If you trained a model on a single copyrighted work, then that would be a copyright violation because it would inevitably produce output similar to that single work.
But if you train it on hundreds of thousands of copyrighted works, that’s no longer a copyright violation, because output won’t closely match any single work.
How is something a crime if you do it once, but not if you do it a million times?
A basic fundamental of copyright law and fair use is if the result is transformative. People literally do stuff like make collages with copyright works and it’s fine in many cases.
Turning pictures into an AI model (and that’s being really generous in my phrasing as if the pictures have anything to do with the math) is just about one of the most transformative things you can do with a picture.
This is like copyright 101 and if you’re shocked you don’t understand what you’re talking about in regards to copyright.
Except it’s not really transformative because the end product is not the model itself. The product is a service that writes code or draws pictures. It is literally the exact same as the input and it is intended specifically to avoid having to buy the inputs.
The product is a service that writes code or draws pictures. It is literally the exact same as the input
Pictures and things that draw pictures aren’t the same thing.
The fact it’s a tool that makes art and completes with you has nothing to do with copyright. That would only apply if this was some convoluted scheme to make actual copies of works, which it isn’t. People just pirate for that. If I wanted to read this person’s books I’d go to pirate Bay, not chat GPT.
It’s not illegal for someone to read your books and start writing similar things. That’s not copyright theft, that’s a genre.
They are selling access to the AI model which draws pictures. Not the original pictures, nor clones of those pictures. A machine to which you can input a prompt that is basically anything and get custom art back as a result.
Also there are companies like stability AI which is providing direct access to the model itself, and I’m sure you’re against them as well.
The “transformation” is the petty technicality in my opinion. Would it be transformative if I sold you a database of base64 encoded images? What about if they were encrypted?
Hell, you can hire me to paint based on prompts you give me. That’s the exact same service an AI provides, no? I’m going to study copyrighted materials to get better at my service. Surely if pictures -> AI model is transformative, then pictures -> knowledge in my brain is transformative as well. So you give me the prompt “Mickey Mouse” and I draw this. This is “custom art”. You think you can use that commercially? And if you realize that you can’t, why do you think I should be able to legally sell you this service?
Have you ever been to the market part of a fan convention? People sell a shitload of copywrited art there, and no one really cared about that. The fact that you wouldnt be able to use a lot of those things commercially doesnt stop people from buying them.
Edit: Also if you sold that database I wouldnt buy it because I dont give a shit about the images the machine was trained on, I give a shit about the art I ask it to make for me, which it consistently does exactly the way I want. Is commissioning humans illegal now?
Would it be transformative if I sold you a database of base64 encoded images? What about if they were encrypted
No.
Also no.
There is a long history of examples set by court cases on what does or doesn’t count as transformative. Law is very good at handling exceptions like this and it’s been handling them for decades.
An encoding is not transformative. It’s just the same information sent a different way. Same with encryption.
Hell, you can hire me to paint based on prompts you give me. That’s the exact same service an AI provides, no? I’m going to study copyrighted materials to get better at my service.
All perfectly legal and commonly done.
So you give me the prompt “Mickey Mouse” and I draw this. This is “custom art”. You think you can use that commercially?
No. Not for you and not with AI generated art either.
Copyright controls your ability to copy and distribute creative works. You can learn to draw Micky mouse, you can even draw Micky mouse, but anyone who tries to sell or distribute that copy can and probably will quickly get sued for it.
And if you realize that you can’t, why do you think I should be able to legally sell you this service?
If AI companies were predominantly advertising themselves as “we make your pictures of Micky mouse” you’d have a valid point.
But at this point you’re basically arguing that it should be impossible to sell a magical machine that can draw anything you ask from it because it could be asked to draw copyright images.
Courts will see that argument, realize it’s absurd, and shut it down.
If AI companies were predominantly advertising themselves as “we make your pictures of Micky mouse” you’d have a valid point.
Doesn’t matter what it’s advertised as. That picture is, you agree, unusable. But the site I linked to above is selling this service and it’s telling me I can use the images in any way I want. I’m not stupid enough to use Mickey Mouse commercially, but what happens when the output is extremely similar to a character I’ve never heard of? I’m going to use it assuming it is an AI-generated character, and the creator is very unlikely to find out unless my work ends up being very famous. The end result is that the copyright of everything not widely recognizable is practically meaningless if we accept this practice.
