9:09 exactly! Copyright gets spun as protecting the creatives but any benefit to the creatives is entirely collateral damage to the publishers' bottom line.
Same can be said for every book author. Aren't they stealing from every author they've ever read? Same for painters or artists. These record labels are praying on people that don't understand how AI works or learns.
A brilliant explanation about the reality of the situation versus what the RIAA would like it to be. The process of learning based on exposure to said media is not a copyright infringement - Otherwise every artist that listened to and learned to play Eruption by Van Halen (non commercially) is breaking copyright (they aren't and are covered in fair use). To learn how to create in the style of is a hallmark of every human artist today in every medium as part of the social construct of learning in and of itself as evidenced by asking who their greatest "influences" are. The sum of human achievement is marked by prior works and transformative application thereof. We applaud such things when it's humans but demonize it when an AI learns to do the same because of biological "exceptionalism". We say "These things were used to 'train' an AI", but don't bat an eyelash when a human learns through example to do the same. What we're talking about is the fundamental exercise of the act of learning overall through mimicry and exposure. Something the entire human race depends on in order to build upon and create new works. Just as a human that learned to play like a certain artist can also play that artists songs when asked, so too an AI is capable, just as well as creating new works in that style. In that the RIAA specifically cited outputs that they input to get such results, and by which they own the copyrights to, is akin to them asking a session musician to play a song the RIAA owns and immediately crying foul for the ability to do so. Ask any capable guitarist to play the opening to Layla by Derek and the Dominos. Does the RIAA get to sue every guitarist for knowing how to do so because they've heard the song, and learned how to play it? No, because that would be absurd. However, just as there would be limitations on what could be published for commercial purposes due to copyright, so too the AI would be subject to the same rules and uses regardless of the learning process. In which case, commercial use cases of non-transformative style transfer which duplicates discernable portions or whole works would require licensing as outputs. This would fall in line with how we operate with human learning and performance output and would not jeopardize the very premise of learning overall.
The litmus test would therefore be: Could a human learn to perform or produce a similar output through exposure to prior works? If the answer is "Yes" then the AI learning process is falling on the same premise but still bound to the same non-commercial output and/or licensing if said output were commercial use and were to be reproducing wholesale or in adequate portions to warrant licensing (sampling or otherwise). The learning process itself is moot if a human can reproduce through exposure a style or output the same. It's the output and intended purpose which is judged the same as a human and whether licensing is needed for prior works or if transformative aspects have been applied. If you asked an AI to reproduce ABBA and it did, after you gave it the lyrics and style... then yes. Licensing is needed if it is for commercial purposes. The same would be true for human created works. But this is very specific and does not cover "general style" - as evidenced by the industry trying to sue John Foggerty for sounding like himself (and CCR) in his solo career.
Thank you for your cogent arguments. This is how I view it too. Anyone who is triggered by AI being fed copyrighted material isn’t considering the fact that you couldn’t feed it any material other than copyrighted material. Even if the music was public domain, the recording isn’t. So, what were they supposed to do? Sing to the AI? Then that would be a Performance Rights issue. People also don’t realize (or consider) the Majors own a small sliver of all copyrighted music. There’s 100,000 records uploaded to streaming a day, all of it copyrighted, little of it by the Major Labels. I don’t need to feed my AI any Major Label music, and it would learn precisely the same things musically, because for every hit owned by a Major there are thousands of songs that do similar things. So the whole argument that feeding the AI copyrighted music is infringement is actually argument absurdium. The argument that you could come up with a way to pay everyone who influenced the AI is equally absurd, and doesn’t take into account that those who came before, should get more than those who built on to after. Prince was way more popular than Robert Johnson, but without Robert Johnson (and all the artists between) should arguably be paid more than for his contributions, than Prince (or their estates at this point). Thanks for your post.
I have multiple problems with your arguments but... 'machine learning' doesn't LISTEN to music any more than a graphic equalizer LISTENS to music... Imagine this... I go to the movie theater and I start recording the screen with my phone. When the usher, inevitably, taps me on the shoulder and says that I'm "stealing" the movie, +I+ say-- "My phone is just WATCHING the movie. What's the difference between ME watching the movie and my phone WATCHING the movie??" "So, my phone REMEMBERS what it saw... so do I. I ALSO remember what I saw. So, what's the difference between me telling my best friend about the plot of the film and me re-playing the movie for them... from my phone?" I'm utterly perplexed by the whole, "Computers are just doing what humans do" argument regarding 'AI'. The mechanism by which an action is performed can completely change the nature of that action. To believe otherwise is a descent into language games that are, insanely, far removed from reality...in my opinion.
The phone and the sampler reproduce what they record, and when that reproduction occurs, it is absolutely infringement of a particular performance. The AI assimilates everything that it hears, and then makes new music, which pulls from all that it’s heard, like influences, the same way that humans do. Perhaps you should make a song or two on Suno or Udio to understand better how it works. If you make a song with Suno, it makes a song that is accessible, that pulls from themes and genres out of millions of possibilities and combines them, but does not pull specifically from anything in particular. This is precisely how people make music. I know it’s how I make music. So, your argument that recording a performance is the same thing is rather weak and nonsensical. My next video will demonstrate a track that I made with Suno, and the smart musical decisions that it made, none of which infringe on anything. It has to specifically steal a melody and a lyric for it to be infringement.
@@MixermanPublishes The truth is that neither you, nor I, have access to the code these sites are working with so it's impossible for you to say, with any certainty, HOW they operate. You're just regurgitating Udio's/Sunos' talking points. (And, you DO understand the incentive for these companies to obscure the truth, right?) It diminishes anybody's argument when they speak so authoritatively about something that is impossible for them to know or verify... When their main source of information is the (possibly) offending party themself... who has all the incentive in the world to be dishonest. I have used both services and read both services' TOS's. What alarmed me was Udio saying that users should be aware that Udio might spit out the +exact same song+ for multiple people on multiple occasions. That doesn't imply a creative, iterative process. It implies an algorithmic "re-combination" of collected pieces. It implies something closer to 'reproduction' than 'assimilation'. Pre-determined outcomes. It indicates a system that has no "memory" to be aware of what it's done in the past... or even an awareness of what it's doing on another open, concurrent, session. These things are NOT analogous to how a human "assimilates" and creates. In the same way that it's a felony to video-record a movie screen (without permission), it should be illegal to aim "machine learning" at copyrighted music. I don't really understand how people see a difference between the two.
@@makers_lab You're speaking in general and I'm talking specifically about two companies that seem to have mastered this procedure in a way that no company before them has been able to accomplish. If "technical papers" could confirm just how Udio and Sunos are operating, there would be no need for a lawsuit.
