One might say that Picasso's Bull is a human equivalent of this: he "evolved" a sequence of images and ended up with something that has very few features of a b

One might say that Picasso's Bull is a human equivalent of this: he "evolved" a sequence of images and ended up with something that has very few features of a bull, but nevertheless gets recognized by humans as such. Then again, unlike the neural networks in the paper, humans would be capable of classifying abstract images into a separate category if asked.