When you use one painting as the training set, then two, then three, it’s easier to track appropriations. When you have billions and trillions of well trained-images it’s not so easy. If it weren’t so then there wouldn’t be “artist signatures” in the field. Just because the appropriation process is more abstract and different doesn’t change anything. Brush strokes, color palettes, items like flowers, style, etc are extracted and trained also so you can steal Rembrandt colors with Lichtenstein style. Look, it’s a neat trick developed by computer scientists but I don’t get to worked up about defining the results. I wrote a companion piece on what an artist and art are and aren’t. An artist once place a crucifix in a jar of urine and called it Piss Christ. Not everyone thought it was art but let’s see AI do that. Something original, groundbreaking, controversial, provocative. Something that smells bad.