To my knowledge, training an AI of this sort does not involve storing copies the data. The program learns from the data (music/songs in this case) i.e., it breaks the data down into noise (not noise as in sound, but noise as in data) and processes it. The training phase is completely separate from the generation (production) phase. Once the training is done, the program can use the patterns in the noise (again, noise as in data) to generate slightly different versions of that original data, modified based on user prompts. The programs in question are not open source so it's hard for me to speak on this specific instance, but I have worked with similar generation models and at no point do they store the original training data in its pure form. It remembers patterns in the data/noise in a way similar to how human brains learn and remember data, often known as an engram.
Very good post. To me worrying about these kinds of technical distinctions is silly. Congress needs to address using copyrighted material to train or produce AI models regardless of whether copies are made. If Congress doesn’t, then unfortunately judges are forced to worry about these distinctions.
@@LawLaughsMusic Noted. I was responding to the fact that you mentioned storing the data in the first part of your video. A lot of people make the same misconception about storing the raw data used to train, I just thought I'll point out the technical bit. I'm not sure about the legal aspects, I'm just a technician/researcher.
At some point would audio files have to be copied? Even if very early on in the training process? Everyone seems to assume this. I used to do a bit of coding, but this is beyond me.
@@LawLaughsMusic Converted is a better term. The training songs would be converted to raw data, if that helps clarify things. Then the raw data is fed to the application and it 'learns' the patterns in the data. It then has/stores a memory of the patterns, not the song or even the raw data.
I love Rick Beato. He's from Rochester, that's cool because he's probably eaten a garbage plate or two. I like mine hamburger, potato, mac, and beans, warm the beans, no onion and Jimmy knows that I know that the beans are extra and he also knows I'm gonna buy a bumper sticker and a couple double extra large tshirts so most times he don't charge me for the beans unless there's custies behind me in line and he'll say loud enough for everyone to hear it and it's a 50 50 shot that he let's me slide on the sweet legumes. And hey most times he knows I'm getting the double extra large t shirts before I do because he's almost always out of the ones that fit a non competitive eater. Sorry. I know this video covered pertinent information concerning serious issues that will have life altering affects on all of the people that entertain us with their creativeness. So, I apologize for the game of 3 degrees of nick tahous
If the record companies win independent artists will be replaced on a scale never seen before. And if training is considered to be protected by copyright , and an AI is considered to be making copies, then the output is _also_ protected by copyright, at which point the record companies _will_ go after the generated content. This is their ultimate goal, and why I’m a proponent for everyone having access to the technology despite the moral gray areas. Because the alternative is so much worse.
Courts that have dealt with this in the AI art space (and it’s just a handful of cases in the early stages of litigation) have so far assumed that copyright infringement in the training phase doesn’t automatically make the output infringing. But, anyway, there are so many big policy considerations here and possible unintended consequences that Congress needs to act and not leave this to judges to base the outcomes on distinctions with older technologies.
I think the biggest aspect of this case that Suno and Udio need to point out is that their songs are being tangibly stolen. The difference between a machine and a human mind is that a machine stores material data that is owned and managed BY human persons, whereas human beings are free to have their own memories and ideas. I wonder if they can argue that the conversion of their songs into a new type of data as the AI analyzes it is a form of “recording”. This is especially true in my view because they have been able to nearly reproduce their own records using the AI, which presumably has its own data stored after listening to their songs. Since digital recording is defined as the registration of sound on a device then converted into digital data, I think they can make the case that this is a form of illicit recording. Also, isn’t it illegal to download copyrighted music in the first place, even if the recording is later destroyed?
Thank you. Good post. It’s pretty clear if there is no fair use defense, there was infringement. And I think fair use is tough. That said, they may have initially obtained copies legally (although may not have). But even if they did, if a file is copied at any point during the training process, a court would likely find that infringing unless there is fair use.
Copyright law is broken and stands in the way of innovation and competition rather than protecting and facilitating it. I sincerely hope that the record companies do not manage to monopolize AI regardless of other outcomes.
Agreed, this laissez-faire approach to business created ripe ground for corporate endeavors to flourish. Less regulations by the people, for the people translates to more regulations by corporations, for corporations.
@@GeneralFailureMajorErrorThe data is in a tangible format, not simply an idea within human consciousness. It becomes material property. You are fully within your right to remember a painting in your head. You’re not allowed to take a photo of it on your phone.