The problem is, that it's not "learning" shit... It's noticing patterns and relating them to parts of a potential worded prompt. It's not learning why this melody sounds sad. Why this one sounds nostalgic. It's noticing that a song with similar melody had a hashtag "sad" in it's description when it was going through the training dataset. Another thing is ... if I sample a song even using it transformatively, I am legally subjected to the possibility of being sued to oblivion. So if judge rules out that "learning is not stealing", I'd like to learn a and use melodies of famous songs in my medley and monetize it. Thank you. And lastly ... there are damages. Huge ones. On sync licensing. Nobody will license Beatles for their background music when they can get close enough approximation from these services for free. That is a monetary hit. (And ofr us small musicians even bigger. Demand will drop to 0.)
That's what learning is. Neural networks that power AIs are basically just a simulation of how neurons in the human brain work. You can find people on UA-cam who make songs in style of X artist that are free and people still buy the actual artists stuff. People still fork out money for hand made stuff even though industrial products are cheaper. Ironically I've actually found artists and bought their content from seeing AI version. Maybe you operate solely on what is cheap and convenient but the entire free market does not.
@@ragnarok7976 I intend to hand make music with the assistance of AI. It’s nowhere near where I need it yet. There is very little control. But once I can direct AI like it’s a human, then I can hand make music with AI as my all purpose, muti-talented assistant, and I will be able to make music far more efficiently, for expanded purposes.
What is the best tool you can use to detect and measure presence of copyrighted material in an a new song? I’m familiar with MIPPIA. Are there even better ones? There needs to be such a tool to detect but most importantly provide copyright holder information. Should someone produce something that happens to contain copyrighted material, such a tool would make convenient to the producer the ability to pursue licensing. One would think that record labels would invest in creation of such a tool rather than in such a lawsuit. They would stand to profit from it. With distributors on board as well, they could apply such a tool to every work before distributing and then offer to the artist the information as to where to pursue licensing if needed.
If there’s a hit song that infringes, then there will be a suit. Many suits. As there almost always is for every hit song. You want to make a tool that can allow someone to find songs that aren’t hits, and worse yet are likely obscure, so that we can all sue each other for statutory damages? That would cause abuse and would clog up the courts so badly that you would force Congress to kill statutory damages, which is actually a critical component of Copyright protections, because it dissuades people from purposely infringing with impunity. Melodic infringement is always largely debatable. So, I don’t know how an algorithm can pick out something that’s debatably infringement. Lyrical infringement requires outright stealing. Nothing else is copyrightable. Not style. Not production. Not arrangement. Not harmony (chord progressions). What you suggest isn’t possible, and if it is, will cause the end of copyright as we know it.
@@MixermanPublishes My only thought was something to allow at least me to determine if a new song I’ve written is highly comparable to another song, meaning 80 to 100% similar in music or lyric or both.
@@jeffhulrichunless what you write is a hit song, you don’t have to worry about being sued for accidental infringement. You have to be given the opportunity to take an infringing work down before you can be sued for the $150,000 in statutory damages.
@@MixermanPublishes thanks for the insight. I really appreciate the video. I just discovered your channel yesterday on this topic and found it very useful. Subscribed.
Thank you for making this video. This is no different than if one of us went to the Coca-cola Headquarters and demanded to see the original recipe because we think it still contains cocaine. Or demanded to know the original herbs and spices in KFC. These are requests for trades secrets and no one would take it seriously. So why do people suddenly think it's alright to demand to see the trade secrets of the Suno or Udio training data?
because people want retribution for the death of their dreams... people dream of living off of their art until they have to do so and they pull their hair out, but now they are mad that robots are taking those dream jobs away and flipping burgers seems like it was the more solid career choice all along except they're making automated McDonalds now. Companies making their money predominantly from AI should be taxed high enough to make Democrats happy. Companies not relying on AI should be taxed low enough to make a Republican blush.
I have to disagree on the point AI learns the same way as humans do. We learn by listening, practicing by actually playing ot on a instrument or singing it, we 'feel' the song as we learn it.
It’s true that we feel and the AI doesn’t. But once you understand what tools to use to make people feel a certain way, it becomes a plug and play process. The difference is, the AI has to be told what makes us feel what way, but then, you actually are taught those things in college level arranging classes too, especially where it comes to scoring. All the composers for movie scoring all use the same phrases snd scales for movie music. There are some that are more inventive than others, but the techniques that are used to make us as viewers feel a certain way are all plug and play techniques. So, yes, it doesn’t have to practice scales, it doesn’t have to discover the magic of music, but if it understands what makes humans feel a certain way, it can plug and play exactly the way humans do, who make enormous volumes of music at a very high level. There’s a reason why movies scorers never sue each other. It’s pointless. Yet, there are movies that have ridiculously similar sounding scores to others (especially in the 90s and aughts), with only slight variations. When I say that AI learns how to make music exactly as humans do, I mean the AI learns the patterns that make effective music that cause a certain physiological reaction in humans. It’s clear that these generative AIs have been taught all of this, because if you give it a sad lyric, it generally makes the music sad to match. The music it spits out is in alignment with the lyric, or the concept, that that the AI is presented by the user. That means it’s making musical decisions, and you may not like that term, but I wouldn’t know what else to call it. I’ll be posting a video demonstrating the good decisions an AI makes based on a lyric that it was provided. I know humans who don’t use some of the good techniques the AI does. The problem is most people are mediocre at music making, and the AI is already superior to it than 80% of them.
Don't get me wrong I think the majors are pure EVIL....But honestly, how many people are NOT going to publish these songs? Putting them onto UA-cam is a form of publishing. Someone makes a cool song using one of these mentioned services (or others) who is not tempted to share it on a mass level?
If these major labels win, I'm gonna send them all a CD with my original music. And then I'll sue them for listening to my music without having the rights to my music. How much was that per song?
Sorry to say this but you're 100% right & These major label are gonna lose this lawsuit because technically these A.i. platforms or tools, instruments that are no different from the MPC samplers technically its just a more advanced version. Put it this way, YOU CANT SUE AKAI FOR MAKING A SAMPLER BUT YOU CAN SUE THE ARTIST & PRODUCER WHO SAMPLE MATERIAL & PUBLISH IT WITHOUT CLEARING SAID SAMPLE. UDIO & SUNO ARE NO DIFFERENT THAN AKAI MPC's if you see what I am saying here. Yeah SUNO & UDIO are making thousands of dollars but so is AKAI for doing the same thing which is Sampling!!!
