4 ms·
> why isn't it that an artist could say, hey I'm letting you see this painting, but you are not allowed to sit down with a canvas and learn how to reproduce it?
by shkkmo 3y ago
> why isn't it that an artist could say, hey I'm letting you see this painting, but you are not allowed to sit down with a canvas and learn how to reproduce it? Because you can do that in galleries - no photos, no reproductions.
But you can't stop people from sitting and studying your painting and then painting stuff similar to it.
One of the core assertions that is being decided in this case is if there is any actual reproduction here. Does a model contain a reproduction of every image it was trained on? Can the model actually create a reproduction of any images it was trained on?
If it turns out that there is no reproduction here, then it comes down to how much legal control we give copyright owners to regulate access.
A gallery can reasonably ban cameras and canvases, but it becomes a lot less reasonable if they try to ban artists.
Let's imagine that this isn't just specifically tuned ML but proper General AI that can learn new skills. Is your argument that this AI would be legally prohibited from viewing any images it doesn't have a specific license for?
I think that drawing hard lines around what kind of processing can be done on publicly available images is going to become problematic. It's better to regulate around what can be done with the results of the processing than that processing itself. That's how our existing laws work. Making a reproduction, even just from memory, of a copyrighted work is restricted. Memorizing a copyrighted work is not.
- kranke155 3y agoI find the whole comparison "it´s just like a person learning" to be a tiring trope. It's demonstrably not. Like I said to another poster - you've probably seen a Picasso. Can you make me a copy? Because a Diffusion model can. But you can't. Why not? Your denial that there is a demonstrable difference between human and machine attention is part of the core obfuscation these companies are using to win this battle, so I reject it entirely. That difference creates the whole issue. If you don't recognise it, then answer me - Why can't you paint me a Picasso? You're saying the Machine is just like a human, yet a simple question of reproduction tells you it's not like a human in any way. It's a machine, and it produces machine reproductions. It learns faster and more accurately than any human, and its purpose is to produce derivative works. If the machine didn't need human data to do this, this discussion would be academic. But it does. So the whole future of the Arts will be decided by investigating what the machine actually does, not the simplistic idea of it´s just like a human. You have to evaluate the machine's abilities and impact onto the world. And that's the tough part. But just saying "hihih it's just a person" while it produces superhuman output is not a solution, it's just a lie that was invented by the people profiting from these models. >Is your argument that this AI would be legally prohibited from viewing any images it doesn't have a specific license for? Yes. You pay for access.
- lt 3y agoA diffusion model can't make a copy. That's the whole point. The original Picasso isn't in the model weights. It has learned to make pixels a particular color to mimic that style, but that's it.
- kranke155 3y agoIf the model didn't learn anything important from Picasso, it wouldn't be in the training data. This whole argument of "ah but it doesnt really need it" doesn't hold up. If the model didn't need it, it wouldn't have used it in the first place. Same thing in Artstation. It was of course propitious for AI scientists to find such a lovely database of high quality imagery, and all so helpfully tagged into categories. All they had to do was take it.
- lt 3y agoOf course it learned, that's the point of training. You claimed the model can reproduce an image from that training data. That's false, and what the judge dismissed. “none of the Stable Diffusion output images provided in response to a particular Text Prompt is likely to be a close match for any specific image in the training data.” “I am not convinced that copyright claims based a derivative theory can survive absent ‘substantial similarity’ type allegations,” the ruling stated. Whether using copyrighted data to train a model is fair use or not is a different discussion.
- shkkmo 3y ago> If the model didn't learn anything important from Picasso, it wouldn't be in the training data. > This whole argument of "ah but it doesnt really need it" doesn't hold up. If the model didn't need it, it wouldn't have used it in the first place. I haven't seen anyone making this argument. There's a pretty clear difference between learning something from an image and memorizing it. There also isn't any illegal with memorizing an image and painting a reproduction. What you aren't allowed to do is sell or distribute that reproduction without a license. I think it makes more sense to restrict what people are allowed do with ML tools than to restrict what ML tools can do.