#so much data used for llms is taken without permission

[cutiemarkofcain]

When I say I hate generative AI, I don't mean, I hate gen AI unless it amuses me or is convenient for me. I don't mean, I hate gen AI only when it involves the art formats I participate in and everything else is fair game. I mean, I HATE GEN AI.

[raimi]

i was going somewhere with that, but i've actually got a better angle here.

the thing is, the art-as-data means that essentially, the things llms are trained on are raw materials. art produced with ai is made out of everything that ai has been trained on, some pieces more than others, much like making a collage (even if the nature of the beast means it's more subtle than that).

and like with a collage, it's not as simple as "transformative = fair use." and fair use is what matters here, not transformative alone.

under us law, there are four factors to be considered when determining if something is fair use. lemme copy them down real quick.

the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;

the nature of the copyrighted work;

the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and

the effect of the use upon the potential market for or value of the copyrighted work.

item one is where transformativeness comes in. and yes, the use of an ai to produce new images from existing ones is transformative. that's a big factor in fair use decisions, but it's not the only one. even as part of this first criteria, we've got that section after the comma. commercial vs non-commercial. i assume you can probably see how using someone else's work as material to make your own and then selling it is kinda bad? that doesn't apply in every case, but there are a lot of people trying to make profit off of ai in some form or another.

item two. nature of the copyrighted work. this is the part that says you can't own an idea or fact. it's not nature as in how good something is or how much effort was put into it. it's about what kind of thing is being used. art is often very personal, and even when it's not it's something that this part says is allowed to be protected. the one exception i can think of would be if someone caught something historically important on camera, because that being in the public domain would be in the public interest.

item three. how much of the thing was used, and how significant it is in whatever it's being used for. this one goes in both directions a bit—the entire piece of art is being used because it's part of what the model draws on. the large amounts of training data mean that it's not usually that significant in the output, but... there's exceptions to that. it's entirely possible to tell an ai to prioritize works by a specific artist, whether by putting a name in the prompt or just by training it on that person's art in particular. in either of those cases, this part is pretty condemning. an argument can also be made that this relies on how much of the training data was taken from artists without permission in general, which also doesn't look that great.

finally, the "does this affect how it sells or how much it's worth" question. i don't think i really need to explain this one, because a lot of artists have talked about it already.

so tl;dr yes ai art is transformative, but it falls flat basically everywhere else when it comes to fair use.

some of yall will be like 'Duchamp's Fountain has been making losers mad for more than 100 years' and then advocate for entire mediums of art to be made illegal through massive expansions of IP law