What is the difference between a person listening to a song and learning from it, and a computer listening to a song and learning from it? You cant copyright learning from music
Good post. This is a tough area that needs to be dealt with thoughtfully, ideally by Congress if it can function. On the one hand, there's obviously nothing wrong with a person listening to a song to learn how to create their own competing music. But on the other hand, computers digesting songs to learn how to create competing music on a very wide scale seems different. But good point.
If I learn to play an song on an instrument, is that also copying it? Or only if I bring out the same exact song? And use what I learned to play only as knowledge on how to play, and create other songs with the knowledge on how to play some songs
Good post. So learning a song in order to be able to create your own music is obviously fine. The only issue would be if you then create music that is substantially similar to someone else's song. Good point.
Interesting, thanks for the info. What I'm wondering is if I use AI to help me finish one of my own songs, can they legally use parts of my music to generate AI music for others? And if I used AI to help me write a song and that song ends up making a lot of money, does the AI platform have rights to demand royalties from me?
Very interesting question the status of works that combine human created elements and AI created elements is either unclear or very fact specific. The human created parts should not be an issue. They should be eligible for copyright. The question is whether the AI created parts are part of the copyright or not.
They should loose, and I think they will. AI is not a database of music. Compensation also doesn't work, and it's not justified if something isn't very close to some original.
I think there are tough questions here and that it’s best if it’s not a judge having to deal with a completely new technology with old legal analogies.
Thanks for simplifying this for a number of us who are busy "making" the music, who need advice; Wow! this can be extremely time-consuming! Glad YOU"RE there to help us figure it all out! LyndaFayeSoriginals
To my knowledge, training an AI of this sort does not involve storing copies the data. The program learns from the data (music/songs in this case) i.e., it breaks the data down into noise (not noise as in sound, but noise as in data) and processes it.
The training phase is completely separate from the generation (production) phase. Once the training is done, the program can use the patterns in the noise (again, noise as in data) to generate slightly different versions of that original data, modified based on user prompts.
The programs in question are not open source so it's hard for me to speak on this specific instance, but I have worked with similar generation models and at no point do they store the original training data in its pure form. It remembers patterns in the data/noise in a way similar to how human brains learn and remember data, often known as an engram.
correction: i meant copies of the data in the first line.
Very good post. To me worrying about these kinds of technical distinctions is silly. Congress needs to address using copyrighted material to train or produce AI models regardless of whether copies are made. If Congress doesn’t, then unfortunately judges are forced to worry about these distinctions.
@@LawLaughsMusic Noted. I was responding to the fact that you mentioned storing the data in the first part of your video. A lot of people make the same misconception about storing the raw data used to train, I just thought I'll point out the technical bit. I'm not sure about the legal aspects, I'm just a technician/researcher.
At some point would audio files have to be copied? Even if very early on in the training process? Everyone seems to assume this. I used to do a bit of coding, but this is beyond me.
@@LawLaughsMusic Converted is a better term. The training songs would be converted to raw data, if that helps clarify things. Then the raw data is fed to the application and it 'learns' the patterns in the data. It then has/stores a memory of the patterns, not the song or even the raw data.
imagine that, the 3 big record companies who have been ripping off artists for years feeling violated.
You don’t think they’re primarily concerned about the musicians??? They’re so nice.
@@LawLaughsMusic if it was then they wouldn’t be interested in another ai company for profit.
@@LawLaughsMusic HAHAHA!
I love Rick Beato. He's from Rochester, that's cool because he's probably eaten a garbage plate or two.
I like mine hamburger, potato, mac, and beans, warm the beans, no onion and Jimmy knows that I know that the beans are extra and he also knows I'm gonna buy a bumper sticker and a couple double extra large tshirts so most times he don't charge me for the beans unless there's custies behind me in line and he'll say loud enough for everyone to hear it and it's a 50 50 shot that he let's me slide on the sweet legumes. And hey most times he knows I'm getting the double extra large t shirts before I do because he's almost always out of the ones that fit a non competitive eater.
Sorry. I know this video covered pertinent information concerning serious issues that will have life altering affects on all of the people that entertain us with their creativeness. So, I apologize for the game of 3 degrees of nick tahous
So if a person says to another make up a riff that sounds a little like Van Halen…….’🤡🥸🤣
Good point, obviously fine unless it’s too close to a specific riff
If the record companies win independent artists will be replaced on a scale never seen before. And if training is considered to be protected by copyright , and an AI is considered to be making copies, then the output is _also_ protected by copyright, at which point the record companies _will_ go after the generated content.