What company has copied YOUR work? Streaming companies are what killed people’s ability to make money with music. This is already done. Our only chance is to become good with AI at this point. We can either accept this, or hate our lives. I make 20 percent of what I made on a per mix basis in the aughts. I make half what I made per mix in 1995. What makes it worth it for me is I can mix faster now, I can mix on my own time schedule, and I never have to leave my house. I can also sell books, which are far more lucrative for me now that I can publishes them myself, That doesn’t make any of it a less of a bitter pill to swallow. This is the reality. There’s nothing you or I or anyone else can do about it, because we don’t have a government that protects our interests, but rather the interests of large corporations. And now Americans are poised to end Democracy, which will only make that worse. We can either play in the sandbox we have, or lament the way it should be. I recommend the former. With all the problems coming down the pike, we need to protect our own interests, at this point.
I stand corrected: A correct and brilliant post by mixerman. :-)) Gratulations, you are completely right. BTW: EVERY person learning to make music uses not-licensed listening to other persons music ...
what? no they don't. not unless you're pirating the music you listen to? if you're listening to music on a streaming service, that music was licensed for you to listen to. you're using the music the way it was licensed for you to use. simple as that.
@@RasmusSchultz so I can’t stream music while you’re in the room? You have a strange understanding of how streaming works. Besides, Suno can buy a subscription specifically for their AI. This argument of yours makes no sense whatsoever.
@@MixermanPublishes you're still going on about things that have nothing to do with what they did. AI is not a person in a room. The model is a product derived from copyright recordings and then sold. You're not allowed to do that. The end.
"played in the presence of their AI" spoken as if you are talking about an intelligence of some sort. This will boil down to that argument. At this point and time, is machine learning the process of teaching an 'intelligence' or will this be deemed the process of a copy writing machine that takes in information and spits out an auto correct/complete. The answer will likely fall on the side of, what happens when you don't feed an intelligence other songs. Can it produce from nothing? The question when it comes to a computer, is simple, it can't do anything without being fed full completed content.
Especially with original lyrics. An AI created song that uses original lyrics is not 'infringing'. You can't 'copyright' popular chord progressions. You can't copyright 'genres' of music. You can't copyright a 'feel'. You can only copyright hard, unambiguous content. If these companies are not COPYING these works, or distributing them, or broadcasting them, then they're NOT running afoul of a copyright. That's silly. That's like trying to say I'm not allowed to learn how to play a Stevie Ray Vaughn song without paying him. The judge in this case will ask, "Did you copy, distribute, or broadcast any copywritten material when training your AI models"? Then ask the plaintiff: "Do you have any proof that these companies copied, distributed, or broadcast any intellectual property owned by you? Because the answer to BOTH of those questions is no......
@@BrockBarr but who am I paying? The Majors only own a small sliver of all copyrighted music. Why should they collect the money for raw materials that come from everyone in music history? And if everyone in music history is to be paid, what mechanism allows for that? And shouldn’t the estates of the music pioneers get more? And who should determine that? And if I’m training my AI, how much should I pay for each track? So each artist is going to get one payment per song for pennies and then the AI is trained, and then that AI engine is licensed out to AI developers? You just cost the developers billions to pay each person less than the price of a sandwich? How does that make any sense? It’s all well and good to make a statement that a business should pay for raw materials, but there isn’t a market in existence for these materials. They’re essentially free to anyone with a streaming subscription. And you need to come up with a way to pay everyone. And you need to explain how that going to amount to anything since and AI need only evaluate any given song one time.
Why should companies be left to being the only ones to train A.I on anything? A.I should be accessible to all the people. It's merely information that can learn and improve. I say impeding that is infringing on Freedom of Speech because, it says what we can and can't do with information as only information. Imagine if the internet was that way and only corporations alone had access to and ran it. I say it's time for corporations to collapse. If we can all do it there's no power nor monopoly focused in one place that can be corrupted. A.I is too powerful to only be in the hands of the few.
I prompted Ecosia chat "The Major Labels are suing Ai companies Udio and Suno for copyright infringement, but are they really infringing?" and this is what it said: "I don't have specific information on the legal case you mentioned involving Udio and Suno. Copyright infringement cases can be complex and often involve detailed analysis of the specific circumstances. It's best to refer to official statements from the parties involved or legal experts for accurate information on the situation. If you're interested in learning more about copyright laws and how they apply in cases like this, I can provide some general information on the topic. Would you like to know more about copyright laws?"
So, it’s very possible the labels have managed to generate some specific instances of infringement. But do we really believe for one second, that even if the labels were able to generate 10 instances of legitimate, clear and obvious infringement, that they are wasting their time suing for $1.5 million dollars? That’s their sushi budget. And do we believe for one second that Udio and Suno, (both companies now with over $125 million in seed money each) wouldn’t just pay that off and make it go away if they could, and teach their AI how to better not infringe? I don’t. Like I said, this is about forcing the AI companies to reveal their training methods through the Discovery process. It’s not about the money. It’s about the money.
@@MixermanPublishes I thought "It’s not about the money. It’s about the money." was a Bill Graham quote but apparently it is actually from H.L. Mencken - ‘When somebody says it’s not about the money, it’s about the money.’ Either way it rings true!
Today these models aren't great at reaching outside their datasets, for example making something extremely novel. They tend to stick to the bounds of what they've consumed. Commentators are latching pretty hard to that. To me, humans aren't much different. Most novices for example do nothing but mimic. I think this limitation in AI today will be quickly behind us, and then the argument in this video will be more widely palatable.
I'm not arguing with the general premise that the major labels are just trying to get some more money out of the AI companies and this is not something that is going to work for artists. And that the actual solutions that works for creators and humanity is probably more complicated and requires thinking outside of the current system. That being said, AI doesn't actually "listen" or "learn", these are analogies that are used. A machine can't "listen" or "learn" in it's truest self as the words exist to describe how humans operate. Additionally, the current norms we have, allowing humans to listen to music freely, to learn and practice established works, etc. are rooted in our physiological limitations and doesn't apply to AI. The whole why can a human listen to music but not AI is silly. Why can a human go to the beach undisturbed but when a bear shows up they get trapped by animal control. Humans and Animals don't have the same rights. Machines and humans don't have the same rights either. I'm not saying specifically if "listening to music" should or shouldn't be allowed for a machine freely. All I'm saying that it's not automatically OK because it's OK for humans to do.
@@FicoosBangaly lots of people get hung up with the terms learn, listen, and decisions, (video on “decisions” next week). But not one person has actually come up with words to describe what the AI is doing that seems more accurate. One person out of probably a hundred gave me an alternative, and it was weak. People get triggered when I use human terms for a machine, but based on what AI does, and how it works, I can’t come up with anything more accurate. So, to me? If the definitions of those words only relate to humans, then the definitions need to change. And I’m not going to use klunky language to describe what I can use one word for, and that conveys my message in an understandable way, just to satisfy people who I find are being rather pedantic about it. Y’all want me to change my language to suit your discomforts, without even providing an alternative that’s accurate and appropriate. I want to redefine the language, since it’s dynamic and ever changing, and that’s what I shall do. Since I’m the one conveying the message, I’m the one who decides how I do it. You’re welcome to use your own terminology.
