It's rich to see people talk about the commons like AI models aren't a massive tragedy of the commons.
Rent-seeking by controlling AI models is anti-commons, the models themselves are not, neither is training them. Don't delude yourself into thinking that Miyazaki didn't study Moebius didn't study Mucha. Even if you got your way and all training on copyrighted material was stopped today and all existing models were deleted it would only be a matter of time before we could synthesize arbitrary styles (at least as well as we can today) by identifying and controlling the correct axes in tensor space using a model purely informed by public domain data.
Your art is mostly not yours, your art is mostly an amalgamation of other people's art which itself is also the same, all the way down. You are a sprinkling of flavor on top. Your art is stolen styles, techniques and vibes mashed together. And here you are, trapped arguing about where we draw the line on what's okay. 75 years? now 95? Look at what you're saying! Introspect and realize that your understanding of the context is defined by Disney and Bono wanting to collect rent on ideas. You're just focused on getting your crumbs of the pie in a game that's rigged to centralize power and keep it centralized.
original or "original"? maybe AI companies need to hire people to make "original" content for AI.
You might think I'm being rude or overly hostile to you, but frankly I think you deserve it. Your views are atrocious.
The "rent-seekers" is the problem. We collectively inherit and own our shared culture, but large corporations have always wanted to sell it back to us. AI companies are arguing they should have no limitations on their usage of the culture, but that the same shouldn't apply to them. Selling tickets to the commons is anti-commons.
Perhaps if these companies were themselves arguing for the end of copyright and IP for everyone, the conversation would be different.
In order to make attribution models we need text+author+timestamp. We can get that from books, newspaper articles, scientific papers and social network posts. Then we extend to the rest of the training set.
But then we can also make AI models that cleverly avoid infringement while the same strict checking is going to be applied to human made content. Humans are not that good at avoiding pitfalls.
Do it then.
> Your art is mostly not yours, your art is mostly an amalgamation of other people's art which itself is also the same, all the way down. You are a sprinkling of flavor on top. Your art is stolen styles, techniques and vibes mashed together.
Eat shit. People deserve to have some ownership over the product of their labor. Imagine doing all that work to learn all these skills and some rat fuck nerd tells you they have the right to alienate you from that work, making some greasy fucking argument that the multi-billion dollar program they made is doing the same thing you do.
Corporate abuse of copyright isn't an excuse for corporate abuse of people who can actually make things.
There's a limited amount of "attention" on the internet that's available in online art spaces. I theorize that artists need to maintain a public portfolio to find work, either creative work for advertising or entertainment companies or commissions or patreon subscription. You need to be out there for people to hire you if you want to make a living doing art.
AI art is disruptive and harmful to that ecosystem in a few ways. It competes with artists not only for dollars and jobs for commercial art (ie ad agency work), but more fundamentally it competes with artists for attention on the internet, which at a macro scale makes it so there's less money in the system for real artists to practice and master their skills. It's also not necessarily that the AI art is better, it's just good enough to pollute the signal to noise ratio. And as you said, it's fast. That can quickly overwhelm sites like DeviantArt or Art Station, which moves those communities and resources away from their original purpose.
Additionally, AI art models need data from the commons to function. I'm skeptical that synthetic data is good enough to use for improving these models, so theoretically, one of the main ways to improve these models is getting more and better data from humans. If the economy of these models is structured in a way that discourages artists from posting their work publicly, or working at all, then the pace of improvement decreases. I think a lot of artists are pretty pissed off that their work is being used to produce commercial software that creates substitutes for their work without their permission or knowledge.
So that's what I mean by the tragedy of the commons. The online art ecosystem is more fragile than many people think it is and ai companies are over exploiting it. The content remains but the thing you're actually trying to sample, human knowledge and skill, withers on the vine.