What's the difference between a human artist walking around in the world, seeing images and videos (many of which are under copyright) and then using those images as inspiration, versus an AI being fed millions of images and videos as part of a large training set?
I am not a lawyer but I don't really see the "it's not fair use" argument.
> What's the difference between a human artist walking around in the world, seeing images and videos (many of which are under copyright) and then using those images as inspiration, versus an AI being fed millions of images and videos as part of a large training set?
What's the difference between retelling someone's story orally, and using a printing press to make an exact replica?
What's the difference between writing down a conversation from memory, and recording that conversation?
What's the difference between making a nude painting of someone, and taking a nude photograph?
What's the difference between going out in public among human beings, and going out in public where there's a facial recognition camera on every corner feeding everyone's movements into a centralized database?
What's the difference between the grandma who knows everyone in the village, vs the social media company that knows everyone in the world?
Social customs evolved in a context of fundamental, sharp limits to human cognition and skill. When technology smashes through those limits, those social customs don't work anymore. Copyright law didn't exist in a world where you couldn't copy things mechanically. If you apply the old rules naively, you end up with a nasty world that nobody wants. So we have to invent new rules to limit how people use the new technology, otherwise people get exploited and life becomes intolerable.
"What does copyright law say about this" doesn't even make sense as a question. Copyright was invented before AI was. The question should be, "what would a society someone would want to live in, say about this?"
> What's the difference between retelling someone's story orally, and using a printing press to make an exact replica?
But that's not what's happening. You can't "crack open" a model to find the bitmap of a piece of training data. It's not there any more than a painting you've seen is "in" your brain. A model sometimes creates things that look similar to existing pieces if it got a ton of copies of the same image, for the same reason that most people's artistic rendition of a tree is going to be more accurate than their rendition of an anteater.
The entire AI art debate is a symptom of the fact that society is structured such that labor-saving technological advancements can harm more people than they help, and it's baffling to see otherwise intelligent and ideologically similar people fixating on this single, relatively minor technology rather than the much more consequential, broader issue of the fact that automation always moves wealth into the hands of a smaller and smaller group of people.
That was a fact during the industrial revolution and it’s quite possible it will happen now with the digital revolution.
The average height of Englishmen actually went down during the industrial rev. presumably due to malnutrition. New histories of the period think of it as potentially as one of the few times we know where there was an enormous economic boom alongside a general collapse of quality of life.
>But that's not what's happening. You can't "crack open" a model to find the bitmap of a piece of training data. It's not there any more than a painting you've seen is "in" your brain. A model sometimes creates things that look similar to existing pieces if it got a ton of copies of the same image, for the same reason that most people's artistic rendition of a tree is going to be more accurate than their rendition of an anteater.
In other words, a technology that has never existed before and doesn't follow any of the categories and schemas we dreamt up to organize the world. So we need to make new rules.
For humans the inspiration can be something else than the medium to use. Watching reality can provide inspiration for creating a video. And I'm pretty sure that's wher most inspiration comes from. It cannot for AI. Of course in a world where everything exists AI doesn't need inspiration from other sources.
Then what was it saying? It said that humans can take inspiration from outside art, from the real world. You can do the same thing with AI by training it on pictures from the real world.
Always back to the same debate then.
Why aren't human minds glorified functions, that perhaps deserve their glory a bit more than current state of AI?
Obviously they're not currently equivalent. But obviously there are similarities.
If I can take copyrighted images and feed them into an AI, why can’t I take any code I find online and ignore the license terms and feed it into a compiler? How do you justify the existence of something like the GPL?
Is it ok to scan copyrighted works. I think Google Books' win at the Supreme Court shows that yes it is.
Is it ok to process them down to the rawest information?
Is it ok for people to generate content from the raw information (and ok for them to charge for it)?
The last two I don't have an answer for, but I understand the fear of artists and the anger that their hard work is being shovelled into the monster that is going affect their work. But also more broadly the make-up industry, the lighting industry, studio spaces, lens makers, paint manufacturers, etc etc etc.
> why can’t I take any code I find online and ignore the license terms and feed it into a compiler?
a compiler is not transformative. You translating a book into another language doesn't make it a new work.
But an AI that takes billions of images, and uses it to synthesize something different and new, is fairly transformative under my eyes, and deserves new copyright. Unless the AI generated image is largely composed of a small number of works, i don't see why anyone should have copyright ownership of such an output!
its actually exactly the same thing, the ai doesnt "synthetize" anything, its following its programming and data based on training sets
if anything I would argue that human taking code from multiple places and making it compile into something more useful is much more synthetiz-ing than AI...
What's the difference between a programmer looking at lots of open source code and github copilot looking at open source code? (And I'm talking MIT licensed code here).
I am not a lawyer but I don't really see the "it's not fair use" argument.