But at this point you’re basically arguing that it should be impossible to sell a magical machine that can draw anything you ask from it because it could be asked to draw copyright images.
Straw man. This is not a magical device that can “draw anything”, and it doesn’t just happen to be able to draw copyrighted images as a side-effect of being able to create every imaginable thing, as you try to make it sound. This is a mundane device whose sole function is to try to copy patterns from its input set, which unfortunately is pirated. If you want to prove me wrong, make your own model without a single image of Micky Mouse or a tag with his name, then try to get it to draw him like I did before. You will fail because this machine’s ability to draw him is dependent on being trained on images of him.
There are many ways this could be done ethically, like:
build it on open datasets, or on datasets you own, instead of pirating
don’t commercialize it
allow non-commercial uses, like research or just messing around (which would be a real transformative use)
Training the AI isn’t a copyright violation though. Producing content from a single source of training information is intuitively different from producing content from a litany of sources. Is there a distinction I’m not understanding that you are pointing out?
“AI” models are, essentially, solvers for mathematical system that we, humans, cannot describe and create solvers for ourselves.
For example, a calculator for pure numbers is a pretty simple device all the logic of which can be designed by a human directly. A language, thought? Or an image classifier? That is not possible to create by hand.
With “AI” instead of designing all the logic manually, we create a system which can end up in a number of finite, yet still near infinite states, each of which defines behavior different from the other. By slowly tuning the model using existing data and checking its performance we (ideally) end up with a solver for some incredibly complex system.
If we were to try to make a regular calculator that way and all we were giving the model was “2+2=4” it would memorize the equation without understanding it. That’s called “overfitting” and that’s something people being AI are trying their best to prevent from happening. It happens if the training data contains too many repeats of the same thing.
However, if there is no repetition in the training set, the model is forced to actually learn the patterns in the data, instead of data itself.
Essentially: if you’re training a model on single copyrighted work, you’re making a copy of that work via overfitting. If you’re using terabytes of diverse data, overfitting is minimized. Instead, the resulting model has actual understanding of the system you’re training it on.
How is something a crime if you do it once, but not if you do it a million times?
Because doing it a million times seriously dilutes the harm to any single content creator (assuming those million sources are from many, many different content creators, of course). Potential harm plays a major role in how copyright cases are determined, and in cases involving such a huge amount of sources, harm can be immeasurably small.
In addition to right and wrong, the practicality of regulation and enforcement is often a part of groundbreaking decisions like these, and I’m not certain this particular issue is something our legal system is equipped to handle.
I’m not sure I agree with the reasoning here, but I see their thinking.
An AI trained on a single image would also probably be fine if it was somehow a generalist AI that didn’t overfit on that single image. The quantity really doesn’t matter.
Imagine this situation if a human replaced the AI.
Imagine a human who wants to write a book. They’ve read hundreds of other books already, and lots of other things besides books. Then they write a book. The final work probably contains an amalgamation of all the other things they’ve read–similar characters, themes, plot points, etc.–but it’s a unique combination, so it’s distinct from those other works. No copyright violation.
Now imagine that same human has only ever read one book. Over and over. They know only the one book. The human wants to write a new book. But they only have experience with the one they’ve read again and again. So the book they write is almost exactly the same as the one book they read. That’s a copyright violation.
Training an AI model is not a crime, any more than reading a book is a crime. You’re not making “copies” or profiting directly from that single work.
How is something a crime if you do it once, but not if you do it a million times?
Companies get to steal from people all the time without repercussions through erroneous fees, ‘mistakes’ in billing, denying coverage, and even outright fraud only gets a slap on the wrist fine at best. But an average person steals $5 and they are thrown in jail.
How is something a crime if you do it once, but not if you do it a million times?
You can dream up other examples of this.
If you’re a DJ performing for a large audience and yell “I want to see you shake it for me!”, that is legal.
If you walk up to one specific woman on the street and pull her aside and say “I want to see you shake it for me”, that’s sexual harassment.
If the government announces that the median income of Detroit residents has gone up by 3%, that’s normal.
If the government public announces that John Fuckface, 36.2 years old, living at 123 Fake Street, had his income increase by 5% in the previous year, that’s a privacy violation.
The whole point of training the AI is to build a model that can’t reproduce a single work.
It may seem superficially strange, but the more works you include, the less capable it is of reproducing one work.
How is something a crime if you do it once, but not if you do it a million times?