@@MixermanPublishes You can use analogies when they are convenient. But you need to remember that they are analogies. People talk about computer viruses and it's fine. They are "viruses" in the way that they are a piece of parasitic code in a host program corrupting it and causing it to make the computer "sick". It is *the* name for it. But if you tell me that because it is a virus it has a certain amount of genetic material, I'll tell you that you are probably getting confused. Additionally, there are words that have a slightly different meaning when used in different contexts. For example the word theory which means different things in the common speech and in a scientific context. There are words and exact descriptions for what different machine learning implementations are doing when they ingest input (i.e. "trained"). Analogies are fine and helpful and people can continue using them as long as they understand that they are "analogies" and that the concept of an algorithm "learning" is different from a person learning in the same way as when an evolutionary biologist saying that a gene "wants" to replicate.
I'm really not fond of the now burgeoning intellectual property industry, but I have to disagree with you. It's not the song created by the AI that's the infringement being sued over, it's the pattern of the song embedded in the AI that's providing value being marketed to users of the AI.
Patterns aren’t protected by copyright. Style is not protected by copyright. Feel is not protected by copyright. Chord changes are not protected by copyright. The only elements of a song that are protected by copyright are melody and lyric. That’s it. So, no. Wrong.
@@MixermanPublishes The melody and lyrics have been incorporated into the database, which is why they can be recreated so easily. Current AI models aren't actual brains, they don't learn. You're essentially making the same argument that by converting video to a different codec it's no longer covered by copyright.
@@o0shad0oo video isn’t the same as melodies. We all create melodies that are similar to other melodies. Video is more like a sound recording or a sample. The AI isn’t combining different samples. I can’t even wrap my head around this argument it’s so fundamentally flawed and has nothing to do with how music is written by humans or AI. Language AI spits out paragraphs that have never been written on the daily. It can write new lyrics too. It doesn’t just take lyrics from other sources and mix them up.
pardon me, but: preposterous. to argue that Udio is just "playing music in the presence of their AI", you would have to argue that the AI is a Person. the license to listen to streaming music is a personal license - it's granted to a person. It's a permissive license, which means the person is allowed to do with that music only the one thing they were permitted to do, which is to listen. nothing else. no license was given to (or even requested by) any business in the licenses you refer to. and they didn't stream the music, they copied it - they had all of it downloaded (which is expressly forbidden by the personal license you consent to on streaming services) before they could analyse it. I'm a (or was) a musician. I use Udio, and I love it! I'd be sad to see it go. But legally, they don't have a leg to stand on. All of that music was illegally downloaded, and there's no way you can argue a legal product comes out of that. Like most people debating this subject, you are way too focused on what people do with products like Udio - but where the problems begin is with the creation of the model in the first place. And yes, I understand you could say the same about large language models - they didn't license any of the training data for those either. the difference here, is there are basically 3 corporations holding all the rights to the content they stole, and those corporations have big money, and an army of lawyers. Udio should have secured the rights to their training data, and they chose not to. They are well and truly f_cked.
How do you know they download it? You can’t really download music without buying it. And if you’re in the room with me when I’m playing my Streaming service, that’s allowed and assumed to happen. I can even have a party and play my streaming service without breaking any terms of service. They could buy a streaming subscription specifically for their AI for purposes of training. As to audio securing permission in advance, under what mechanism? Major label music is likely only one source of the music they need to feed AI. The Majors aren’t going to win this one. You can believe what you like, but Udio couldn’t have dealt with these issues in advance. That needs to be dealt with by Congress. That’s why Title 17 exists. Udio and Suno aren’t fucked, and they are just the first to get into this game. They had to train their AI on copyrighted works because thats all that exists. Should they have sung to it? Even that could be technically deemed a performance. So, without any rules to the road, this was exactly the way it was going to go down. So, not fucked.
It's always been a gambit by Udio/Sunos. Their bet is that they will make WAY more money, with a perfected model, than they will ever (successfully) be sued for. They are, probably, correct. When the years of litigation are done, there'll be no way to "remove" all the copyrighted material (and the subsequent learning) from the model. Otherwise, you're absolutely correct.
@@nilespeshay1734 there is such a thing as a cease and desist order. if they violated that, they would be fined daily, I doubt they would be able to survive that.
@@RasmusSchultz none of that is going to happen. Guaranteed. The labels will lose their suit. You do realize, like, the totality of all Major Label music is just a tiny fraction of all copyrighted music out there. It would ultimately have to be a class action suit, and it's never going to be a winner, and even if it was, nobody other than lawyers would get squatola.
@@RasmusSchultz ++You may, very well, be right++... I just think that there is SO much money on the 'other side' of a perfected model that it negates any amount of money they can feasibly be fined or punished with. And I think enough people (i.e. potential investors) believe that that Udio (as an example) is playing with, figuratively, unlimited capital. (As an example, just +one+ of Udio's investor companies (Andreessen Horowitz) has access to 42 billion dollars.) When you look at the biggest fines that any private company has EVER been hit with (Bank of America, British Petroleum, etc) you're still within a range of money that would only be a few years worth of earnings... for a perfected model. And I don't see any world where Udio gets fined more than, let's say, a massive oil spill in the ocean.
Good to hear someone talk intelligently about this without screaming and cursing at AI.
Major labels suing themselves into extinction, now that sounds good
Thank you for making this.
You don't see art consumers barging into museums and cursing at art students copying the Masters.
Very interesting times ahead!
9:09 exactly! Copyright gets spun as protecting the creatives but any benefit to the creatives is entirely collateral damage to the publishers' bottom line.
we need more level headed real discourse like this. I'm impressed
Same can be said for every book author. Aren't they stealing from every author they've ever read?
Same for painters or artists.
These record labels are praying on people that don't understand how AI works or learns.
A brilliant explanation about the reality of the situation versus what the RIAA would like it to be.
The process of learning based on exposure to said media is not a copyright infringement - Otherwise every artist that listened to and learned to play Eruption by Van Halen (non commercially) is breaking copyright (they aren't and are covered in fair use). To learn how to create in the style of is a hallmark of every human artist today in every medium as part of the social construct of learning in and of itself as evidenced by asking who their greatest "influences" are. The sum of human achievement is marked by prior works and transformative application thereof.
We applaud such things when it's humans but demonize it when an AI learns to do the same because of biological "exceptionalism". We say "These things were used to 'train' an AI", but don't bat an eyelash when a human learns through example to do the same.
