"What they mean their filters is like if I had a security camera outside my house, and your car was parked in my parking lot, and I drew with sharpie on the security camera so you couldn't see your car getting stolen, and I was like 'there's a filter up there, your car wasn't stolen." -- Benn Jordan, from "The Truth About Music AI Companies" on Top Music Attorney's channel
@@YoPaulieMusic Heh, yeah, that could use some context. ;) It was part of a bigger discussion about the way that Sudo & Udio apply a Content-ID style filtering mechanism to their output, and eliminate anything that seems too closely related to copyrighted material before they even let you listen to it. The trouble is that they can't make any guarantees about how good that process is; consider that Content-ID systems on something like RUclips are kind of a mess. So, even if you weren't trying to make something that sounds like Taylor Swift's "Shake It Off" or the Beatles or whatever, they can't guarantee that something a jury would find infringing (or a label with big $$$ could just threaten you and you wouldn't have the resources to fight it to bring it to a jury anyway) will slip in from their training data. So unless you're using an AI music company that has a pretty clear documentary record of their training data -- so if something "infringing" in that data slips through, you are still covered -- using it in something like a movie or TV show or video game is awfully risky.
@@Lantertronics I think content-id systems need some more focus by their owners. I've had original music of mine flagged by content ID, the technology as lots of room for improvement. Same applies to any AI-tools being used to flag content on Facebook or RUclips, and the difficulty of being able to find a human to speak with about the issues makes it frustrating for end users.
Good breakdown, Paul, and I think Mikey's got a valid defense, which I support wholeheartedly. I can understand why the record labels are so scared; what I can't understand is why MUSICIANS are so frightened of the new technology. The past has shown us that the sooner one embraces and integrates emerging technologies (hello synthesizers, drum machines, and MIDI, for example), the better one will be positioned to take full advantage of the tech as it matures. That's just smart business. I, for one, can't wait to see where this new paradigm leads us. I'm behind Udio and Suno all the way. -Jay Scott @DeadRockStar49
The notion that these algorithms learn "the same way that humans learn" represents a fundamental misunderstanding of how these algorithms work. It's not unreasonable for someone outside the field to make such an assumption, but it's frustrating to hear people like this Suno CEO, who knows better, make that claim. 12:25: "If Suno is found guilty of illegally copying music, everyone listening to music on RUclips or Spotify for free is also guilty." This is a false comparison. No matter how much you or I may listen to DJ Khalid saying "We the Best Music," no matter how inspiration we may find it, no matter how much we try to sound like DJ Khalid, WE ARE NOT GOING TO SOUND LIKE DJ KHALID THE WAY SUNO CAN USING ITS FEATURE SPACE.
This is not related to learning or training algorithms, but rather to the anatomy and physics of humans. We can replicate DJ Khaled's sound instrumentally 100%, but not his unique own voice - unless you happen to have identical anatomy needed to speak and sing like him. Digital audio does not have this limitation, so it can replicate any imaginable sound. So IF the training algorithm recreates the sound, you have issue with how good the algorithm is rather than what it does or how it does it. And Suno's CEO is kind of right that if we deem this illegal, then we are saying computer/AI training has different rules and laws than humans do, as otherwise any person who imitates another human "too well" should be sued as well.
As someone inside the field, I find your lack of undertanding disturbing. Do you understand AI training algorithms, have you done any research in the specifics of AI training models. If so I would love to better understand your point. We are not talking about traditional electrical engineering concepts. While engineering relies on deterministic rules and programming, AI models for music use large datasets to learn patterns through statistical methods and neural networks. This process is more about teaching than explicit programming, allowing AI to adapt and generate creative compositions. Modern AI in music leverages advanced techniques that go beyond traditional engineering principles. Shulman graduated Columbia with a 3.95 in Math and Physics and has a PhD in quantum computing. He was head of Machine Learning at a major tech firm before leaving to found Suno. My bet is that his understanding of AI is more current than yours, it would be great to hear the two of you engage in a polite discussion on the topic.
@@YoPaulieMusic In all honesty, I'd *love* to hear more about the details of their algorithms. I'm not doubting his credentials or experience at all, but he is running a company (with incredibly impressive technology) and has particular incentives for framing things a certain way.
@@Widiar0 I probably should have been clearer in my initial comments about what the main problem I see is. If a training algorithm recreates a copyrighted sound that's in a training set so that it's largely indistinguishable from a straight sample, something might slip through in one of your AI creations and you wouldn't even know it until another artist sued you. And they'd sue *you*, not Udio and Suno. Napster couldn't defend itself my saying "well, it's not the original 44.1 kHz 16-bit stereo file, it's an MP3 encoding, that's totally different."
I'm not an attorney, but I've also thought from the start that Suno has a good chance of winning, and that it might not even go to trial. The label's team clearly violated Suno's TOS by inputting copyrighted lyrics and intentionally misusing the tool to create facsimiles of copyrighted works. Suno's lyric generation tool uses a seemingly limited set of mundane lyrical tropes that can generally be strung together in about any order and still make sense. This is just smart programing, but there is no way Suno can possibly re-create a copyrighted song lyric unless it was pasted into the lyric prompt. It is literally impossible. That smells like bad faith to me. They had to cheat to make thier case. It's literally fabricated evidence.
