's a there's a big chain of reactions that happens, when music is licensed.
Dennis Kastrup:And that's a it's a month and that's a monthly fee?
Matthias Strobel:I think it's a monthly or yearly fee. I think you can decide on stable audio what you what you pay for stable, audio. It's like, monthly or yearly fee. And I think that if I'm if I'm correct, it's I I know that they they did that at the beginning. I don't know if they still do that, that they forward, a percentage of the the the subscriptions that people do for their pro version to the people who provided, the music, but they also, may might have may have different, models in place right now.
Matthias Strobel:Another way is, like, weighing the music. So if you have, like, a model and you you basically, you're getting music from a label and, use like, where we what you can do as a model is, like, you can weigh music in a different different ways. Let's say, like, there's, I don't know, a certain amount of, let's go big, like, a certain amount of Taylor Swift songs in there. And then there's a certain amount of, like, indie folk musicians from from Nashville. Obviously, the music that, is from the Taylor Swift catalog will be probably more prominent in the outcome from the the AI system.
Matthias Strobel:So it's more important to have that kind of music in there. So you will you want to pay more to the person who provided the more important, bits and pieces of music in that model than to the person who provided bits and pieces that are not as relevant as, as the other ones. So you kinda like, kinda like categorize different parts of the mod of the catalog that you provide for the model. And then when you when people generate music for, like, I think in the in soon on audio, you have, like, points, and you use different points to regenerate models. And then you basically can see how much points were used to to this part and that part in the song.
Matthias Strobel:So at the end of the day, it's pretty for you as an AI model, for someone who owns an AI model, it's pretty transparent to see how much, of of of impact had different parts of the catalog that were used to train a certain model for a certain song in the outcome that comes out. But there's tons of different different ways of doing that, I think. And then there's artifacts, and, people are trying to, to to figure out how they can detect these different artifacts in the outcome. Then there's tokenization. Like, one of the most stupidest ideas is, like, to tokenize every kind of music that's out there and then trying to find those tokens in the outcome of the songs.
Matthias Strobel:There's there's no one one solution right now, and I don't know if there will ever be a one solution in in that sense. But, the first step would be, I mean, like like labels operate, like, in an advance that you say, okay, here's a catalog of 50,000 songs. And the AI company pays an advance upfront to the licensing company and says, okay, 50,000 songs. We give you a €100,000 and you pay that forward to the catalog owners. And, we use that kind of music for, I don't know, a year.
Matthias Strobel:And after a year, we see how much revenue we made. And then we make a new deal about, like, another advance or another kind of, like, deal that we make with the license holders. So there's, like, there's probably a lot of different options that need to be, discovered or need to be explored in a way to make sure that there's a fair system in place at the end. But, yeah, I'm not a I'm not a a licensing, professional here. Mhmm.
Dennis Kastrup:But, do you think with, do you think in the end this would be fair? I mean, that is it fair and will artists I mean, to to be honest, if you think about this again and we think about the fact that, like, that the dilemma we have with, with Spotify where the artists are not being paid fair. And if you are, let's say, a small artist who has been, part of a dataset and you probably get a just a little share of, what you have done all of your life, let's say 20 songs into I don't know. Maybe you wrote 20 songs in 2 years and you spent a lot of spent a lot of time, to to to to be creative and do this. Excuse me.
Dennis Kastrup:And then, you will not get paid at all a lot of money. And for example, those 20 songs you just put in there, people use to create a song that they would have asked you before to create because they think, they, you they like your music, but then they don't pay you anymore the right way. Do you think that will solve, a lot of questions about spending money in that field? Because I have a doubt that it will be. I really see another big wave of, I know all these critics that are coming out there from musicians.
Dennis Kastrup:I see another big wave of the money will not end where it's supposed to end. And tech companies will just gain a lot of money from from all those even if they are paying licenses and they make a lot of money. And the money will be distributed in a way that smaller artists I know they always suffered. People always give this argument. It was always like this.
