#AiArt #AdversarialMachineLearning #AdversarialAi

I can't *find* it now, but I read this thing where someone analyzed several training images of a certain subject, found pixels they had in common, and added that almost-invisible pattern to unrelated images to fool the AI image classifier into thinking they were that subject.

What if a group of artists with similar styles added a watermark to their work that'd be noticable to AI software but not to human viewers?

1/3