However this lawsuit goes, the trouble with these tools is you will never know if music you created with them has material a jury would find infringing or not (see the examples of wild "Cash Money AP" producer tags appearing).
You say there is no way Suno can possibly recreate copyrighted song lyrics unprompted but it already proved it can with the producer tags it spat out verbatim with no prompting from the user. Instead of saying the copyright owners had to cheat to make their case then how about we flip it and say that Suno and Udio had to cheat to make their product which they have commercialised and profited from.
@@royaltyfreemusiccollective8662 Producer tags aren't song lyrics, but maybe they should have used that as thier evidence instead fabricating fake "evidence." It is literally impossible for Suno to generate the lyrics to Johnny B Good (for example) unless they are entered verbatim into the prompt. Whataboutism doesn't let them off the hook. Moreover, the law already protects artists from having thier work copied and sold without thier permission. I suspect the reason no artists have sued is because no artists have been harmed, so they have no standing. The label's standing is dubious, at best. They can't demonstrate that any material/financial harm has come to them, or any of the artists they have contracts with.
@@Lantertronics Suno has probably already fixed this issue, but the answer to your concern is simple; if you ask Suno to generate a hip-hop beat and it spits out a producer tag, delete the tag. Why would anyone want leave some other producer's tags in their track?
Thought question: Drake's song "Taylor Made Freestyle" included an AI recreation of Tupac Shakur's voice. Tupac's estate threatened legal action, so Drake took it down to avoid complications. Did Tupac's estate actually have a valid legal case against Drake?
I think any copies of voices are infirignement and legit claims by those artists. Adelle won a lawsuit over that topic. The issue gets murky when voices sound "similar" to voices. There is a massive differenve between influemnced by versus copying, and I think the courts are not friendly to musicians in this regard. We can easily find singers that sound like other singers... and also apply the same concept to guitar players, sax players, etc. This is certainly going to get interesting.
@@YoPaulieMusic there’s several Peter Gabriel era Genesis cover bands where the singers are spot on (Gabriel said something about taking his kids to see one to show his kids what his dad used to get up to). There’s a band that covers early-era Gary Numan going for the full look - again the singer is spot on. I suppose an AI company could pay them for their voices. Yeah… it gets murky. Apparently there’s something called “publicity rights” separate from usual copyright that comes into play somehow but that’s now well outside my wheelhouse…
The problem with the lawsuit from the majors is, that they want to own and take over something that is not their property. How are they going to prove that hundred percent of Suno is all their property, because it is not and that is where the case will end without success. Because it is actually theft if they will win the case. they will never be able to proof that every song on earth is their copyright property. But however, the majors are free to build their own model, which they can’t so that is where it all will end . Only Suno knows what percentage they used from the majors, and they can say whatever they want what percentage it is, they could say it is only 2%. The majors don’t have the experts to find out. There simply is no law for this kind of case. If you stab someone with a knife, you cannot sue the knife factory, but only the user of the knife himself . The same story counts for Suno. The same counts for cars as well, you could drive it or kill with it, it depends on the person who uses it. I could copy a Beatles song with only stock plug-ins, that means we need to sue all stock plug-ins. No Suno is needed for that . This case is going nowhere.
@@YoPaulieMusic to correctly sue the bad guys in this case we need to sue Logic Pro as well for making it possible to maybe infringement of a major copyright , then we need to sue apple MacBook and Microsoft PC for accomplishment and their responsibility to make it possible for someone to maybe do such copyright theft. Then we need to sue every drummer on earth for copying patterns from majors copyright protected property. Where else can we get money from ? Let’s do all guitarists as well , and don’t forget all USB sticks companies , that copy is even worse. I smell big money here , let’s do it ! We can’t prove we own all the copyright of the planet but let’s just start this and see where it ends just for the fun of it. With all this woke these days , there might even be a good chance to win this nut case ?
@@elmentor8919 Your analogy only makes any sense if Apple included the collected works on the major labels with Logic Pro along with a license for you to use them in your own works.
Copyright laws was made to protect something and the reduplication/theft of that something. Data training and Generative work are not a reduplication of anything. This lawsuits from the get go was all about finding means to make money for themselves, in this case, the record labels. They just want royalty in the training data area, because they know their property are being used to trained the AI. Despite there being no such law that says the owner permission is required to trained an AI system with their data. They know this as well, they just hope no one else does.
