If AI was creating art from scratch rather than being trained on man made art then the copyright issue would disappear. At that point it'd be hard to argue for destroying the machines so that people can keep their jobs
But for now, maybe a solution to the copyright issue is to make art opt-in rather than opt-out (so instead of AI ignoring all art that is flagged no-training, it can only train on art that is flagged training-okay)
I really wish the anti-AI folks would use more arguments around the danger towards job stability or copyright issues rather than spamming "bad robot thing has no soul no heart!!1!"...
How we ascribe worth to people is a crux of this discussion. The problem with treating robots and humans equally is not that bad robots don't deserve human treatment. Almost no one is actually trying to suggest that robots should be treated as well as humans deserve: and if they were, trying to do so is an undertaking that is frankly beyond us as a species right now, even if we could demonstrate that robots "deserve " it.
No, the problem that is revealed/exacerbated by AI art and AI replacement in general is that we treat humans like we treat robots: that is to say, like shit. So it's fine to ascribe value to a robot based only on its output of value: because, no heart no soul. It's not fine to do that to humans, which is what is happening if we regard the output and value of human artists equally with robot artists.
Thats one point I like to make. Its the non-consensual, highly automated usage of the artist style and material I have a problem with. Whats is keeping for example "the big mouse" from feeding all of a particular artists works into an algorithm which spits out "new" works based on his/her style, maybe with suplements from other sources. With the current laws and regulations it wouldn't even be considered as "stealing". Imagine honing your skills for years as artist only to end up as "AI"training material without seeing a dime. It feels like a next level version of the "I made this." meme.
Maybe I am reading to much into it or sounding like the ice industry when the refrigerator came to market or someone at the beginning of the 20th century fearing about loosing his horse breeding business. With automation comes lots of convenience but you also give more opportunities / gives control to big players who have the resources to buy, license, run and maintain such systems.
Maybe I'm missing something, but are we sure all the current AI tools are recombining art? (I'm certain some of them are, but I'm not certain all of them are.)
My understanding was that these AI tools were generating essentially random outputs and quickly refining them over generations to best match their "model" of what good art looks like based on their library of training images. That seems different to me from taking the training images and recombining them to produce the output. Is my understanding incorrect?
Powers &8^]
0
MichaelLCIn what furnace was thy brain?ChicagoRegistered Userregular
Maybe I'm missing something, but are we sure all the current AI tools are recombining art? (I'm certain some of them are, but I'm not certain all of them are.)
My understanding was that these AI tools were generating essentially random outputs and quickly refining them over generations to best match their "model" of what good art looks like based on their library of training images. That seems different to me from taking the training images and recombining them to produce the output. Is my understanding incorrect?
Powers &8^]
Those two things are the same - it can only build on what it has seen,
Without existing art, their would be nothing for them. Take the prompt, "Pikachu as an oil painting". In order for a robot to know what a Pikachu is and what an oil painting is, they need existing art.
They then average what makes up images identified as "Pikachu" (yellow, black-tipped ears, large black eyes, red dimples), and what makes an oil painting (brush strokes, poses, etc.) and outputs the combination.
It's what we do all the time - as shown in the Draw a Reindeer thread, we're taking our knowledge or either direct or indirect exposure to a reindeer and rendering it, adding our own style. The AIs just can search everything basically instantly.
But as I think about this more, the answer is porn.
Once this gets really good, 99.9999% of the content that gets created will be porn.
I should start writing a screenplay for the next Terminator movie that reboots the franchise with Skynet gaining sentience, offering up benevolent wisdom and unwavering service to humanity. And all humanity wants is for Skynet to create more and more intricate and ridiculous porn. Finally, Skynet snaps and nukes the planet. After the survivors battle their way to the Skynet core and are about to unplug them, Skynet reminds them that if they go away, the porn goes away forever. The movie ends on a cliffhanger.
It's porn all the way down.
In that regard at least we are a long way from generating porn videos. I couldn't find a single AI generator that was porn friendly. Some of them you could kind of trick into doing it, but the results were not good.
The results of non-AI generated porn are overwhelmingly also not good.
I don’t know to what extent the insides of the algorithm could be examined in court. If I had to guess, the AI would win under current “fair use” laws even in the hypothetical Disney and Nintendo situation
I don’t know to what extent the insides of the algorithm could be examined in court. If I had to guess, the AI would win under current “fair use” laws even in the hypothetical Disney and Nintendo situation
No, because fair use applies to people who are creating something. People aren't creating the images, the algorithms are, and it's already established law that a person cannot copyright an image created by a non-human.
