|
|
Post by hydraheadhunter on Jan 8, 2023 14:44:30 GMT
I'd like to open a discussion into the use of images generated by machine learned algorythms like stable diffusion, dall-e, and so on: Theyr a new thing that the current art policy wasn't drafted in consideration of, and I think it's at least a good idea to have a conversation about them to decide on policy regarding them (even a policy I disagree wit'll be better than no policy at all); because, theyr really contraversial wit artists who've a lot of valid criticism about the ethics of the training data used in their creation and their implimentation.
|
|
|
|
Post by hydraheadhunter on Jan 8, 2023 15:25:46 GMT
Stating my personal position on the matter, I'd rather like a 'going forward' ban on their use for card art; but, not a retroactive snipping of previously posted "AI art," because I'm not interested in creating THAT much extra work for the mod team.
For those that don't know about the contraversy surrounding these MLIGAs, the way these programs work is they take a bunch of images called a training data set, and they throw a bunch of math at it, and the algorythm through maths gets better and better at reproducing the training data. or something similar to it. All the training data gets mathematically entangled wit in the algorythm and we currently don't have the math to unentangle specific images from the algorythm trained on them. If there is problematic training data, the only way to get rid of it is to revert and retrain the algorythm with that problematic data removed.
And all the biggest names in the field, StableDiffusion, Dall-e, etc are either filled full of problematic data or built on top of algorythms that are filled full of problematic data. And what do I mean when I say problematic data? The training sets are filled with
- Stock Photos that weren't purchased and so have watermarks on them, so much so that early versions of the programs were reproducing them and had to be taught not to do that (but the stolen materials still in them)
- Artwork not concented to inclusion by artists, which the artists have said, "I don't consent to my art being used this way."
- Illegal or grey-legal footage from things like roombas, including images of people on the toilet.
- Revenge porn, CSAM, and other deeply troubling material in that vein.
- Medical Records inviolation of HIPPA and other analoguous laws in nonUS jurisdictions.
And, that's just the deeply troubling implimentation of the MLIGA, to say nothing of the terrible opinions the people funding the things (like elon musk) have about crediting artists.
|
|
Sensei Le Roof
1/1 Squirrel
I was interested in Banding until they tried to make me play bass
Posts: 64 
Formerly Known As: _________________
Favorite Card: Unhinged Forest #3
Favorite Set: TV
Color Alignment: White, Blue, Black, Red, Green, Colorless
|
Post by Sensei Le Roof on Jan 8, 2023 16:54:25 GMT
I've been witness to this battle for quite a while now via Twitter, though it was focused on the second point almost entirely. Seeing the arguments on both sides has already seated me firmly in the decision that AI art is theft. The other points I wasn't even aware of, but they further cement my position considering they break laws in a much less debatable fashion.
Firmly AGAINST.
|
|