Dennis Kastrup:But if they are in a data set, they're part of something, of generating something. They are part of this because an AI learns from everything that's in there. So do you think, I have fear that this will go in a direction where, again, like the smaller artists, what we see already around us, which is people are struggling more and more, will have a chance to live by this? Or is this a chance to make some money? Or will it in the end again be just the big the big the big, artist?
Matthias Strobel:I I don't I don't think that an art like, I don't know. Maybe there's artists out there that's gonna, like, like, that start producing music only to train, AI models with that. That could be a new kind of, like, kind of like a new job. Like, you as a music producer, you you you you produce music with the intention that this music is only be used to train an AI model because this is what an AI model needs right now in order to, have the output that, that's that's required or demanded. So that could be a new chop, at the end of the day.
Matthias Strobel:If you are just a, like, I don't know, a street musician, a folk musician that has, like, a a couple of songs and, then, like, what's the point? Like, if you expect to, like, make a living from, generative AI model licenses, then something is wrong with you. You should expect to make a living by playing your music, selling your music. That's why you make music. Like, this is can only be an additional revenue stream.
Matthias Strobel:And, yeah, you're right. I think that, if we don't get systems in place, it's gonna be the same same story again that, the big guys are kind of, like, ripening all the fruits, and the the people, on the at the end of the food chain, they will only get what's left over. And, so that's the the same story again. That's what that's why I think that, there's we have now a big opportunity to kind of, like, change the system and make a shift in in the industry also, because these AI systems get so so accessible. I don't know, like, I don't know if you're aware of, like, there's there's been Loop Fiasco made an AI, rap lyric generator, I think, already 2 years ago.
Matthias Strobel:And, so he's no tech guy at all, but, he had, I don't know, people in his surroundings. I don't know. The website is still on. So you can basically type in stuff, and, the the the website would generate rap verses, in his style for you. And, he just did that because he saw an opportunity, and he used his own catalog, I guess, to train that model in order to come up with new verses.
Matthias Strobel:And I think there's a lot of, like, opportunities for artists out there to actually do the same, especially with open source models to, like, have access to them, make them, own them, create your own models, train with your own catalog, and then sell these models. I think, anyway, that the future is for artists selling their own models to fans and that fans can then create their own versions, remixes, and stuff of that music. Only
Dennis Kastrup:models it will be called.
Matthias Strobel:Yeah. That's a good call. We should save that name. That's like that name. Yeah.
Matthias Strobel:Only models. Because, yeah, I think that's that's that's also a great way of engaging with the fans and, having, like, making sure that, you get the the money that you deserve to get. And obviously, small, really small artists don't will not have an opportunity to to train their own models now. But in the future, it will be as easy as building a WordPress website these days. I'm pretty sure that, everybody can train their own models.
Dennis Kastrup:At least then you have enough songs.
Matthias Strobel:Yeah. Exactly. Yeah. And still that. Like, there's, like, there's models that are so small these days that you can train them already with, like, I don't know, 500 songs.
Matthias Strobel:Obviously, that's already a lot. But, if you if you hang out with your friends and you find, like, a couple of friends that all come together to train a model with, like, I don't know. You go to a to a punk festival here, and you have, like, 40 punk bands. And if they all come together and say, let's train a model with all of our songs, we can all share. There's different ways of dealing with that, but that's what I meant before with, like, there's so many things that we don't know what's gonna come.
Matthias Strobel:And I think one of the big things that's gonna come in the next 2 years is, artists having their own models, labels having their own models, selling these models to their fans, fans using these models, and using the output to actually do free marketing and communications for the actual artist because they are proud of what they generated through that thing. The artist is going to share it. There's a lot of opportunity with, like, these open source models. And, yeah, I think that's that's where we're going to go. And, yeah.
Dennis Kastrup:I totally agree also in the sense of a broader picture, of course, that we will all have our assistant AI model who will be focused on us, on everything we do, on everything we say, on everything we play, on everything maybe in the end everything we think. Or, we might, be able to detect what we think one day. I mean, to forecast in a rule in a long in a further future. And, yes, and this will also be happening with music. I mean, I don't know.