Because we are talking about a generalized knowledge base vs a specific one? Is it not obvious from the explanation you quoted that instructing an AI to respond off of millions of sources means that it isn’t biased off of one person’s work?
So judges are saying:
If you trained a model on a single copyrighted work, then that would be a copyright violation because it would inevitably produce output similar to that single work.
But if you train it on hundreds of thousands of copyrighted works, that’s no longer a copyright violation, because output won’t closely match any single work.
How is something a crime if you do it once, but not if you do it a million times?
It reminds me of the scheme from Office Space: https://youtu.be/yZjCQ3T5yXo
A basic fundamental of copyright law and fair use is if the result is transformative. People literally do stuff like make collages with copyright works and it’s fine in many cases.
Turning pictures into an AI model (and that’s being really generous in my phrasing as if the pictures have anything to do with the math) is just about one of the most transformative things you can do with a picture.
This is like copyright 101 and if you’re shocked you don’t understand what you’re talking about in regards to copyright.
deleted by creator
Man, its refreshing that Lemmy seems to have people with more nuanced takes on AI than the rest of.the internet
Except it’s not really transformative because the end product is not the model itself. The product is a service that writes code or draws pictures. It is literally the exact same as the input and it is intended specifically to avoid having to buy the inputs.
Pictures and things that draw pictures aren’t the same thing.
The fact it’s a tool that makes art and completes with you has nothing to do with copyright. That would only apply if this was some convoluted scheme to make actual copies of works, which it isn’t. People just pirate for that. If I wanted to read this person’s books I’d go to pirate Bay, not chat GPT.
It’s not illegal for someone to read your books and start writing similar things. That’s not copyright theft, that’s a genre.
And that’s completely irrelevant because “things that draw pictures” is not the work being sold. You’re buying pictures.
Seems like a petty technicality to me.
They are selling access to the AI model which draws pictures. Not the original pictures, nor clones of those pictures. A machine to which you can input a prompt that is basically anything and get custom art back as a result.
Also there are companies like stability AI which is providing direct access to the model itself, and I’m sure you’re against them as well.
The “transformation” is the petty technicality in my opinion. Would it be transformative if I sold you a database of base64 encoded images? What about if they were encrypted?
Hell, you can hire me to paint based on prompts you give me. That’s the exact same service an AI provides, no? I’m going to study copyrighted materials to get better at my service. Surely if pictures -> AI model is transformative, then pictures -> knowledge in my brain is transformative as well. So you give me the prompt “Mickey Mouse” and I draw this. This is “custom art”. You think you can use that commercially? And if you realize that you can’t, why do you think I should be able to legally sell you this service?
Have you ever been to the market part of a fan convention? People sell a shitload of copywrited art there, and no one really cared about that. The fact that you wouldnt be able to use a lot of those things commercially doesnt stop people from buying them.
Edit: Also if you sold that database I wouldnt buy it because I dont give a shit about the images the machine was trained on, I give a shit about the art I ask it to make for me, which it consistently does exactly the way I want. Is commissioning humans illegal now?
No.
Also no.
There is a long history of examples set by court cases on what does or doesn’t count as transformative. Law is very good at handling exceptions like this and it’s been handling them for decades.
An encoding is not transformative. It’s just the same information sent a different way. Same with encryption.
All perfectly legal and commonly done.
No. Not for you and not with AI generated art either.
Copyright controls your ability to copy and distribute creative works. You can learn to draw Micky mouse, you can even draw Micky mouse, but anyone who tries to sell or distribute that copy can and probably will quickly get sued for it.
If AI companies were predominantly advertising themselves as “we make your pictures of Micky mouse” you’d have a valid point.
But at this point you’re basically arguing that it should be impossible to sell a magical machine that can draw anything you ask from it because it could be asked to draw copyright images.
Courts will see that argument, realize it’s absurd, and shut it down.
Doesn’t matter what it’s advertised as. That picture is, you agree, unusable. But the site I linked to above is selling this service and it’s telling me I can use the images in any way I want. I’m not stupid enough to use Mickey Mouse commercially, but what happens when the output is extremely similar to a character I’ve never heard of? I’m going to use it assuming it is an AI-generated character, and the creator is very unlikely to find out unless my work ends up being very famous. The end result is that the copyright of everything not widely recognizable is practically meaningless if we accept this practice.