What we're talking about is the fundamental exercise of the act of learning overall through mimicry and exposure.
Something the entire human race depends on in order to build upon and create new works.
Just as a human that learned to play like a certain artist can also play that artists songs when asked, so too an AI is capable, just as well as creating new works in that style. In that the RIAA specifically cited outputs that they input to get such results, and by which they own the copyrights to, is akin to them asking a session musician to play a song the RIAA owns and immediately crying foul for the ability to do so.
Ask any capable guitarist to play the opening to Layla by Derek and the Dominos. Does the RIAA get to sue every guitarist for knowing how to do so because they've heard the song, and learned how to play it? No, because that would be absurd.
However, just as there would be limitations on what could be published for commercial purposes due to copyright, so too the AI would be subject to the same rules and uses regardless of the learning process. In which case, commercial use cases of non-transformative style transfer which duplicates discernable portions or whole works would require licensing as outputs.
This would fall in line with how we operate with human learning and performance output and would not jeopardize the very premise of learning overall.
The litmus test would therefore be:
Could a human learn to perform or produce a similar output through exposure to prior works? If the answer is "Yes" then the AI learning process is falling on the same premise but still bound to the same non-commercial output and/or licensing if said output were commercial use and were to be reproducing wholesale or in adequate portions to warrant licensing (sampling or otherwise).
The learning process itself is moot if a human can reproduce through exposure a style or output the same. It's the output and intended purpose which is judged the same as a human and whether licensing is needed for prior works or if transformative aspects have been applied.
If you asked an AI to reproduce ABBA and it did, after you gave it the lyrics and style... then yes. Licensing is needed if it is for commercial purposes. The same would be true for human created works. But this is very specific and does not cover "general style" - as evidenced by the industry trying to sue John Foggerty for sounding like himself (and CCR) in his solo career.
Thank you for your cogent arguments. This is how I view it too.
Anyone who is triggered by AI being fed copyrighted material isn’t considering the fact that you couldn’t feed it any material other than copyrighted material. Even if the music was public domain, the recording isn’t. So, what were they supposed to do? Sing to the AI?
Then that would be a Performance Rights issue. People also don’t realize (or consider) the Majors own a small sliver of all copyrighted music. There’s 100,000 records uploaded to streaming a day, all of it copyrighted, little of it by the Major Labels. I don’t need to feed my AI any Major Label music, and it would learn precisely the same things musically, because for every hit owned by a Major there are thousands of songs that do similar things. So the whole argument that feeding the AI copyrighted music is infringement is actually argument absurdium. The argument that you could come up with a way to pay everyone who influenced the AI is equally absurd, and doesn’t take into account that those who came before, should get more than those who built on to after. Prince was way more popular than Robert Johnson, but without Robert Johnson (and all the artists between) should arguably be paid more than for his contributions, than Prince (or their estates at this point).
Thanks for your post.
I have multiple problems with your arguments but... 'machine learning' doesn't LISTEN to music any more than a graphic equalizer LISTENS to music...
Imagine this... I go to the movie theater and I start recording the screen with my phone. When the usher, inevitably, taps me on the shoulder and says that I'm "stealing" the movie, +I+ say-- "My phone is just WATCHING the movie. What's the difference between ME watching the movie and my phone WATCHING the movie??"
"So, my phone REMEMBERS what it saw... so do I. I ALSO remember what I saw. So, what's the difference between me telling my best friend about the plot of the film and me re-playing the movie for them... from my phone?"
I'm utterly perplexed by the whole, "Computers are just doing what humans do" argument regarding 'AI'. The mechanism by which an action is performed can completely change the nature of that action. To believe otherwise is a descent into language games that are, insanely, far removed from reality...in my opinion.
no no, my sampler was just "listening" that day, I was *well* within my right to sample that song 🤪
The phone and the sampler reproduce what they record, and when that reproduction occurs, it is absolutely infringement of a particular performance. The AI assimilates everything that it hears, and then makes new music, which pulls from all that it’s heard, like influences, the same way that humans do. Perhaps you should make a song or two on Suno or Udio to understand better how it works.
If you make a song with Suno, it makes a song that is accessible, that pulls from themes and genres out of millions of possibilities and combines them, but does not pull specifically from anything in particular. This is precisely how people make music. I know it’s how I make music.
So, your argument that recording a performance is the same thing is rather weak and nonsensical.
My next video will demonstrate a track that I made with Suno, and the smart musical decisions that it made, none of which infringe on anything. It has to specifically steal a melody and a lyric for it to be infringement.
@@MixermanPublishes The truth is that neither you, nor I, have access to the code these sites are working with so it's impossible for you to say, with any certainty, HOW they operate. You're just regurgitating Udio's/Sunos' talking points. (And, you DO understand the incentive for these companies to obscure the truth, right?)
It diminishes anybody's argument when they speak so authoritatively about something that is impossible for them to know or verify... When their main source of information is the (possibly) offending party themself... who has all the incentive in the world to be dishonest.
I have used both services and read both services' TOS's. What alarmed me was Udio saying that users should be aware that Udio might spit out the +exact same song+ for multiple people on multiple occasions. That doesn't imply a creative, iterative process. It implies an algorithmic "re-combination" of collected pieces. It implies something closer to 'reproduction' than 'assimilation'. Pre-determined outcomes.
It indicates a system that has no "memory" to be aware of what it's done in the past... or even an awareness of what it's doing on another open, concurrent, session. These things are NOT analogous to how a human "assimilates" and creates.
In the same way that it's a felony to video-record a movie screen (without permission), it should be illegal to aim "machine learning" at copyrighted music. I don't really understand how people see a difference between the two.
@@nilespeshay1734 There are plenty of technical papers on the techniques.
@@makers_lab You're speaking in general and I'm talking specifically about two companies that seem to have mastered this procedure in a way that no company before them has been able to accomplish.
If "technical papers" could confirm just how Udio and Sunos are operating, there would be no need for a lawsuit.
The problem is, that it's not "learning" shit... It's noticing patterns and relating them to parts of a potential worded prompt. It's not learning why this melody sounds sad. Why this one sounds nostalgic. It's noticing that a song with similar melody had a hashtag "sad" in it's description when it was going through the training dataset.
Another thing is ... if I sample a song even using it transformatively, I am legally subjected to the possibility of being sued to oblivion. So if judge rules out that "learning is not stealing", I'd like to learn a and use melodies of famous songs in my medley and monetize it. Thank you.
And lastly ... there are damages. Huge ones. On sync licensing. Nobody will license Beatles for their background music when they can get close enough approximation from these services for free. That is a monetary hit. (And ofr us small musicians even bigger. Demand will drop to 0.)