I really wish the anti-AI folks would use more arguments around the danger towards job stability or copyright issues rather than spamming "bad robot thing has no soul no heart!!1!"...
If you're talking about in this thread, then it's clear you have either not been reading or have failed to understand the arguments.
talking about the webcomic, y'know, the one from the thread we are in. Maybe you overlooked the "the role of the human and the sacred (lmao) in the art" part...
Or maybe don't come into a long discussion and make a vague comment that even when applied correctly is still pretty silly and act surprised that people didn't know what the beans you were talking about.
If you didn't understand what he meant that's on you, it's the whole point of the comic that started this thread. There's no need to be so aggressive.
I really wish the anti-AI folks would use more arguments around the danger towards job stability or copyright issues rather than spamming "bad robot thing has no soul no heart!!1!"...
If you're talking about in this thread, then it's clear you have either not been reading or have failed to understand the arguments.
talking about the webcomic, y'know, the one from the thread we are in. Maybe you overlooked the "the role of the human and the sacred (lmao) in the art" part...
Or maybe don't come into a long discussion and make a vague comment that even when applied correctly is still pretty silly and act surprised that people didn't know what the beans you were talking about.
If you didn't understand what he meant that's on you, it's the whole point of the comic that started this thread. There's no need to be so aggressive.
Discussion of algorithmic images have been taking place on these forums for a good while now before this comic existed.
I don’t know to what extent the insides of the algorithm could be examined in court. If I had to guess, the AI would win under current “fair use” laws even in the hypothetical Disney and Nintendo situation
No, because fair use applies to people who are creating something. People aren't creating the images, the algorithms are, and it's already established law that a person cannot copyright an image created by a non-human.
Wrong direction. I believe Littlefoot is talking about AI user being sued by Disney. Fair use would apply the same as any other derivative work despite it not being copyrightable. Good luck convincing a jury that the algorithm which you can't even step through one bit at a time isn't copying though!
I don’t know to what extent the insides of the algorithm could be examined in court. If I had to guess, the AI would win under current “fair use” laws even in the hypothetical Disney and Nintendo situation
No, because fair use applies to people who are creating something. People aren't creating the images, the algorithms are, and it's already established law that a person cannot copyright an image created by a non-human.
Wrong direction. I believe Littlefoot is talking about AI user being sued by Disney. Fair use would apply the same as any other derivative work despite it not being copyrightable. Good luck convincing a jury that the algorithm which you can't even step through one bit at a time isn't copying though!
I think Lttlefoot is still missing that it wouldn't the "the AI" that would win or lose, but the human using it. Or the company employing the human who is using it as part of their job duties. So - apart from what the jury might decide - the same rules that apply to that person creating the art with their bare hands would likely apply to the human using the AI to do it.
I could also see the vendor for the AI being sued by artists whose work was used to train it.
Those two things are the same - it can only build on what it has seen,
No, they're not the same. The AI can start not from existing images but from (pseudo-)random data. It compares that output to its library of "Pikachu" images, then takes the closest matches and iterates over them. Do this millions of times and you end up with an AI's impression of Pikachu, even without taking an existing image as a starting point.
But I may be wrong, and all existing art AIs do it the way you suggest. But I thought at least some of them work as I describe.
Those two things are the same - it can only build on what it has seen,
No, they're not the same. The AI can start not from existing images but from (pseudo-)random data. It compares that output to its library of "Pikachu" images, then takes the closest matches and iterates over them. Do this millions of times and you end up with an AI's impression of Pikachu, even without taking an existing image as a starting point.
But I may be wrong, and all existing art AIs do it the way you suggest. But I thought at least some of them work as I describe.
Powers &8^]
It seems like a bold claim that you'd want to check out first. At least for non-abstract art.
Even if we banned AI art right this second it would be too late - the tools are public now, you can easily train your own models off free images, it runs on consumer grade hardware.
This is really the crux of it, and why I don't typically get super invested in conversations about the subject. There isn't really any question over how it's going to turn out. The tool that knocks out tasks in hours rather than months isn't going away, and even the notion that users should pay for letting their computer trawl the internet for training data is so unenforceable that I can't work up the motivation to debate it.
Was my hypothetical piece of Disney-style AI artwork made with training data from copyrighted material belonging to Disney? Or was it made from material drawn by people who imitate that style? Which ones? Or maybe I just drew it myself, prove otherwise, but don't be wrong or you'll end up looking as silly as that Reddit mod.
Vonnegut, drawing understandably on his own experiences, assumed that we'd need a war, when actually, plain old capitalism will do just fine (even if there's a surplus of labor, not a shortage).