Dennis Kastrup:Maybe one day, you will just have your assistant. You sing every morning a melody in your phone because you you like to sing melodies in your phone. And, at the end of the day, or at end of the week, the AI will create a song out of what you've just sung, for a couple of days. Yeah.
Matthias Strobel:And then still made the song of it. Yeah. And then it's still just an inspiration for yourself. And then you can continue doing that in a in a professional way by using a digital audio workstation, using your instrument. Yeah.
Matthias Strobel:But it's just like it makes things so much more accessible. And there's always a difference between the amateur and the professional. And this will always be the case. And I think that, yeah, that's basically, the bottom line is that, that's tech and it can be used by an amateur, by professional, and you will see the difference and you will hear the difference.
Dennis Kastrup:Yeah. I mean, I you would probably agree. I mean, you just said it. Like, I totally agree with all the and that's why I started this podcast with all the possibilities, for musicians and non musicians. It's an exciting time.
Dennis Kastrup:And it's, it's amazing to see what everybody can do and that I can send a funny song to my mother for her birthday and she she loves and she's, happy about this, although it's crappy. But, I made her smile, which is a which is a nice way to give to give something, little, like a little song. So and I was never able to to to to create something. And I already talked to somebody also. I had a coffee in Berlin with and he also said, like, you know, I'm I'm a texter.
Dennis Kastrup:I like to write poems and, I like to it really makes me happy to have my songs being sung by somebody, which is even if it's an AI. And then I put it out there and he puts out really political texts out there and he says, I I have I have the feeling I can change something with it. I don't know. In the end, there may be just 30 people listening to it. But if he has the feeling, oh, I love to do this because I can say something in a different way.
Dennis Kastrup:It's beautiful. And I, I embrace this and I like this a lot with AI and I like this, the things that are happening. So if everybody who's listening always thinks, oh, Dennis is always criticizing the the the way artists are paying. Yes. I do.
Dennis Kastrup:And we have to find a way to, to do this. But I criticize also Spotify and I criticize I also criticized downloads. I didn't in the beginning, but I do now when all this m p three mania came up. I was a downloader. I loved it.
Dennis Kastrup:And then now I think differently about it, which is also strange when you change with with age that you think, wow, that's not nice what you did. And now I know artists who who complain because they are really great artists. And it's also also also one thing, like, we have so many artists out there right now. This is also something we have to keep in mind. It will be more and more difficult to make money with music anyhow.
Dennis Kastrup:I mean, I don't know if you've seen it this, this, thing that, this this headline that was like 120,000 songs a day are generated not generated are uploaded on on streaming, portals like Spotify, teaser, or any any DSP that's that's out there. And, it's like and I think the one guy from Spotify said, it's like in one day more than than a whole year of 1989 or something. What's the quote? Yeah. And that's that's that's that's also where we are going because we will have more and more producers who think they are producers because they have a I I to help them to create something.
Dennis Kastrup:And there will be a a shit load really a a shit load of more music coming out, and it will be even more difficult to, to be to be heard. That's the thing.
Matthias Strobel:That's correct. There's, like, one thing that is really gonna be, like, like, somebody needs to a lot of people need to, produce tools that, like better recommendation tools, filtering tools, make sure to, like, give you only the music that you wanna listen to. And then there's nothing wrong with AI generated music if it, if it serves a purpose. Like, if if people wanna listen to AI generated music because they like it, then, why not? It's it's always been the case that the music who people wanna listen to, that music survives.
Matthias Strobel:And it doesn't matter how and who has, has produced it at the end of the day. That doesn't that doesn't make a difference on the consumer side. And on the artist side, like, if you are a good artist, you had in the entire like, it was always, like, you had to be lucky. You had to had, like, either these days have, like, a viral TikTok thing, TikTok thing, or you had, like, have contacts to, like, A and R people who were lucky to be discovered by an A and R person, get a deal with a label, and get, like, the support in order to make it and to make a living from music. I I like, the whole, like, there's this like, I mean, I'm I was a musician back in the days as well, and I think that the first step why you make music is because you enjoy being creative and expressing yourself through music.