Algorithms aren't exempt from copyright and you don't know how copyright works or what the law considers copying/derivative works. What constitutes a copy in law is not based on how information in a computer or program is copied or a computer science definition of "copying", copyright precedes the invention of the modern computer. Someone making a sculpture based on an existing photograph isn't a reduplication or copy but has still been considered copyright infringement in court. And a for profit company scraping the web for copyrighted material to train an algorithm model is 100% copying and reduplication. This is the equivalent of stealing source code from other programs to create another commercial product. The software is already a derivative work. The machine learning on said data and algorithmic model may be considered infringement regardless if its "actually copying" on a computational level. It just has to be recognized by the court as copying.
@@coffin29 Its called copyright law. It's like saying "theres no current law regarding what book I can or can't take the story and characters from to develop my new film." Yes there's "no law" setting guard rails for filmmakers either but you can still be sued by the original author and be found liable for copyright infringement.
@@gensoustudio4703 Yes, that says you cannot copy or infringe on someone else's work. But what about data training for AI? It's a completely new thing. Because the output is not a clone or copy of something that already existed.
@@coffin29 it is not completely new tech at all. There's just not case law established for the mass scale data laundering companies are engaged in that use it.
Great video. I think the idea of having AI generated music is sound. I just think their current methodology and training is flawed. I would prefer a tool that I could interact with iteratively and which would allow me to tweak it. For example the bass sucks try this thing instead. I would also like to ask it what the track was influenced by. Also when I used the tool I want to know that what it is generating is totally new and avant garde and that it was trained in such a way that no one gets pissed off and that no copyright was infringed. I want to be able to talk to it in musical terms and be more involved in the creative process and for it to be more of a collaboration. Right now it doesn’t really resonate with me. The fact that I can’t copyright the stuff generated because it would partially be generated by AI is kind of problematic. So until the tools improve and until copyright law changes or until a better way of training it so that it has better ways of generating brand new progressive musical forms I probably won’t be using these tools. I would almost prefer to have a general chat with ChatGPT about general compositional arrangement and get inspiration that way right now, rather than the AI taking complete control. ChatGPT is great at explaining scales, modes, chords, etc and what could work together without just puking out melodies and chord progressions. I am experimenting with new modes and scales and microtonality because it was suggested to me by AI. I find that way more compelling right now.
Great ideas... all good feature suggestions. Remember, these tools are less than a year old, there is a LONG way to go, but I'm confident that they will get far more impressive, as long as litigation doesn't get in the way.
@@YoPaulieMusicI can imagine a day when you can truly collaborate inside your DAW with an AI that can help you with harmonies or melodies and you can converse with it and collaborate with it like another human. It could generate audio or midi for a single stem that could be manipulated with effects and other processing plugins and VST instruments and you could build the track together. That would be compelling.
@@YoPaulieMusic this case is going to have big consequences for the entire law. Now people can be arrested for what they maybe could do. And that is where we are going now. You are guilty because you may be could do something. Just watch. This is how they want to play it.
Totally disagree with you Paulie. You can't compare training an AI and human learning to decide that what they have done is legal. Humans can't absorb petaflops of data and then regurgitate it ad infinitum at warp speed. Suno and Udio took a risk and recklessly trained on copyrighted material whilst others paid the likes of Pond5 for training data and did it fairly. Is it any surprise that Udio and Suno produce better outputs? They've ingested decades of the best of the best music, and if allowed to continue, there's going to be users landing in legal troubles when it inevitably infringes. I don't agree that folks should be able to make any commercially viable music just by typing a prompt either. Learn music if you wanna make music like every musician before you. I expect the record labels to win although not for altruistic reasons
I get where you are coming from for sure, I've heard this same concern a lot since I started talking about this. I think it's important to differentiate between the capabilities of AI and the principles behind their training methods. Technological advancements, by nature, are usually at a different scale and speed compared to human abilities. This isn't a new phenomenon; many innovations in history have faced similar criticisms. Comparing AI's capacity to process vast amounts of data quickly with human learning isn't quite an apples-to-apples comparison. The main issue of the lawsuit involves the legality and ethics of using copyrighted material for training AI. The fact that AI can learn and reproduce at high speeds doesn't inherently make the training process unfair or illegal. Learning is different than copying or stealing. I appreciate your comment.
That's like saying learn to walk like everybody else, instead of driving a car. It's a tool. Tools are meant to provide helpful options to the users of said tools. That's it.
@@YoPaulieMusic Udio and Suno, of course, haven't revealed the details of their algorithms, but it's probably something along the lines of a souped-up MusicLM, which uses SoundStream as an underlying codec. SoundStream's just a generic audio codec, like MP3, and the "Cash Money AP" producer tag examples have shown that at least Suno can produce sound samples that a jury would find indistinguishable from a straight sample. Napster couldn't get away with a defense like "these files aren't the original stereo 16-bit 44.1 kHz files from the CD, they're MP3 encodings, so that's different data, so it's not infringing," so I wouldn't expect Sudo or Udio (or more worrying, a user of them) to get away with it either. Whatever happens with this case, you will never really know the output you get from these systems has something a jury would find infringing or not, especially when there's lawsuits floating around over a few seconds of a sample.
@@coffin29 No. It's like saying learn to drive like everyone else and earn your drivers licence. Instead of getting a self driving car and calling yourself a driver.
