YouTube also blatantly violates US Copyright Law...
But yes, reviews and parody are valid commercial applications of the Fair Use doctrine. Neither applies to AI generated words, since they don't explicitly reference to original work.
AI generation is outright copying that pretends it is not.
None of the original artists, not the companies that bought their souls, are receiving any credit or income for the images/sounds that were imputed into the machines. And the machines can no currently work without those inputs.
Citation needed.
Also, the New York Post is suing Microsoft right now so I wouldn't be so sure about text generation being any different than music or image generation. In fact, no publically available model in existence cited a full copy of a copyrighted work (part of it) in the output. Language models do this all the freaking time.
If YouTube is breaking the copyright law why Google is not sued into bankruptcy?
US list of allowed uses is not exhaustive, you can't say "yep parody, review and education are allowed but not the thing you suggest". All 4 factors of fair use need to be evaluated before you can claim that.
Do you know that under the US copyright law I can take a bunch of images, glue them together in a collage, draw some stuff over it, and exhibit the resulting creative work for money? And it all
may be the Fair Use depending on the evaluation in the court. Which depends on how creative it is, the extent of the stuff I added, what % of copyrighted works I used, and many other factors.
For image generation models. I'd say that both 1st and 3rd heavily weigh in favor of Fair Use evaluation, while 2nd and 4th weigh in other direction.
I think that heavily transformative use (Images and software that draws images are very different things) and the fact that copyrighted data is incredibly diluted (except in cases of blatant overfitting, even then it is diluted) We are talking about a ridiculous amount of images + tags compressed into mere gigabytes. It would be a very, very lossy compression even if it contained no original data but obviously, it does contain other data except heavily converted and compressed images.
It is nowhere close to being a clear example of Fair Use but there are good chances that it is. And, most definitely, it is not a clear example of not Fair Use.