Straw man. This is not a magical device that can “draw anything”, and it doesn’t just happen to be able to draw copyrighted images as a side-effect of being able to create every imaginable thing, as you try to make it sound. This is a mundane device whose sole function is to try to copy patterns from its input set, which unfortunately is pirated. If you want to prove me wrong, make your own model without a single image of Micky Mouse or a tag with his name, then try to get it to draw him like I did before. You will fail because this machine’s ability to draw him is dependent on being trained on images of him.
There are many ways this could be done ethically, like:
Training the AI isn’t a copyright violation though. Producing content from a single source of training information is intuitively different from producing content from a litany of sources. Is there a distinction I’m not understanding that you are pointing out?
Nope, I think you nailed it.
I’ve trained my personal AI, my brain, by ingesting 1,000+ books. So now I can’t write a book?
Suppose I use a Stephen King phrase, “friends and neighbors”. Can’t use that? Of course I can.
“AI” models are, essentially, solvers for mathematical system that we, humans, cannot describe and create solvers for ourselves.
For example, a calculator for pure numbers is a pretty simple device all the logic of which can be designed by a human directly. A language, thought? Or an image classifier? That is not possible to create by hand.
With “AI” instead of designing all the logic manually, we create a system which can end up in a number of finite, yet still near infinite states, each of which defines behavior different from the other. By slowly tuning the model using existing data and checking its performance we (ideally) end up with a solver for some incredibly complex system.
If we were to try to make a regular calculator that way and all we were giving the model was “2+2=4” it would memorize the equation without understanding it. That’s called “overfitting” and that’s something people being AI are trying their best to prevent from happening. It happens if the training data contains too many repeats of the same thing.
However, if there is no repetition in the training set, the model is forced to actually learn the patterns in the data, instead of data itself.
Essentially: if you’re training a model on single copyrighted work, you’re making a copy of that work via overfitting. If you’re using terabytes of diverse data, overfitting is minimized. Instead, the resulting model has actual understanding of the system you’re training it on.
Because doing it a million times seriously dilutes the harm to any single content creator (assuming those million sources are from many, many different content creators, of course). Potential harm plays a major role in how copyright cases are determined, and in cases involving such a huge amount of sources, harm can be immeasurably small.
In addition to right and wrong, the practicality of regulation and enforcement is often a part of groundbreaking decisions like these, and I’m not certain this particular issue is something our legal system is equipped to handle.
I’m not sure I agree with the reasoning here, but I see their thinking.
An AI trained on a single image would also probably be fine if it was somehow a generalist AI that didn’t overfit on that single image. The quantity really doesn’t matter.
Imagine this situation if a human replaced the AI.
Imagine a human who wants to write a book. They’ve read hundreds of other books already, and lots of other things besides books. Then they write a book. The final work probably contains an amalgamation of all the other things they’ve read–similar characters, themes, plot points, etc.–but it’s a unique combination, so it’s distinct from those other works. No copyright violation.
Now imagine that same human has only ever read one book. Over and over. They know only the one book. The human wants to write a new book. But they only have experience with the one they’ve read again and again. So the book they write is almost exactly the same as the one book they read. That’s a copyright violation.
Training an AI model is not a crime, any more than reading a book is a crime. You’re not making “copies” or profiting directly from that single work.
Thank you for putting into words what my entire point is to people always claiming AI Art is theft
Companies get to steal from people all the time without repercussions through erroneous fees, ‘mistakes’ in billing, denying coverage, and even outright fraud only gets a slap on the wrist fine at best. But an average person steals $5 and they are thrown in jail.
You can dream up other examples of this.
If you’re a DJ performing for a large audience and yell “I want to see you shake it for me!”, that is legal. If you walk up to one specific woman on the street and pull her aside and say “I want to see you shake it for me”, that’s sexual harassment.
If the government announces that the median income of Detroit residents has gone up by 3%, that’s normal. If the government public announces that John Fuckface, 36.2 years old, living at 123 Fake Street, had his income increase by 5% in the previous year, that’s a privacy violation.
The whole point of training the AI is to build a model that can’t reproduce a single work. It may seem superficially strange, but the more works you include, the less capable it is of reproducing one work.
Because we are talking about a generalized knowledge base vs a specific one? Is it not obvious from the explanation you quoted that instructing an AI to respond off of millions of sources means that it isn’t biased off of one person’s work?
Here is an alternative Piped link(s):
https://piped.video/yZjCQ3T5yXo
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I’m open-source; check me out at GitHub.