@@royaltyfreemusiccollective8662 You've only taken into consideration the post-car era. Not so much when cars weren't a standard yet. Let's say AI becomes standardized in the future, 20-40 years from now. In that era, no one would say, "Stop using AI". Because it'll be a standard at that point. That's my argument. When a tool exists, people can and will use it, especially when it's not illegal. Current Example: If my kid nephew wanted to make me a song for my Birthday, he's like 6 years old, he wants to make me a gift, why should he have to study music and learn music for 10-20 years just to make me one track as a celebration? By then I'd be dead. So in this scenario, you see how it could be useful as a tool?
@@YoPaulieMusic Suno have publicly admitted to training on copyrighted works. The "AI learns the same way humans do therefore it's fair-use" defense is ridiculous. First off, everyone knows that computers and humans are not comparable. Second, humans get sued all the time for copyright infringement. Third, one of the tests of fair-use is that it cannot undercut the makers of said copyright, which Suno, Udio, and other companies want to do.
Training is not illegal. LIstening to music is not illegal. Analyzing music is not illegal, nor is backwards engineering music. Whether by machine or human, none of that is illegal. This is how musicians learn, and this is how machines learn. What is illegal is copying and distributing pre-existing works. Suno does not do that. Fair use is absolutely legitimate... because listening and analyzing are not the same as copying. Humans that get sued for copyright are being sued for duplicating someone elses work. Many of those cases were absurd bastardizations of the legal system... juries have been persuaded to say that if a song "sounds like" someone elses, or is in the same mood, that it's "close enough." Truly absurd. Copying melodies and lyrics is illegal. Many lawsuits are about who wrote the song first. That's unrelated to what Suno is doing. Ironically, users that paste copyrighted lyrics into Suno are actually the ones breaking Suno's T&C's, as well as actual copyright law. How is Suno undercutting copyright owners? Are you implying that Suno is causing financial damage to Drake, or Taylor Swift, or Bruno Mars, et al? Asbout 120,000 new songs are uploaded to streaming media services every DAY. The market is saturated with music of all qualities and genres. There are no financial damages whatsoever, people are just angry due to ignorance and/or emotions. I appreciate the dialogue.
@@YoPaulieMusic with respect, the arguments you've put forth - the same ones the tech companies often employ to justify stealing other's IP to profit from, are disingenuous at best. Particularly so the favorite argument of these companies that AI is doing the same thing humans do. We all know on a fundamental level that's not the case. The problem is in defining why. However, I think there's a parallel to be drawn from a famous Supreme Court in the 60s concerning obscenity. When asked what his test for obscenity was Justice Stewart replied' "I know it when I see it."
@@SeanGould The quote was about obscenity, not obesity... perhaps an autocorrect error. ;) Learning how music works is different than stealing intellectual property... the two concepts are quite different actually. Intellectual property could be software code, or technical blueprints for specific products. Learning how music works is not stealing. Is learning how to cook stealing? Learning how to draw, or how to lay brick or make cabinets? These tools are analyzing music to understand how it works: tempo, melody, harmony, chords, drums, rhythm, etc. Learning the language of music is drastically different than illegal duplication of someone's musical work.
Suno has been proved to be out putting cut-and-paste songs which is why they are being sued. They use the exact voices and beats and music from the original. This is not AI learning it is theft. This is why Suno will settle before a trial. Most AI is fake and is just cut-and-paste.
This is incorrect. The ability to recreate producer tags is not the same thing as cut-and-paste. The audio is being created, not copied. The producer tags were ingested and "learned" to be part of the genre during the learning process. It's a flaw in the training, but not an example of sampling or copy/paste.
@@coffin29 On a technical level, maybe whats happening on the computer isn't "copying" the way a direct file transfer is copying. But its still a copy. Taking a screenshot of an image on your PC is not "reduplicating what exist" on a computational level the computer is creating an entirely new image with unique data. But no one in their right mind would say that's not a copy of the image you took a screenshot of. David is also correct that digital learning does not exist.
@@coffin29 Doesn't matter how different you personally think it is, or how the tech works. It just needs to be considered a copy by an average person who isn't tech savvy. And that average person is a judge in the case of copyright infringement.
@@gensoustudio4703 Yes, but even that "average person", in this case, the judges, they'd still have to see and verify whether the output is copyright infringement or not. If it's not, then what?
"What they mean their filters is like if I had a security camera outside my house, and your car was parked in my parking lot, and I drew with sharpie on the security camera so you couldn't see your car getting stolen, and I was like 'there's a filter up there, your car wasn't stolen." -- Benn Jordan, from "The Truth About Music AI Companies" on Top Music Attorney's channel
What does this even mean, professor?
@@YoPaulieMusic Heh, yeah, that could use some context. ;)
It was part of a bigger discussion about the way that Sudo & Udio apply a Content-ID style filtering mechanism to their output, and eliminate anything that seems too closely related to copyrighted material before they even let you listen to it.
