I think you misunderstand something. The same thing many AI enthusiasts and critics often choose to not understand. Regenerative AIs aren‘t just born from plain code and they don’t just imitate. They use a ton of data as reference points. It’s literally in the name of the technology.
You could claim „well maybe they used different voices and mixed them together“ but that is highly unlikely, given how much of a wild west approach most regenerative AI services have. it‘s more likely they used protected property here in a way it was not intended to be used. In which case SJ does indeed have a legal case here.
They use a ton of data as reference points. It’s literally in the name of the technology.
Reference is the wrong word.
They learn the patterns that exist in data and are able to predict future patterns.
They don’t actually reference the source material during generation (barring over itting which can happen and is roughly akin to a human memorizing something and reproducing it).
Weather or not the copyrighted data shows up in the final model is utterly irrelevant though. It is illegal to use copyrighted material period outside of fair use, and this is most certainly not. This is civil law, not criminal, the standard is more likely than not rather than beyond a reasonable doubt. If a company cannot provide reasonable evidence that they created the model entirely with material they own the rights to use for that purpose, than it is a violation of the law.
Math isn’t a person, doesn’t learn in anything approaching the same method beyond some unrelated terminology, and has none of the legal rights that we afford to people. If it did, than this would be by definition a kidnapping and child abuse case not a copyright case.
It is illegal to use copyrighted material period outside of fair use, and this is most certainly not.
Yeah it is. Even assuming fair use applied, fair use is largely a question of how much a work is transformed and (a billion images) -> AI model is just about the most transformative use case out there.
And this assumes this matters when they’re literally not copying the original work (barring over fitting). It’s a public internet download. The “copy” is made by Facebook or whoever you uploaded the image to.
The model doesn’t contain the original artwork or parts of it. Stable diffusion literally has one byte per image of training data.
I never understood why so many from the more techbro political alignment find this argument so convincing.
It doesn’t really matter whether the original data is present in the model or if it was reduced to such an abstract form that we cannot find it anymore. The model only can exist because of the original data being used to make it, and it was used without proper license. It doesn’t matter how effective nor how lossy your compression is, mere compression is not transformation and does not wash away copyright.
The argument that it is in some way transformative is more relevant. But it’s also got a pretty heavy snort of “thinking like a cop” in it, fundamentally. Yes, the law protects transformative works, so if we only care what the written rules of the law says, then if we can demonstrate that what the AI does is transformative, the copyright issues go away. This isn’t a slam dunk argument that there’s nothing wrong with what an AI does even if we grant it is transformative. It may also simply be proving that the copyright law we have fails to protect artists in the new era of AI.
In a truly ideal world, we wouldn’t have copyright. At all. All these things would be available and offered freely to everyone. All works would be public domain. And artists who contributed to the useful arts and sciences would be well-fed, happy, and thriving. But we don’t live in that ideal world, so instead we have copyright law. The alternative is that artists cannot earn a living on their works.
It doesn’t really matter whether the original data is present in the model
Yeah it does. One of the arguments people make is that AI models are just a form of compression, and as a result distributing the model is akin to distributing all the component parts. This fact invalidates that argument.
This isn’t a slam dunk argument that there’s nothing wrong with what an AI does even if we grant it is transformative. It may also simply be proving that the copyright law we have fails to protect artists in the new era of AI.
If we change the law to make it illegal it’s illegal.
The number of bytes per image doesn’t necessarily mean there’s no copying of the original data. There are examples of some images being “compressed” (lossily) by Stable Diffusion; in that case the images were specifically sought out, but I think it does show that overfitting is an issue, even if the model is small enough to ensure it doesn’t overfit for every image.
It’s illegal to reproduce copyrighted material*. That includes changing the format as well as things which fall under “derivative” works, but not creating a new work in the style of someone else’s (unless it falls under the derivative definition). Many voice impersonators exist and the way you impersonate a voice is to listen to (usually) recordings of that person and practice producing the same sounds that they use for common phonemes (as well as vocal tract shape and larynx positioning to alter the vocal pitch production and overtones which represent vowel shapes). ML does, effectively the same thing without requiring a human to do the listening and practicing.
That said, I think this type of use should be strictly prohibited. In fact, I think it should have severe criminal penalties for any specific voice, not just celebrities. Having the ability to simulate accurate, regional-sounded voices is extremely valuable in the general sense, but imitating or mimicking a specific person’s voice without their explicit consent and/or direction has very few, if any, legitimate uses.
* I didn’t think that voice mimicking would count as valid for any law, but Google tells me of the “Right of Publicity” and there is (again according to Google) case law involving Ford and Bette Midler. So while it’s not a copyright violation to reproduce a voice, it may still run afoul of some laws.
Again: how is that different from an imitation? What exactly differentiates a human watching a movie to imitate a voice from a machine doing the same thing?