This is their ultimate goal, and why I’m a proponent for everyone having access to the technology despite the moral gray areas. Because the alternative is so much worse.
Courts that have dealt with this in the AI art space (and it’s just a handful of cases in the early stages of litigation) have so far assumed that copyright infringement in the training phase doesn’t automatically make the output infringing. But, anyway, there are so many big policy considerations here and possible unintended consequences that Congress needs to act and not leave this to judges to base the outcomes on distinctions with older technologies.
Interesting discussion. I like your channel and hope you will continue to discuss this issue in greater depth.
Thank you, I appreciate it. It’s always a question for me how deep to go.
I think the biggest aspect of this case that Suno and Udio need to point out is that their songs are being tangibly stolen. The difference between a machine and a human mind is that a machine stores material data that is owned and managed BY human persons, whereas human beings are free to have their own memories and ideas.
I wonder if they can argue that the conversion of their songs into a new type of data as the AI analyzes it is a form of “recording”. This is especially true in my view because they have been able to nearly reproduce their own records using the AI, which presumably has its own data stored after listening to their songs. Since digital recording is defined as the registration of sound on a device then converted into digital data, I think they can make the case that this is a form of illicit recording.
Also, isn’t it illegal to download copyrighted music in the first place, even if the recording is later destroyed?
Thank you. Good post. It’s pretty clear if there is no fair use defense, there was infringement. And I think fair use is tough. That said, they may have initially obtained copies legally (although may not have). But even if they did, if a file is copied at any point during the training process, a court would likely find that infringing unless there is fair use.
Copyright law is broken and stands in the way of innovation and competition rather than protecting and facilitating it. I sincerely hope that the record companies do not manage to monopolize AI regardless of other outcomes.
Agreed, this laissez-faire approach to business created ripe ground for corporate endeavors to flourish.
Less regulations by the people, for the people translates to more regulations by corporations, for corporations.
Definitely agree. Some type of compulsory license scheme is probably a good idea like in the case of cover songs.
Is listening to (and 'save') the music in our brain also a copyright infringement?
Luckily not!
@@LawLaughsMusic Why should a machine that is listening and learning from music be a copyright infringement? What is the difference?
@@GeneralFailureMajorErrorThe data is in a tangible format, not simply an idea within human consciousness. It becomes material property. You are fully within your right to remember a painting in your head. You’re not allowed to take a photo of it on your phone.
What is the difference between a person listening to a song and learning from it, and a computer listening to a song and learning from it?
You cant copyright learning from music
Good post. This is a tough area that needs to be dealt with thoughtfully, ideally by Congress if it can function. On the one hand, there's obviously nothing wrong with a person listening to a song to learn how to create their own competing music. But on the other hand, computers digesting songs to learn how to create competing music on a very wide scale seems different. But good point.
If I learn to play an song on an instrument, is that also copying it? Or only if I bring out the same exact song?
And use what I learned to play only as knowledge on how to play, and create other songs with the knowledge on how to play some songs
Good post. So learning a song in order to be able to create your own music is obviously fine. The only issue would be if you then create music that is substantially similar to someone else's song. Good point.
Interesting, thanks for the info.
What I'm wondering is if I use AI to help me finish one of my own songs, can they legally use parts of my music to generate AI music for others?
And if I used AI to help me write a song and that song ends up making a lot of money, does the AI platform have rights to demand royalties from me?
Very interesting question the status of works that combine human created elements and AI created elements is either unclear or very fact specific. The human created parts should not be an issue. They should be eligible for copyright. The question is whether the AI created parts are part of the copyright or not.
Great content, thanks for your insights!
Thank you very much for the support. It’s much appreciated.
They should loose, and I think they will. AI is not a database of music. Compensation also doesn't work, and it's not justified if something isn't very close to some original.
I think there are tough questions here and that it’s best if it’s not a judge having to deal with a completely new technology with old legal analogies.
Thanks for simplifying this for a number of us who are busy "making" the music, who need advice; Wow! this can be extremely time-consuming! Glad YOU"RE there to help us figure it all out!
LyndaFayeSoriginals
Thank you very much. I appreciate the support, and I’m glad I can help even in a little way.