The trouble is that they can't make any guarantees about how good that process is; consider that Content-ID systems on something like RUclips are kind of a mess.
So, even if you weren't trying to make something that sounds like Taylor Swift's "Shake It Off" or the Beatles or whatever, they can't guarantee that something a jury would find infringing (or a label with big $$$ could just threaten you and you wouldn't have the resources to fight it to bring it to a jury anyway) will slip in from their training data.
So unless you're using an AI music company that has a pretty clear documentary record of their training data -- so if something "infringing" in that data slips through, you are still covered -- using it in something like a movie or TV show or video game is awfully risky.
@@Lantertronics I think content-id systems need some more focus by their owners. I've had original music of mine flagged by content ID, the technology as lots of room for improvement. Same applies to any AI-tools being used to flag content on Facebook or RUclips, and the difficulty of being able to find a human to speak with about the issues makes it frustrating for end users.
@@YoPaulieMusic TRUTH.
Good breakdown, Paul, and I think Mikey's got a valid defense, which I support wholeheartedly. I can understand why the record labels are so scared; what I can't understand is why MUSICIANS are so frightened of the new technology. The past has shown us that the sooner one embraces and integrates emerging technologies (hello synthesizers, drum machines, and MIDI, for example), the better one will be positioned to take full advantage of the tech as it matures. That's just smart business. I, for one, can't wait to see where this new paradigm leads us. I'm behind Udio and Suno all the way. -Jay Scott @DeadRockStar49
The notion that these algorithms learn "the same way that humans learn" represents a fundamental misunderstanding of how these algorithms work. It's not unreasonable for someone outside the field to make such an assumption, but it's frustrating to hear people like this Suno CEO, who knows better, make that claim.
12:25: "If Suno is found guilty of illegally copying music, everyone listening to music on RUclips or Spotify for free is also guilty." This is a false comparison. No matter how much you or I may listen to DJ Khalid saying "We the Best Music," no matter how inspiration we may find it, no matter how much we try to sound like DJ Khalid, WE ARE NOT GOING TO SOUND LIKE DJ KHALID THE WAY SUNO CAN USING ITS FEATURE SPACE.
This is not related to learning or training algorithms, but rather to the anatomy and physics of humans. We can replicate DJ Khaled's sound instrumentally 100%, but not his unique own voice - unless you happen to have identical anatomy needed to speak and sing like him. Digital audio does not have this limitation, so it can replicate any imaginable sound. So IF the training algorithm recreates the sound, you have issue with how good the algorithm is rather than what it does or how it does it. And Suno's CEO is kind of right that if we deem this illegal, then we are saying computer/AI training has different rules and laws than humans do, as otherwise any person who imitates another human "too well" should be sued as well.
As someone inside the field, I find your lack of undertanding disturbing. Do you understand AI training algorithms, have you done any research in the specifics of AI training models. If so I would love to better understand your point. We are not talking about traditional electrical engineering concepts. While engineering relies on deterministic rules and programming, AI models for music use large datasets to learn patterns through statistical methods and neural networks. This process is more about teaching than explicit programming, allowing AI to adapt and generate creative compositions. Modern AI in music leverages advanced techniques that go beyond traditional engineering principles. Shulman graduated Columbia with a 3.95 in Math and Physics and has a PhD in quantum computing. He was head of Machine Learning at a major tech firm before leaving to found Suno. My bet is that his understanding of AI is more current than yours, it would be great to hear the two of you engage in a polite discussion on the topic.
@@YoPaulieMusic In all honesty, I'd *love* to hear more about the details of their algorithms.
I'm not doubting his credentials or experience at all, but he is running a company (with incredibly impressive technology) and has particular incentives for framing things a certain way.
@@YoPaulieMusic I've been meaning to make a video about this issue for a while now, this is a good push for me to get off my butt and do it. :)
@@Widiar0 I probably should have been clearer in my initial comments about what the main problem I see is. If a training algorithm recreates a copyrighted sound that's in a training set so that it's largely indistinguishable from a straight sample, something might slip through in one of your AI creations and you wouldn't even know it until another artist sued you. And they'd sue *you*, not Udio and Suno. Napster couldn't defend itself my saying "well, it's not the original 44.1 kHz 16-bit stereo file, it's an MP3 encoding, that's totally different."
100 agreement!
I'm not an attorney, but I've also thought from the start that Suno has a good chance of winning, and that it might not even go to trial. The label's team clearly violated Suno's TOS by inputting copyrighted lyrics and intentionally misusing the tool to create facsimiles of copyrighted works. Suno's lyric generation tool uses a seemingly limited set of mundane lyrical tropes that can generally be strung together in about any order and still make sense. This is just smart programing, but there is no way Suno can possibly re-create a copyrighted song lyric unless it was pasted into the lyric prompt. It is literally impossible. That smells like bad faith to me. They had to cheat to make thier case. It's literally fabricated evidence.
