Would it make sense to compare the output images to their nearest neighbors in the training set in order to establish some metric of "originality"? Sorry if that is a basic question, I haven't read much on image synthesis with neural nets.
Yes, and they don't usually look too much alike (see the BigGAN appendix nearest-neighbors for example of doing this) but some people argue this isn't a good enough check because the GAN could just be shuffling around patches of textures or something, which is a little goal-post-move-y for my taste.
I think "humans also do X" is fallacious. Humans also draw paintings like The Black Square by Malevich (and he wasn't the first to do it) and there's a poem consisting a single letter "i" by John Curry and another consisting of one word "lighght". Does this mean a printer capable of printing random letters is a poet and one that can produce a black page is an artist?
It's true it is hard to define what art is. But I'm pretty sure /dev/random isn't an artist, even if a human artist can produce random things.
> Does this mean a printer capable of printing random letters is a poet and one that can produce a black page is an artist?
You're asking that rhetorically, assuming people will just agree there's a difference because a human artist did it and a computer doing it isn't an artist, but you know a lot of people would be happy to take one of those forks.
Well, I guess for these people general AI already exists, or at least AI capable of producing art. Just take a random pattern, imbue it with whatever meaning you please, and treat it as artistic intelligence. With this approach, any innovation (at least as far as producing art goes) does not carry any value beyond novelty value - if /dev/random has everything one ever would need in art, then what's the difference between any new sophisticated model and just another dump of output of /dev/random?
True, but humans are generally able to start from scratch and make something. It doesn’t seem like that’s entirely what is happening here, but to be honest, it would be hard to say given the quality of the output.
I really enjoyed the videos that transition through the different options for Holo.
> From this story, though it was the shortest and the most simple too of all those I have recorded in this book, I feel that I learned several things of some importance. First of all, how much of our speech, which we think freshly minted in our own mouths, consists of set locutions. The Ascian seemed to speak only in sentences he had learned by rote, though until he used each for the first time we had never heard them. Foila seemed to speak as women commonly do, and if I had been asked whether she employed such tags, I would have said that she did not---but how often one might have predicted the ends of her sentences from their beginnings.
You could look through the various sets of samples and the more extreme psi examples and the interpolation videos and perhaps get a better idea of the extent to which the StyleGAN 'understands' faces and is being creative: https://twitter.com/gwern/status/1093701790971953152