@metin The network needs to know how to 'fill in the blanks' though, so I'd consider this generative AI as well. So the question is: what has this been trained on?
@BartV I'd rather nuance this to restorative AI, as the trained models are only used to remove artifacts and other issues in existing images, and are beneficial and free to all image creators, while generative AI blatantly steals distinct artists images and styles, and makes them generative to paying users of an OS or other platform as a "feature", to the disadvantage of all artists.
@metin I agree there's a difference, but what if this was a paid service?
You're skipping over my main question: these networks can't remove artifacts or upscale without having contextual knowledge to fill in the blanks. What are they trained on and did they get permission for that?
@BartV As far as I know, the Open Model Database is trained on freely available image datasets that have been released for AI training, probably CC0 textures, images, et cetera.
@metin@BartV All training is fair use, regardless of source. It is the very definition of transformational. No copy of the original exists in the model. It is functionally no different from an artist visiting a museum and observing art. The difference is in scale and volume. You can argue that the scale/volume makes this qualitatively different, but that's a philosophical question that I don't think holds up to scrutiny.
@wagesj45@BartV I don't think "fair use" covers billionaire companies parasitizing hard-working artists, sucking up their carefully developed styles, and serving them to their paying customers as a feature, while the sources of that feature lose customers, motivation and income.
@metin@BartV That is pretty much exactly what fair use of the cultural commons means. Hard work and exploitation separate concepts. I don't deny that artists work hard. I don't deny that billion dollar companies are exploitative. But training neutral networks is not those things. Entering your work into the cultural commons comes with a social contact. If an artist rejects that, the only option is to not enter it.
@wagesj45 What you seem to say is: "Once you've exposed your work, you lose your intellectual property," 🙂
More and more jobs will become obsolete: image creation, coding, and with AI-powered robots also the physical jobs.
I guess it would be OK as long as there would be compensation, such as a universal basic income. But that's not how capitalism works, so we'll see how it goes in the next decades.
@metin Sounds like you have more beef with capitalism than a philosophical difference about the transformational nature of training. But in the same breath you defend it by clinging to the idea of owning your ideas. So I dunno. Hard to break out of that mode of thinking, I suppose.
@wagesj45 Well, owning your ideas is the main way to make a living as a creative in the capitalist system. I don't like the system at all, but I need to pay my mortgage. Adapting doesn't equal supporting.
What do you do for a living? Coding? What if your boss or clients tell you "We don't need you anymore, because through the transformational nature of AI your skills aren't necessary anymore. Good luck paying your rent!"
Will you respond: "No problem, I totally support that" ?
@metin I am a coder, and I use AI. I'm also an artist, having recorded music and played live shows and festivals in the past, as well as doing 3D work and professional videography today. It isn't like I don't understand that perspective.
I'm not saying that we should defend the capitalist hellscape. I'm saying that you don't defeat capitalism with more capitalism. What's right and correct doesn't change according to my personal situation.
@metin To address your direct question, I don't like it when anyone is discarded in our system. Regardless of the reason. UBI would indeed ease this burden, and I think it's a good and necessary thing. I also believe that we should evolve beyond our system and think beyond it. We can and should think beyond capitalism even as we live under its thumb.
Add comment