However this lawsuit goes, the trouble with these tools is you will never know if music you created with them has material a jury would find infringing or not (see the examples of wild "Cash Money AP" producer tags appearing).
You say there is no way Suno can possibly recreate copyrighted song lyrics unprompted but it already proved it can with the producer tags it spat out verbatim with no prompting from the user. Instead of saying the copyright owners had to cheat to make their case then how about we flip it and say that Suno and Udio had to cheat to make their product which they have commercialised and profited from.
I can exactly re-create a Beatles song with just stock plug-ins and no Suno is needed for that, so now we can sue all stock plug-ins as well.
@@royaltyfreemusiccollective8662 Producer tags aren't song lyrics, but maybe they should have used that as thier evidence instead fabricating fake "evidence." It is literally impossible for Suno to generate the lyrics to Johnny B Good (for example) unless they are entered verbatim into the prompt. Whataboutism doesn't let them off the hook. Moreover, the law already protects artists from having thier work copied and sold without thier permission. I suspect the reason no artists have sued is because no artists have been harmed, so they have no standing. The label's standing is dubious, at best. They can't demonstrate that any material/financial harm has come to them, or any of the artists they have contracts with.
@@Lantertronics Suno has probably already fixed this issue, but the answer to your concern is simple; if you ask Suno to generate a hip-hop beat and it spits out a producer tag, delete the tag. Why would anyone want leave some other producer's tags in their track?
Thought question: Drake's song "Taylor Made Freestyle" included an AI recreation of Tupac Shakur's voice. Tupac's estate threatened legal action, so Drake took it down to avoid complications. Did Tupac's estate actually have a valid legal case against Drake?
I think any copies of voices are infirignement and legit claims by those artists. Adelle won a lawsuit over that topic. The issue gets murky when voices sound "similar" to voices. There is a massive differenve between influemnced by versus copying, and I think the courts are not friendly to musicians in this regard. We can easily find singers that sound like other singers... and also apply the same concept to guitar players, sax players, etc. This is certainly going to get interesting.
@@YoPaulieMusic there’s several Peter Gabriel era Genesis cover bands where the singers are spot on (Gabriel said something about taking his kids to see one to show his kids what his dad used to get up to). There’s a band that covers early-era Gary Numan going for the full look - again the singer is spot on. I suppose an AI company could pay them for their voices.
Yeah… it gets murky. Apparently there’s something called “publicity rights” separate from usual copyright that comes into play somehow but that’s now well outside my wheelhouse…
The problem with the lawsuit from the majors is, that they want to own and take over something that is not their property. How are they going to prove that hundred percent of Suno is all their property, because it is not and that is where the case will end without success. Because it is actually theft if they will win the case. they will never be able to proof that every song on earth is their copyright property. But however, the majors are free to build their own model, which they can’t so that is where it all will end . Only Suno knows what percentage they used from the majors, and they can say whatever they want what percentage it is, they could say it is only 2%. The majors don’t have the experts to find out. There simply is no law for this kind of case. If you stab someone with a knife, you cannot sue the knife factory, but only the user of the knife himself . The same story counts for Suno. The same counts for cars as well, you could drive it or kill with it, it depends on the person who uses it. I could copy a Beatles song with only stock plug-ins, that means we need to sue all stock plug-ins. No Suno is needed for that . This case is going nowhere.
Absolutely correct... it's a mess because emotions and greed are blocking rational thinking.
@@YoPaulieMusic to correctly sue the bad guys in this case we need to sue Logic Pro as well for making it possible to maybe infringement of a major copyright , then we need to sue apple MacBook and Microsoft PC for accomplishment and their responsibility to make it possible for someone to maybe do such copyright theft. Then we need to sue every drummer on earth for copying patterns from majors copyright protected property. Where else can we get money from ? Let’s do all guitarists as well , and don’t forget all USB sticks companies , that copy is even worse. I smell big money here , let’s do it ! We can’t prove we own all the copyright of the planet but let’s just start this and see where it ends just for the fun of it. With all this woke these days , there might even be a good chance to win this nut case ?
@@elmentor8919 Love the idea... we gonna be RICH! :)
@@YoPaulieMusic 🤣
@@elmentor8919 Your analogy only makes any sense if Apple included the collected works on the major labels with Logic Pro along with a license for you to use them in your own works.
Copyright laws was made to protect something and the reduplication/theft of that something. Data training and Generative work are not a reduplication of anything. This lawsuits from the get go was all about finding means to make money for themselves, in this case, the record labels. They just want royalty in the training data area, because they know their property are being used to trained the AI. Despite there being no such law that says the owner permission is required to trained an AI system with their data. They know this as well, they just hope no one else does.
Algorithms aren't exempt from copyright and you don't know how copyright works or what the law considers copying/derivative works. What constitutes a copy in law is not based on how information in a computer or program is copied or a computer science definition of "copying", copyright precedes the invention of the modern computer. Someone making a sculpture based on an existing photograph isn't a reduplication or copy but has still been considered copyright infringement in court. And a for profit company scraping the web for copyrighted material to train an algorithm model is 100% copying and reduplication. This is the equivalent of stealing source code from other programs to create another commercial product. The software is already a derivative work. The machine learning on said data and algorithmic model may be considered infringement regardless if its "actually copying" on a computational level. It just has to be recognized by the court as copying.
