My problem with AI is that it has humanity mediocre, filtered for the masses culture baked into it by definition.
This is machinery that can not suggest experimental jazz- unless it is already mainstream. It can not go to the fringes and shove the species towards new and wild discoveries.
I find it hilarious though, that it may destroy the cultural behemoths with their IPs by mashUp. Death by a million beatles-clone songs, yesterday came suddenly indeed. And after all that is sad and done for- some of us, might even venture out to the fringe and find the weird and wild parts again.
I'm not AI's biggest fan, but it is training it's virtual brain on images and data, just like a creative human would, yet it is just far quicker at that.
Do we need to limit scraping and learning to only free publicly available data? Therefore is a Reuters photo still not publicly available for learning style of photo, composition, lighting, even with a watermark. What you pay for is the license to reproduce the image, you can still look at them beforehand.
Yes, all that data that ai is sucking up belongs to Google, twitter and Facebook! Lol. They are the owners of it. Those corporates didn't commit legal theft!
So ridiculous.
The whole idea of copyright is wrong. Anything that is popular and therefore successful, owes that to the crowd that made it popular. The crowd has is what gives it the interest.
I personally wouldn't even be averse to some sort period of copyright, say 2 years, with the possibility to extend to say 5, but these things are common.
One of two legal outcomes must follow from OpenAI's piracy.
1. OpenAI must be fined according to law.
2. Piracy is decriminalized.
Failing to do neither is an admission that the US has become a corporatocracy. That is, a form of oligarchy where the rule of law is not by a majority or plurality of people, but by a number or corporations you can count with one hand.
I don't understand the example image. The description says:
> The yellow circles highlight areas of similarity between the original photo and the AI-generated photo
What about the person in the middle, doing the throwing? It's exactly the same. Why don't they have a circle around them? Why highlight some lens flares & distorted faces when the actual subject is the same? Or was this some kind of image-to-image generation?
He gives two examples that are meant to prove his point, but they don't really convince me.
The first example is the image: the AI has seen the famous photograph of the Ferguson riots, and (with 6 prompts?!) manages to get something fairly similar. But suppose a human had seen that photo, and then you asked that human to draw you a photo of the riots; and then continued prompting them to make it look similar. Is it really unrealistic that the human could generate something that looks as similar? Is the human themselves therefore inherently a violation of copyright?
The NYT article to begin with looks a bit more damning -- except that, it appears that they prompted the AI directly with the beginning of the article. My son, when a toddler, for a long time could recite nearly the full text of his favorite story books with minimal prompting -- does that mean he's inherently a violation of copyright? Because he can recite The Gruffalo almost verbatim when prompted, is he a walking violation of Julia Donaldson's copyright? What about people with photographic memory, that can recite long sections of books verbatim -- are they inherently violating copyright?
Now sure, in both cases, the output might be a violation of copyright, if it's clearly derived from it -- both for humans and for AI. But I don't think the fact that AI can be prompted to generate copyright-violating material is proof that the AI training itself has violated copyright, any more than the fact that a human can be prompted to generate copyright-violating material is proof that human training has violated copyright.
As much as I think that what these companies is doing has moral and legal issues, reframing copyright violations as "theft" will cut both ways and make the arguments for open culture more difficult.
The fact that they called this captioned as "The bottom photo is an AI-generated image created in six prompts using OpenAI’s ChatGPT", without actually releasing the 6 prompts is quite telling. Because that will show they were prompting things to match the original image.
Iconic images (mona lisa, tank man, widely reported news stories, styles like ghibli) would of course be incorporated in as styles. It doesn't refute fair use.
So, you can't say "draw this person in the mona lisa pose, in simpsons style" and then act surprised and shocked when the model does exactly that. That's not theft.
I think most people are already pretty annoyed with how copyright law makes very reasonable applications of technology impossible. What these whiny retards want to do is put copyright between everyone and a technical revolution.
Do that and copyright will go away entirely. If you like copyright, try to come up with practical and reasonable compromises. Don't try to break things.
It's obvious to me that the converse is true here as well... Fair Use is NOT theft.