I don’t know why certain chords make me feel a certain way either. I just learned which ones do, and now I use them for that purpose.
That's what learning is. Neural networks that power AIs are basically just a simulation of how neurons in the human brain work.
You can find people on UA-cam who make songs in style of X artist that are free and people still buy the actual artists stuff. People still fork out money for hand made stuff even though industrial products are cheaper.
Ironically I've actually found artists and bought their content from seeing AI version. Maybe you operate solely on what is cheap and convenient but the entire free market does not.
@@ragnarok7976 I intend to hand make music with the assistance of AI. It’s nowhere near where I need it yet. There is very little control. But once I can direct AI like it’s a human, then I can hand make music with AI as my all purpose, muti-talented assistant, and I will be able to make music far more efficiently, for expanded purposes.
Great video, this was my take on everything as well. It's great to hear someone talking about it in the same ways that I think.
Before a T800 tries to kill me I'm gonna make ot sing me a song.
Brilliant. 👏🏻 👏🏻 👏🏻
finally somoene who can articulate my view perfectly
Awesome explanation, great video, thanks
Never thought about it like that, you make a good point.
What is the best tool you can use to detect and measure presence of copyrighted material in an a new song? I’m familiar with MIPPIA. Are there even better ones? There needs to be such a tool to detect but most importantly provide copyright holder information. Should someone produce something that happens to contain copyrighted material, such a tool would make convenient to the producer the ability to pursue licensing. One would think that record labels would invest in creation of such a tool rather than in such a lawsuit. They would stand to profit from it. With distributors on board as well, they could apply such a tool to every work before distributing and then offer to the artist the information as to where to pursue licensing if needed.
If there’s a hit song that infringes, then there will be a suit. Many suits. As there almost always is for every hit song. You want to make a tool that can allow someone to find songs that aren’t hits, and worse yet are likely obscure, so that we can all sue each other for statutory damages? That would cause abuse and would clog up the courts so badly that you would force Congress to kill statutory damages, which is actually a critical component of Copyright protections, because it dissuades people from purposely infringing with impunity.
Melodic infringement is always largely debatable. So, I don’t know how an algorithm can pick out something that’s debatably infringement. Lyrical infringement requires outright stealing. Nothing else is copyrightable. Not style. Not production. Not arrangement. Not harmony (chord progressions).
What you suggest isn’t possible, and if it is, will cause the end of copyright as we know it.
@@MixermanPublishes My only thought was something to allow at least me to determine if a new song I’ve written is highly comparable to another song, meaning 80 to 100% similar in music or lyric or both.
@@jeffhulrichunless what you write is a hit song, you don’t have to worry about being sued for accidental infringement. You have to be given the opportunity to take an infringing work down before you can be sued for the $150,000 in statutory damages.
@@MixermanPublishes thanks for the insight. I really appreciate the video. I just discovered your channel yesterday on this topic and found it very useful. Subscribed.
Excellent points. Some I've been trying to communicate previously, and some very good new ones.
Thank you for making this video. This is no different than if one of us went to the Coca-cola Headquarters and demanded to see the original recipe because we think it still contains cocaine. Or demanded to know the original herbs and spices in KFC. These are requests for trades secrets and no one would take it seriously. So why do people suddenly think it's alright to demand to see the trade secrets of the Suno or Udio training data?
because people want retribution for the death of their dreams... people dream of living off of their art until they have to do so and they pull their hair out, but now they are mad that robots are taking those dream jobs away and flipping burgers seems like it was the more solid career choice all along except they're making automated McDonalds now.
Companies making their money predominantly from AI should be taxed high enough to make Democrats happy. Companies not relying on AI should be taxed low enough to make a Republican blush.
You've said it like on other I have heard before. Excellent video.
I have to disagree on the point AI learns the same way as humans do. We learn by listening, practicing by actually playing ot on a instrument or singing it, we 'feel' the song as we learn it.
It’s true that we feel and the AI doesn’t. But once you understand what tools to use to make people feel a certain way, it becomes a plug and play process. The difference is, the AI has to be told what makes us feel what way, but then, you actually are taught those things in college level arranging classes too, especially where it comes to scoring. All the composers for movie scoring all use the same phrases snd scales for movie music. There are some that are more inventive than others, but the techniques that are used to make us as viewers feel a certain way are all plug and play techniques. So, yes, it doesn’t have to practice scales, it doesn’t have to discover the magic of music, but if it understands what makes humans feel a certain way, it can plug and play exactly the way humans do, who make enormous volumes of music at a very high level.
There’s a reason why movies scorers never sue each other. It’s pointless. Yet, there are movies that have ridiculously similar sounding scores to others (especially in the 90s and aughts), with only slight variations.
When I say that AI learns how to make music exactly as humans do, I mean the AI learns the patterns that make effective music that cause a certain physiological reaction in humans. It’s clear that these generative AIs have been taught all of this, because if you give it a sad lyric, it generally makes the music sad to match. The music it spits out is in alignment with the lyric, or the concept, that that the AI is presented by the user. That means it’s making musical decisions, and you may not like that term, but I wouldn’t know what else to call it. I’ll be posting a video demonstrating the good decisions an AI makes based on a lyric that it was provided. I know humans who don’t use some of the good techniques the AI does.
The problem is most people are mediocre at music making, and the AI is already superior to it than 80% of them.
@@MixermanPublishes beautiful response and I appreciate it.
Dudes a lawyer now
lol! Aw.
I couldn't agree more! I've been learning and being INFLUENCED by popular music forever. AI should be afforded the same leeway.
Yes! I so agree!
The genie is out 🧞
Don't get me wrong I think the majors are pure EVIL....But honestly, how many people are NOT going to publish these songs? Putting them onto UA-cam is a form of publishing. Someone makes a cool song using one of these mentioned services (or others) who is not tempted to share it on a mass level?
I don’t think you grapes what I was saying.
Excellent commentary
If these major labels win, I'm gonna send them all a CD with my original music. And then I'll sue them for listening to my music without having the rights to my music. How much was that per song?
Sorry to say this but you're 100% right & These major label are gonna lose this lawsuit because technically these A.i. platforms or tools, instruments that are no different from the MPC samplers technically its just a more advanced version. Put it this way, YOU CANT SUE AKAI FOR MAKING A SAMPLER BUT YOU CAN SUE THE ARTIST & PRODUCER WHO SAMPLE MATERIAL & PUBLISH IT WITHOUT CLEARING SAID SAMPLE. UDIO & SUNO ARE NO DIFFERENT THAN AKAI MPC's if you see what I am saying here. Yeah SUNO & UDIO are making thousands of dollars but so is AKAI for doing the same thing which is Sampling!!!