@@gensoustudio4703 Except there's no current law regarding what can or can't be used to train AI. If we narrow it down to that specifics.
@@coffin29 Its called copyright law.
It's like saying "theres no current law regarding what book I can or can't take the story and characters from to develop my new film."
Yes there's "no law" setting guard rails for filmmakers either but you can still be sued by the original author and be found liable for copyright infringement.
@@gensoustudio4703 Yes, that says you cannot copy or infringe on someone else's work. But what about data training for AI? It's a completely new thing. Because the output is not a clone or copy of something that already existed.
@@coffin29 it is not completely new tech at all. There's just not case law established for the mass scale data laundering companies are engaged in that use it.
I want ai to win.
Great video. I think the idea of having AI generated music is sound. I just think their current methodology and training is flawed. I would prefer a tool that I could interact with iteratively and which would allow me to tweak it. For example the bass sucks try this thing instead. I would also like to ask it what the track was influenced by. Also when I used the tool I want to know that what it is generating is totally new and avant garde and that it was trained in such a way that no one gets pissed off and that no copyright was infringed. I want to be able to talk to it in musical terms and be more involved in the creative process and for it to be more of a collaboration. Right now it doesn’t really resonate with me. The fact that I can’t copyright the stuff generated because it would partially be generated by AI is kind of problematic. So until the tools improve and until copyright law changes or until a better way of training it so that it has better ways of generating brand new progressive musical forms I probably won’t be using these tools. I would almost prefer to have a general chat with ChatGPT about general compositional arrangement and get inspiration that way right now, rather than the AI taking complete control.
ChatGPT is great at explaining scales, modes, chords, etc and what could work together without just puking out melodies and chord progressions. I am experimenting with new modes and scales and microtonality because it was suggested to me by AI. I find that way more compelling right now.
Great ideas... all good feature suggestions. Remember, these tools are less than a year old, there is a LONG way to go, but I'm confident that they will get far more impressive, as long as litigation doesn't get in the way.
@@YoPaulieMusicI can imagine a day when you can truly collaborate inside your DAW with an AI that can help you with harmonies or melodies and you can converse with it and collaborate with it like another human. It could generate audio or midi for a single stem that could be manipulated with effects and other processing plugins and VST instruments and you could build the track together. That would be compelling.
@@YoPaulieMusic this case is going to have big consequences for the entire law. Now people can be arrested for what they maybe could do. And that is where we are going now. You are guilty because you may be could do something. Just watch. This is how they want to play it.
@@martinhanson4281 I think a *great* example of proper (and non-legally-ethically-problematic) use of AI is the various iZotope tools.
Totally disagree with you Paulie. You can't compare training an AI and human learning to decide that what they have done is legal. Humans can't absorb petaflops of data and then regurgitate it ad infinitum at warp speed. Suno and Udio took a risk and recklessly trained on copyrighted material whilst others paid the likes of Pond5 for training data and did it fairly. Is it any surprise that Udio and Suno produce better outputs? They've ingested decades of the best of the best music, and if allowed to continue, there's going to be users landing in legal troubles when it inevitably infringes. I don't agree that folks should be able to make any commercially viable music just by typing a prompt either. Learn music if you wanna make music like every musician before you. I expect the record labels to win although not for altruistic reasons
I get where you are coming from for sure, I've heard this same concern a lot since I started talking about this. I think it's important to differentiate between the capabilities of AI and the principles behind their training methods. Technological advancements, by nature, are usually at a different scale and speed compared to human abilities. This isn't a new phenomenon; many innovations in history have faced similar criticisms. Comparing AI's capacity to process vast amounts of data quickly with human learning isn't quite an apples-to-apples comparison. The main issue of the lawsuit involves the legality and ethics of using copyrighted material for training AI. The fact that AI can learn and reproduce at high speeds doesn't inherently make the training process unfair or illegal. Learning is different than copying or stealing. I appreciate your comment.
That's like saying learn to walk like everybody else, instead of driving a car. It's a tool. Tools are meant to provide helpful options to the users of said tools. That's it.
@@YoPaulieMusic Udio and Suno, of course, haven't revealed the details of their algorithms, but it's probably something along the lines of a souped-up MusicLM, which uses SoundStream as an underlying codec. SoundStream's just a generic audio codec, like MP3, and the "Cash Money AP" producer tag examples have shown that at least Suno can produce sound samples that a jury would find indistinguishable from a straight sample. Napster couldn't get away with a defense like "these files aren't the original stereo 16-bit 44.1 kHz files from the CD, they're MP3 encodings, so that's different data, so it's not infringing," so I wouldn't expect Sudo or Udio (or more worrying, a user of them) to get away with it either. Whatever happens with this case, you will never really know the output you get from these systems has something a jury would find infringing or not, especially when there's lawsuits floating around over a few seconds of a sample.