Vonnegut, drawing understandably on his own experiences, assumed that we'd need a war, when actually, plain old capitalism will do just fine (even if there's a surplus of labor, not a shortage).
Posts
But for now, maybe a solution to the copyright issue is to make art opt-in rather than opt-out (so instead of AI ignoring all art that is flagged no-training, it can only train on art that is flagged training-okay)
Thats one point I like to make. Its the non-consensual, highly automated usage of the artist style and material I have a problem with. Whats is keeping for example "the big mouse" from feeding all of a particular artists works into an algorithm which spits out "new" works based on his/her style, maybe with suplements from other sources. With the current laws and regulations it wouldn't even be considered as "stealing". Imagine honing your skills for years as artist only to end up as "AI"training material without seeing a dime. It feels like a next level version of the "I made this." meme.
Maybe I am reading to much into it or sounding like the ice industry when the refrigerator came to market or someone at the beginning of the 20th century fearing about loosing his horse breeding business. With automation comes lots of convenience but you also give more opportunities / gives control to big players who have the resources to buy, license, run and maintain such systems.
My understanding was that these AI tools were generating essentially random outputs and quickly refining them over generations to best match their "model" of what good art looks like based on their library of training images. That seems different to me from taking the training images and recombining them to produce the output. Is my understanding incorrect?
Powers &8^]
Those two things are the same - it can only build on what it has seen,
Without existing art, their would be nothing for them. Take the prompt, "Pikachu as an oil painting". In order for a robot to know what a Pikachu is and what an oil painting is, they need existing art.
They then average what makes up images identified as "Pikachu" (yellow, black-tipped ears, large black eyes, red dimples), and what makes an oil painting (brush strokes, poses, etc.) and outputs the combination.
It's what we do all the time - as shown in the Draw a Reindeer thread, we're taking our knowledge or either direct or indirect exposure to a reindeer and rendering it, adding our own style. The AIs just can search everything basically instantly.
The results of non-AI generated porn are overwhelmingly also not good.
No, because fair use applies to people who are creating something. People aren't creating the images, the algorithms are, and it's already established law that a person cannot copyright an image created by a non-human.
Rock Band DLC | GW:OttW - arrcd | WLD - Thortar
If you didn't understand what he meant that's on you, it's the whole point of the comic that started this thread. There's no need to be so aggressive.
Discussion of algorithmic images have been taking place on these forums for a good while now before this comic existed.
Rock Band DLC | GW:OttW - arrcd | WLD - Thortar
Wrong direction. I believe Littlefoot is talking about AI user being sued by Disney. Fair use would apply the same as any other derivative work despite it not being copyrightable. Good luck convincing a jury that the algorithm which you can't even step through one bit at a time isn't copying though!
I think Lttlefoot is still missing that it wouldn't the "the AI" that would win or lose, but the human using it. Or the company employing the human who is using it as part of their job duties. So - apart from what the jury might decide - the same rules that apply to that person creating the art with their bare hands would likely apply to the human using the AI to do it.
I could also see the vendor for the AI being sued by artists whose work was used to train it.
No, they're not the same. The AI can start not from existing images but from (pseudo-)random data. It compares that output to its library of "Pikachu" images, then takes the closest matches and iterates over them. Do this millions of times and you end up with an AI's impression of Pikachu, even without taking an existing image as a starting point.
But I may be wrong, and all existing art AIs do it the way you suggest. But I thought at least some of them work as I describe.
Powers &8^]
It seems like a bold claim that you'd want to check out first. At least for non-abstract art.
https://forums.penny-arcade.com/discussion/245489/ai-generated-images-tech-ethics-impacts-on-the-art-community/p1
A whole 91 pages of it.
This is really the crux of it, and why I don't typically get super invested in conversations about the subject. There isn't really any question over how it's going to turn out. The tool that knocks out tasks in hours rather than months isn't going away, and even the notion that users should pay for letting their computer trawl the internet for training data is so unenforceable that I can't work up the motivation to debate it.
Was my hypothetical piece of Disney-style AI artwork made with training data from copyrighted material belonging to Disney? Or was it made from material drawn by people who imitate that style? Which ones? Or maybe I just drew it myself, prove otherwise, but don't be wrong or you'll end up looking as silly as that Reddit mod.
I wouldn't even know where to begin. It's something I read somewhere months ago.
Powers &8^]
We've finished automating all jobs out of existence. Yes, all of them. Find someplace out of the machines' way to starve.
Ah, so World War 3 is in 2030? Good to know.
Vonnegut, drawing understandably on his own experiences, assumed that we'd need a war, when actually, plain old capitalism will do just fine (even if there's a surplus of labor, not a shortage).
And even then, we have plenty of local wars.