I've seen plenty of photos of "The Bean" in Chicago, most of which aren't mine if I look really, really closely, but that's all ok. I'd be an idiot if I got upset by any other similar photos.
Clearly copyright needs to be updated for the modern era, but perhaps it's reputation that should replace it? Instead of focusing on the content, let's instead work to secure the sources of information, so that we know a story is from a given source, and everything has a digital provenance.
13 comments
[ 3.3 ms ] story [ 32.4 ms ] threadThis is machinery that can not suggest experimental jazz- unless it is already mainstream. It can not go to the fringes and shove the species towards new and wild discoveries.
I find it hilarious though, that it may destroy the cultural behemoths with their IPs by mashUp. Death by a million beatles-clone songs, yesterday came suddenly indeed. And after all that is sad and done for- some of us, might even venture out to the fringe and find the weird and wild parts again.
Do we need to limit scraping and learning to only free publicly available data? Therefore is a Reuters photo still not publicly available for learning style of photo, composition, lighting, even with a watermark. What you pay for is the license to reproduce the image, you can still look at them beforehand.
So ridiculous.
The whole idea of copyright is wrong. Anything that is popular and therefore successful, owes that to the crowd that made it popular. The crowd has is what gives it the interest.
I personally wouldn't even be averse to some sort period of copyright, say 2 years, with the possibility to extend to say 5, but these things are common.
1. OpenAI must be fined according to law.
2. Piracy is decriminalized.
Failing to do neither is an admission that the US has become a corporatocracy. That is, a form of oligarchy where the rule of law is not by a majority or plurality of people, but by a number or corporations you can count with one hand.
> The yellow circles highlight areas of similarity between the original photo and the AI-generated photo
What about the person in the middle, doing the throwing? It's exactly the same. Why don't they have a circle around them? Why highlight some lens flares & distorted faces when the actual subject is the same? Or was this some kind of image-to-image generation?
Yes, be damned and fuck them. And stop pursuing individuals too.
You can have omniscient AI without feeding all of the data into it.
The first example is the image: the AI has seen the famous photograph of the Ferguson riots, and (with 6 prompts?!) manages to get something fairly similar. But suppose a human had seen that photo, and then you asked that human to draw you a photo of the riots; and then continued prompting them to make it look similar. Is it really unrealistic that the human could generate something that looks as similar? Is the human themselves therefore inherently a violation of copyright?
The NYT article to begin with looks a bit more damning -- except that, it appears that they prompted the AI directly with the beginning of the article. My son, when a toddler, for a long time could recite nearly the full text of his favorite story books with minimal prompting -- does that mean he's inherently a violation of copyright? Because he can recite The Gruffalo almost verbatim when prompted, is he a walking violation of Julia Donaldson's copyright? What about people with photographic memory, that can recite long sections of books verbatim -- are they inherently violating copyright?
Now sure, in both cases, the output might be a violation of copyright, if it's clearly derived from it -- both for humans and for AI. But I don't think the fact that AI can be prompted to generate copyright-violating material is proof that the AI training itself has violated copyright, any more than the fact that a human can be prompted to generate copyright-violating material is proof that human training has violated copyright.
As much as I think that what these companies is doing has moral and legal issues, reframing copyright violations as "theft" will cut both ways and make the arguments for open culture more difficult.
Iconic images (mona lisa, tank man, widely reported news stories, styles like ghibli) would of course be incorporated in as styles. It doesn't refute fair use.
So, you can't say "draw this person in the mona lisa pose, in simpsons style" and then act surprised and shocked when the model does exactly that. That's not theft.
Do that and copyright will go away entirely. If you like copyright, try to come up with practical and reasonable compromises. Don't try to break things.
I've seen plenty of photos of "The Bean" in Chicago, most of which aren't mine if I look really, really closely, but that's all ok. I'd be an idiot if I got upset by any other similar photos.
Clearly copyright needs to be updated for the modern era, but perhaps it's reputation that should replace it? Instead of focusing on the content, let's instead work to secure the sources of information, so that we know a story is from a given source, and everything has a digital provenance.