@@coffin29 No. It's like saying learn to drive like everyone else and earn your drivers licence. Instead of getting a self driving car and calling yourself a driver.
@@royaltyfreemusiccollective8662 You've only taken into consideration the post-car era. Not so much when cars weren't a standard yet. Let's say AI becomes standardized in the future, 20-40 years from now. In that era, no one would say, "Stop using AI". Because it'll be a standard at that point. That's my argument. When a tool exists, people can and will use it, especially when it's not illegal. Current Example: If my kid nephew wanted to make me a song for my Birthday, he's like 6 years old, he wants to make me a gift, why should he have to study music and learn music for 10-20 years just to make me one track as a celebration? By then I'd be dead. So in this scenario, you see how it could be useful as a tool?
I don't know if using an absurd defense is "punching back", per se...
What do you consider absurd about his stance?
@@YoPaulieMusic Suno have publicly admitted to training on copyrighted works. The "AI learns the same way humans do therefore it's fair-use" defense is ridiculous. First off, everyone knows that computers and humans are not comparable. Second, humans get sued all the time for copyright infringement. Third, one of the tests of fair-use is that it cannot undercut the makers of said copyright, which Suno, Udio, and other companies want to do.
Training is not illegal. LIstening to music is not illegal. Analyzing music is not illegal, nor is backwards engineering music. Whether by machine or human, none of that is illegal. This is how musicians learn, and this is how machines learn. What is illegal is copying and distributing pre-existing works. Suno does not do that. Fair use is absolutely legitimate... because listening and analyzing are not the same as copying.
Humans that get sued for copyright are being sued for duplicating someone elses work. Many of those cases were absurd bastardizations of the legal system... juries have been persuaded to say that if a song "sounds like" someone elses, or is in the same mood, that it's "close enough." Truly absurd. Copying melodies and lyrics is illegal. Many lawsuits are about who wrote the song first. That's unrelated to what Suno is doing. Ironically, users that paste copyrighted lyrics into Suno are actually the ones breaking Suno's T&C's, as well as actual copyright law.
How is Suno undercutting copyright owners? Are you implying that Suno is causing financial damage to Drake, or Taylor Swift, or Bruno Mars, et al? Asbout 120,000 new songs are uploaded to streaming media services every DAY. The market is saturated with music of all qualities and genres. There are no financial damages whatsoever, people are just angry due to ignorance and/or emotions.
I appreciate the dialogue.
@@YoPaulieMusic with respect, the arguments you've put forth - the same ones the tech companies often employ to justify stealing other's IP to profit from, are disingenuous at best. Particularly so the favorite argument of these companies that AI is doing the same thing humans do. We all know on a fundamental level that's not the case. The problem is in defining why. However, I think there's a parallel to be drawn from a famous Supreme Court in the 60s concerning obscenity. When asked what his test for obscenity was Justice Stewart replied' "I know it when I see it."
@@SeanGould The quote was about obscenity, not obesity... perhaps an autocorrect error. ;)
Learning how music works is different than stealing intellectual property... the two concepts are quite different actually. Intellectual property could be software code, or technical blueprints for specific products. Learning how music works is not stealing. Is learning how to cook stealing? Learning how to draw, or how to lay brick or make cabinets? These tools are analyzing music to understand how it works: tempo, melody, harmony, chords, drums, rhythm, etc. Learning the language of music is drastically different than illegal duplication of someone's musical work.
👍👍👍
Suno has been proved to be out putting cut-and-paste songs which is why they are being sued. They use the exact voices and beats and music from the original. This is not AI learning it is theft. This is why Suno will settle before a trial. Most AI is fake and is just cut-and-paste.
This is incorrect. The ability to recreate producer tags is not the same thing as cut-and-paste. The audio is being created, not copied. The producer tags were ingested and "learned" to be part of the genre during the learning process. It's a flaw in the training, but not an example of sampling or copy/paste.
Digital Learning does not exist - a Digital Copy is a Copy, there is no difference.
To say it is a copy implies it is reduplicating what existed. This isn't that.
@@coffin29 On a technical level, maybe whats happening on the computer isn't "copying" the way a direct file transfer is copying. But its still a copy. Taking a screenshot of an image on your PC is not "reduplicating what exist" on a computational level the computer is creating an entirely new image with unique data. But no one in their right mind would say that's not a copy of the image you took a screenshot of.
David is also correct that digital learning does not exist.
@@gensoustudio4703 But Generative AI isn't an exact "screenshot" of what existed, is it? Generative tech works differently.
@@coffin29 Doesn't matter how different you personally think it is, or how the tech works. It just needs to be considered a copy by an average person who isn't tech savvy. And that average person is a judge in the case of copyright infringement.
@@gensoustudio4703 Yes, but even that "average person", in this case, the judges, they'd still have to see and verify whether the output is copyright infringement or not. If it's not, then what?