In the age of companies copying my work for free while I don’t even own what I buy, this is not a good title
What company has copied YOUR work? Streaming companies are what killed people’s ability to make money with music. This is already done. Our only chance is to become good with AI at this point. We can either accept this, or hate our lives.
I make 20 percent of what I made on a per mix basis in the aughts. I make half what I made per mix in 1995. What makes it worth it for me is I can mix faster now, I can mix on my own time schedule, and I never have to leave my house. I can also sell books, which are far more lucrative for me now that I can publishes them myself, That doesn’t make any of it a less of a bitter pill to swallow. This is the reality. There’s nothing you or I or anyone else can do about it, because we don’t have a government that protects our interests, but rather the interests of large corporations. And now Americans are poised to end Democracy, which will only make that worse. We can either play in the sandbox we have, or lament the way it should be. I recommend the former. With all the problems coming down the pike, we need to protect our own interests, at this point.
I stand corrected: A correct and brilliant post by mixerman. :-)) Gratulations, you are completely right. BTW: EVERY person learning to make music uses not-licensed listening to other persons music ...
You are correcting yourself? Did you comment half way through or something?
what? no they don't. not unless you're pirating the music you listen to? if you're listening to music on a streaming service, that music was licensed for you to listen to. you're using the music the way it was licensed for you to use. simple as that.
@@RasmusSchultz so I can’t stream music while you’re in the room? You have a strange understanding of how streaming works.
Besides, Suno can buy a subscription specifically for their AI. This argument of yours makes no sense whatsoever.
@@marc.levinson that’s a big assumption that it’s recorded.
@@MixermanPublishes you're still going on about things that have nothing to do with what they did. AI is not a person in a room. The model is a product derived from copyright recordings and then sold. You're not allowed to do that. The end.
Well said!
"played in the presence of their AI" spoken as if you are talking about an intelligence of some sort. This will boil down to that argument. At this point and time, is machine learning the process of teaching an 'intelligence' or will this be deemed the process of a copy writing machine that takes in information and spits out an auto correct/complete.
The answer will likely fall on the side of, what happens when you don't feed an intelligence other songs. Can it produce from nothing? The question when it comes to a computer, is simple, it can't do anything without being fed full completed content.
I think it will boil down to the argument that it’s not outputting copyrighted works. It’s outputting original works, whether people like it or not.
Especially with original lyrics. An AI created song that uses original lyrics is not 'infringing'. You can't 'copyright' popular chord progressions. You can't copyright 'genres' of music. You can't copyright a 'feel'.
You can only copyright hard, unambiguous content. If these companies are not COPYING these works, or distributing them, or broadcasting them, then they're NOT running afoul of a copyright. That's silly. That's like trying to say I'm not allowed to learn how to play a Stevie Ray Vaughn song without paying him.
The judge in this case will ask, "Did you copy, distribute, or broadcast any copywritten material when training your AI models"?
Then ask the plaintiff: "Do you have any proof that these companies copied, distributed, or broadcast any intellectual property owned by you?
Because the answer to BOTH of those questions is no......
Yes. But people seem to think that you can copyright all those things, and they are sadly mistaken.
If you're using a raw material to create an end product, surely you should pay for the raw materials? Basic business right?
@@BrockBarr but who am I paying? The Majors only own a small sliver of all copyrighted music. Why should they collect the money for raw materials that come from everyone in music history? And if everyone in music history is to be paid, what mechanism allows for that? And shouldn’t the estates of the music pioneers get more? And who should determine that? And if I’m training my AI, how much should I pay for each track? So each artist is going to get one payment per song for pennies and then the AI is trained, and then that AI engine is licensed out to AI developers? You just cost the developers billions to pay each person less than the price of a sandwich? How does that make any sense?
It’s all well and good to make a statement that a business should pay for raw materials, but there isn’t a market in existence for these materials. They’re essentially free to anyone with a streaming subscription. And you need to come up with a way to pay everyone. And you need to explain how that going to amount to anything since and AI need only evaluate any given song one time.
Why should companies be left to being the only ones to train A.I on anything? A.I should be accessible to all the people. It's merely information that can learn and improve. I say impeding that is infringing on Freedom of Speech because, it says what we can and can't do with information as only information. Imagine if the internet was that way and only corporations alone had access to and ran it. I say it's time for corporations to collapse. If we can all do it there's no power nor monopoly focused in one place that can be corrupted. A.I is too powerful to only be in the hands of the few.
Very well said
Spot on!
They would also sue the teachers for smuggling knowledge 😀
Subscribed
I prompted Ecosia chat "The Major Labels are suing Ai companies Udio and Suno for copyright infringement, but are they really infringing?" and this is what it said: "I don't have specific information on the legal case you mentioned involving Udio and Suno. Copyright infringement cases can be complex and often involve detailed analysis of the specific circumstances. It's best to refer to official statements from the parties involved or legal experts for accurate information on the situation. If you're interested in learning more about copyright laws and how they apply in cases like this, I can provide some general information on the topic. Would you like to know more about copyright laws?"
So, it’s very possible the labels have managed to generate some specific instances of infringement. But do we really believe for one second, that even if the labels were able to generate 10 instances of legitimate, clear and obvious infringement, that they are wasting their time suing for $1.5 million dollars? That’s their sushi budget.
And do we believe for one second that Udio and Suno, (both companies now with over $125 million in seed money each) wouldn’t just pay that off and make it go away if they could, and teach their AI how to better not infringe? I don’t.
Like I said, this is about forcing the AI companies to reveal their training methods through the Discovery process.
It’s not about the money. It’s about the money.
@@MixermanPublishes I thought "It’s not about the money. It’s about the money." was a Bill Graham quote but apparently it is actually from H.L. Mencken - ‘When somebody says it’s not about the money, it’s about the money.’ Either way it rings true!
I thought it might be amusing to see what an AI chatbot though but er not so much. LOL
@@alphaomega6062 yeah, I asked it too. Lol.
Today these models aren't great at reaching outside their datasets, for example making something extremely novel. They tend to stick to the bounds of what they've consumed. Commentators are latching pretty hard to that. To me, humans aren't much different. Most novices for example do nothing but mimic. I think this limitation in AI today will be quickly behind us, and then the argument in this video will be more widely palatable.
I'm not arguing with the general premise that the major labels are just trying to get some more money out of the AI companies and this is not something that is going to work for artists.
And that the actual solutions that works for creators and humanity is probably more complicated and requires thinking outside of the current system.
That being said, AI doesn't actually "listen" or "learn", these are analogies that are used. A machine can't "listen" or "learn" in it's truest self as the words exist to describe how humans operate.
Additionally, the current norms we have, allowing humans to listen to music freely, to learn and practice established works, etc. are rooted in our physiological limitations and doesn't apply to AI.