Matthias Strobel:That is like the first step. And the second step is if you do it in a band, you like making music with other people together. And the 3rd step is then you wanna go on stage and share it with other people. And that's actually a 4th step because if you see that people like your music and you like to be on stage, then you wanna continue doing that and go on more stages. And then you realize, okay, I wanna make that my living, and I wanna make money out of that.
Matthias Strobel:But starting with, like, buying an Ableton license and, on your computer and then thinking that you this is you can make a living from that, that's a really stupid way of that's like me buying a a soccer ball and thinking that, Bayern Munich can assign me in the next 2 months. So that's that's extremely stupid and unlikely. So at the end of the day, we shouldn't forget what music is about. Music is mainly about, like, the joy of expressing yourself, in, like, in a musical way and also expressing your thoughts and feelings and all that kind of stuff. And, if the if you like, there's the whole functional music thing that we didn't talk about, adaptive music and games and stuff.
Matthias Strobel:There's a lot of purposes for Gen AI Music that make sense. And, as I said, there's, like, lots of things that we don't think about right now because we're so busy with, like, thinking about the traditional system of the music industry and how AI can can affect it rather than thinking about, like, how AI can actually open up new horizons for jobs, for, music to be heard, for access, for people who don't have the money, the the capabilities, and so on. There's, like, tons of different opportunities on that side, and also for the musicians to actually generate an income that they wouldn't generate before. So there's, yeah, it's an endless topic where we can talk forever about, like, what what AI is actually good or bad for for music in that sense.
Dennis Kastrup:I always give this example. I I also have to believe already that we have heard in the in the in the charts all over the world, songs who are in the top ten, who have been at least generated have generated melodies by an AI, and they just used it in in a different context and maybe maybe made somebody sing it or or whatever. But in the end, and that's what you also said in the end, it is what the artist makes out of it. I mean, I can find I can find a melody which is amazing and which will stuck in your head for the next 3 years. But if I don't find anybody who is more, beautiful than me singing on stage and has more talent to be an entertainer, this song will stay forever in the darkness.
Dennis Kastrup:Or I will make a I or I will use an avatar of me and there will be something else. But I always believe, like, if you listen to a song I gave this example. Maybe I've I've mentioned this on this podcast also sometimes before. I always give this example. If you listen to a song, you make something out of the songs by yourself.
Dennis Kastrup:So I talk to many musicians and I always ask them, so what what was the idea of the song? And they a lot of them tell me, you know, it's your song. I I had an idea what I wrote when I wrote the song but if you take the song it is what you make out of this. If you think I'm singing about my my my husband, my ex wife, my ex husband, my lover, my cat, my dog, my my bird, whatever. It's up to you.
Dennis Kastrup:It's your projection what you take of it. And if that song makes you happy, if I sing about somebody and you think it's your it's your dog, that's your dog. So that's the same thing also about if you're listening to an AI song. If you don't know this song was written by an AI and you still make something out of the song which touches you and you sing this song which you sing this song every day 24 hours. You're feeling sadness by it with it or if you're happy and it makes you just happy.
Dennis Kastrup:And you think maybe that person has sung it and then somebody tells you, oh, it wasn't AI that wrote it. Does it change you? Does it change the experience?
Matthias Strobel:Like, at the end of the day, it doesn't matter. At the end
Dennis Kastrup:of the day It doesn't matter. It's what you said. Yeah. It's what you do with song. And and it and it face it in the end, how many songs of these artists in the top ten have been really written by the people who are singing it.
Matthias Strobel:Yeah.
Dennis Kastrup:Right. So it's already a projection of our us as listeners of people. It's been always a ghostwriter maybe in the background as a producer. The person who's singing it in the ad is just the person who has the best entertainer quality. So everything the product of pop music has always been a mixture of many things.