And that is, what you misunderstand. AI is not magic, it’s computation. Nothing more. In no other context would it even matter, whether the source data was intended for the use case, if no infringement is being committed by the end product.
I think you misunderstand something. The same thing many AI enthusiasts and critics often choose to not understand. Regenerative AIs aren‘t just born from plain code and they don’t just imitate. They use a ton of data as reference points. It’s literally in the name of the technology.
You could claim „well maybe they used different voices and mixed them together“ but that is highly unlikely, given how much of a wild west approach most regenerative AI services have. it‘s more likely they used protected property here in a way it was not intended to be used. In which case SJ does indeed have a legal case here.
Reference is the wrong word.
They learn the patterns that exist in data and are able to predict future patterns.
They don’t actually reference the source material during generation (barring over itting which can happen and is roughly akin to a human memorizing something and reproducing it).
Weather or not the copyrighted data shows up in the final model is utterly irrelevant though. It is illegal to use copyrighted material period outside of fair use, and this is most certainly not. This is civil law, not criminal, the standard is more likely than not rather than beyond a reasonable doubt. If a company cannot provide reasonable evidence that they created the model entirely with material they own the rights to use for that purpose, than it is a violation of the law.
Math isn’t a person, doesn’t learn in anything approaching the same method beyond some unrelated terminology, and has none of the legal rights that we afford to people. If it did, than this would be by definition a kidnapping and child abuse case not a copyright case.
Yeah it is. Even assuming fair use applied, fair use is largely a question of how much a work is transformed and (a billion images) -> AI model is just about the most transformative use case out there.
And this assumes this matters when they’re literally not copying the original work (barring over fitting). It’s a public internet download. The “copy” is made by Facebook or whoever you uploaded the image to.
The model doesn’t contain the original artwork or parts of it. Stable diffusion literally has one byte per image of training data.
I never understood why so many from the more techbro political alignment find this argument so convincing.
It doesn’t really matter whether the original data is present in the model or if it was reduced to such an abstract form that we cannot find it anymore. The model only can exist because of the original data being used to make it, and it was used without proper license. It doesn’t matter how effective nor how lossy your compression is, mere compression is not transformation and does not wash away copyright.
The argument that it is in some way transformative is more relevant. But it’s also got a pretty heavy snort of “thinking like a cop” in it, fundamentally. Yes, the law protects transformative works, so if we only care what the written rules of the law says, then if we can demonstrate that what the AI does is transformative, the copyright issues go away. This isn’t a slam dunk argument that there’s nothing wrong with what an AI does even if we grant it is transformative. It may also simply be proving that the copyright law we have fails to protect artists in the new era of AI.
In a truly ideal world, we wouldn’t have copyright. At all. All these things would be available and offered freely to everyone. All works would be public domain. And artists who contributed to the useful arts and sciences would be well-fed, happy, and thriving. But we don’t live in that ideal world, so instead we have copyright law. The alternative is that artists cannot earn a living on their works.
Yeah it does. One of the arguments people make is that AI models are just a form of compression, and as a result distributing the model is akin to distributing all the component parts. This fact invalidates that argument.
If we change the law to make it illegal it’s illegal.
deleted by creator
The number of bytes per image doesn’t necessarily mean there’s no copying of the original data. There are examples of some images being “compressed” (lossily) by Stable Diffusion; in that case the images were specifically sought out, but I think it does show that overfitting is an issue, even if the model is small enough to ensure it doesn’t overfit for every image.
Over fitting is an issue for the images that were overfit. But note in that article that those images mostly appeared many times in the data set.
People who own the rights to one of those images have a valid argument. Everyone else doesn’t.
It’s illegal to reproduce copyrighted material*. That includes changing the format as well as things which fall under “derivative” works, but not creating a new work in the style of someone else’s (unless it falls under the derivative definition). Many voice impersonators exist and the way you impersonate a voice is to listen to (usually) recordings of that person and practice producing the same sounds that they use for common phonemes (as well as vocal tract shape and larynx positioning to alter the vocal pitch production and overtones which represent vowel shapes). ML does, effectively the same thing without requiring a human to do the listening and practicing.
That said, I think this type of use should be strictly prohibited. In fact, I think it should have severe criminal penalties for any specific voice, not just celebrities. Having the ability to simulate accurate, regional-sounded voices is extremely valuable in the general sense, but imitating or mimicking a specific person’s voice without their explicit consent and/or direction has very few, if any, legitimate uses.
* I didn’t think that voice mimicking would count as valid for any law, but Google tells me of the “Right of Publicity” and there is (again according to Google) case law involving Ford and Bette Midler. So while it’s not a copyright violation to reproduce a voice, it may still run afoul of some laws.
Again: how is that different from an imitation? What exactly differentiates a human watching a movie to imitate a voice from a machine doing the same thing?
And that is, what you misunderstand. AI is not magic, it’s computation. Nothing more. In no other context would it even matter, whether the source data was intended for the use case, if no infringement is being committed by the end product.