The whole why can a human listen to music but not AI is silly. Why can a human go to the beach undisturbed but when a bear shows up they get trapped by animal control.
Humans and Animals don't have the same rights. Machines and humans don't have the same rights either. I'm not saying specifically if "listening to music" should or shouldn't be allowed for a machine freely. All I'm saying that it's not automatically OK because it's OK for humans to do.
@@FicoosBangaly lots of people get hung up with the terms learn, listen, and decisions, (video on “decisions” next week). But not one person has actually come up with words to describe what the AI is doing that seems more accurate. One person out of probably a hundred gave me an alternative, and it was weak.
People get triggered when I use human terms for a machine, but based on what AI does, and how it works, I can’t come up with anything more accurate. So, to me? If the definitions of those words only relate to humans, then the definitions need to change. And I’m not going to use klunky language to describe what I can use one word for, and that conveys my message in an understandable way, just to satisfy people who I find are being rather pedantic about it. Y’all want me to change my language to suit your discomforts, without even providing an alternative that’s accurate and appropriate. I want to redefine the language, since it’s dynamic and ever changing, and that’s what I shall do. Since I’m the one conveying the message, I’m the one who decides how I do it. You’re welcome to use your own terminology.
@@MixermanPublishes You can use analogies when they are convenient. But you need to remember that they are analogies. People talk about computer viruses and it's fine. They are "viruses" in the way that they are a piece of parasitic code in a host program corrupting it and causing it to make the computer "sick". It is *the* name for it. But if you tell me that because it is a virus it has a certain amount of genetic material, I'll tell you that you are probably getting confused.
Additionally, there are words that have a slightly different meaning when used in different contexts. For example the word theory which means different things in the common speech and in a scientific context.
There are words and exact descriptions for what different machine learning implementations are doing when they ingest input (i.e. "trained"). Analogies are fine and helpful and people can continue using them as long as they understand that they are "analogies" and that the concept of an algorithm "learning" is different from a person learning in the same way as when an evolutionary biologist saying that a gene "wants" to replicate.
But, I thought the problem was these AI companies are Publishing...and getting the Ad Revenue?
Honestly asking. They're making Money no?
They aren’t publishing anything. Users are publishing original works spot out by the AI.
I'm really not fond of the now burgeoning intellectual property industry, but I have to disagree with you. It's not the song created by the AI that's the infringement being sued over, it's the pattern of the song embedded in the AI that's providing value being marketed to users of the AI.
Patterns aren’t protected by copyright. Style is not protected by copyright. Feel is not protected by copyright. Chord changes are not protected by copyright. The only elements of a song that are protected by copyright are melody and lyric. That’s it. So, no. Wrong.
@@MixermanPublishes The melody and lyrics have been incorporated into the database, which is why they can be recreated so easily. Current AI models aren't actual brains, they don't learn. You're essentially making the same argument that by converting video to a different codec it's no longer covered by copyright.
@@o0shad0oo video isn’t the same as melodies. We all create melodies that are similar to other melodies. Video is more like a sound recording or a sample. The AI isn’t combining different samples. I can’t even wrap my head around this argument it’s so fundamentally flawed and has nothing to do with how music is written by humans or AI.
Language AI spits out paragraphs that have never been written on the daily. It can write new lyrics too. It doesn’t just take lyrics from other sources and mix them up.
pardon me, but: preposterous.
to argue that Udio is just "playing music in the presence of their AI", you would have to argue that the AI is a Person.
the license to listen to streaming music is a personal license - it's granted to a person. It's a permissive license, which means the person is allowed to do with that music only the one thing they were permitted to do, which is to listen. nothing else. no license was given to (or even requested by) any business in the licenses you refer to.
and they didn't stream the music, they copied it - they had all of it downloaded (which is expressly forbidden by the personal license you consent to on streaming services) before they could analyse it.
I'm a (or was) a musician. I use Udio, and I love it! I'd be sad to see it go.
But legally, they don't have a leg to stand on. All of that music was illegally downloaded, and there's no way you can argue a legal product comes out of that.
Like most people debating this subject, you are way too focused on what people do with products like Udio - but where the problems begin is with the creation of the model in the first place.
And yes, I understand you could say the same about large language models - they didn't license any of the training data for those either. the difference here, is there are basically 3 corporations holding all the rights to the content they stole, and those corporations have big money, and an army of lawyers.
Udio should have secured the rights to their training data, and they chose not to. They are well and truly f_cked.
How do you know they download it? You can’t really download music without buying it. And if you’re in the room with me when I’m playing my Streaming service, that’s allowed and assumed to happen. I can even have a party and play my streaming service without breaking any terms of service. They could buy a streaming subscription specifically for their AI for purposes of training.
As to audio securing permission in advance, under what mechanism? Major label music is likely only one source of the music they need to feed AI.
The Majors aren’t going to win this one. You can believe what you like, but Udio couldn’t have dealt with these issues in advance. That needs to be dealt with by Congress. That’s why Title 17 exists.
Udio and Suno aren’t fucked, and they are just the first to get into this game. They had to train their AI on copyrighted works because thats all that exists. Should they have sung to it? Even that could be technically deemed a performance. So, without any rules to the road, this was exactly the way it was going to go down. So, not fucked.
It's always been a gambit by Udio/Sunos. Their bet is that they will make WAY more money, with a perfected model, than they will ever (successfully) be sued for.
They are, probably, correct.
When the years of litigation are done, there'll be no way to "remove" all the copyrighted material (and the subsequent learning) from the model.
Otherwise, you're absolutely correct.
@@nilespeshay1734 there is such a thing as a cease and desist order. if they violated that, they would be fined daily, I doubt they would be able to survive that.
@@RasmusSchultz none of that is going to happen. Guaranteed. The labels will lose their suit.
You do realize, like, the totality of all Major Label music is just a tiny fraction of all copyrighted music out there. It would ultimately have to be a class action suit, and it's never going to be a winner, and even if it was, nobody other than lawyers would get squatola.
@@RasmusSchultz ++You may, very well, be right++... I just think that there is SO much money on the 'other side' of a perfected model that it negates any amount of money they can feasibly be fined or punished with.
And I think enough people (i.e. potential investors) believe that that Udio (as an example) is playing with, figuratively, unlimited capital.
(As an example, just +one+ of Udio's investor companies (Andreessen Horowitz) has access to 42 billion dollars.)
When you look at the biggest fines that any private company has EVER been hit with (Bank of America, British Petroleum, etc) you're still within a range of money that would only be a few years worth of earnings... for a perfected model.
And I don't see any world where Udio gets fined more than, let's say, a massive oil spill in the ocean.