Dennis Kastrup:And now AI comes in there and, yeah, it's it's just another player. And, let's see where it he, she will take us, whatever we will call the AI. I'm really excited to see where it's going. And I think, we're gonna finish with this enthusiastic, last It's gonna
Matthias Strobel:be exciting times next year. Let's do that again by the end of next year and see where we got from there.
Dennis Kastrup:Yeah. And then we'll quote each other and, and say, okay. You were totally wrong there. And you say you were totally right there. Yeah.
Dennis Kastrup:I'm I'm really excited. And, yeah, as as there are, good good sides of it, there are many bad sides of it. And we talked about
Matthias Strobel:I would make a prediction that by the end of next year, you will have your own, AI model. Like, you personally, Dennis Kaszrup, is gonna have, like, his own AI model, by the
Dennis Kastrup:Like like a general one? I mean, I Yeah.
Matthias Strobel:No. You have your own LLM that you train with your own content. The audio stuff, music stuff, picture stuff, whatever this is. You're gonna have your own like, and most of us people who are interested in that topic will have their own models.
Dennis Kastrup:And you know, I tell you what. Some years ago, I already asked somebody from Google I knew. Like, hey. Can you take all I I've done around, I don't know, 2,000 interviews in my life. And if all of them all of them are on my computer.
Dennis Kastrup:And I asked him, can you make a model of me doing interviews? So train on how I am in interviews so I
Matthias Strobel:can do.
Dennis Kastrup:Yeah. And so I can do I like, I my my other self can do interviews for me. And he said, yes, Dennis. Of course, we can do this. But it will take us how many days?
Dennis Kastrup:I don't remember. Many days. And it will cost you a lot of money. So I said no. But maybe this, this will be easier and quicker and not that expensive in the next month.
Dennis Kastrup:We'll see. And yeah. Maybe you will have a model too. And maybe we just
Matthias Strobel:have our 2 bubble stock. Yeah. Yeah. Let's do that.
Dennis Kastrup:Is it already next year that we'll have the 2 models stuck? I don't think so next year. But, one day.
Matthias Strobel:Who knows?
Dennis Kastrup:Or or maybe the 4 of us. I think the 4 of us. We still have
Matthias Strobel:to keep the human touch. Interesting. Yeah.
Dennis Kastrup:We have to have the human touch. Keep the human touch alive in the music, in the AI world, anywhere.
Matthias Strobel:That's
Dennis Kastrup:true. That's true.
Matthias Strobel:Yeah. That's a really good ending.
Dennis Kastrup:Thank you, Matthias, for talking to me and, giving Thanks
Matthias Strobel:so much for having
Dennis Kastrup:me. Appreciate it. Insights of, AI and music. And as you said, we just opened little, little doors. There's so many.
Dennis Kastrup:I would have talked also about maybe because we always talk about the surface, and and AI is also causing some, economical ecological problems we never talk about. I'm always a little bit sad about that I don't talk about it because I'm so fascinated by the subject in general. But I promise on one of my next episode, I will talk about the ecological impact of AI and which is also a big thing.
Matthias Strobel:No. That's yeah. That's a really big topic. We could talk about that. Let's do that another time.
Matthias Strobel:Thank you very much. Thanks for having me. It was really amazing and fun talking to you.
Dennis Kastrup:This was it. The last episode of the Iliac suite in 2024. What a year it was for AI and music. What a year it was for AI in general. What a year it was in general.
Dennis Kastrup:I wish you a very good 2,025. I hope you will stay with this podcast. If you ever have anything to tell me, you can mail me at mail at the iliacsuite.com, mail at the iliacsuite.com, or you can find me on Instagram at the iliac suite podcast. You will also find the episode as a stream on YouTube. You will find the iliac suite anywhere online if you want to listen to the last episodes.
Dennis Kastrup:Thank you for listening. Take care and behave humans.
Creators and Guests
