The problem is not in discrimination at all, but in generation of novel and correct output. (grammatically and syntactically) Discerning correct syntax and grammar has eluded all AI so far. (Even specialized translation systems are worse than babies at it.)
The whole idea of using the terms “correct” and “check” with things like art seems flawed out of the gate.
Maybe it needs to pass the Turing test only halfway. Then humans will fill in the rest and will think of it highly as art. For instance, take a computer which can create photo-realistic image - boring, not art. Take a computer that can, because of its imperfections, create crazily wrong, impressionist or cubist type of stuff - art!
What if computer produces a Picasso painting? Or impressionist one? And it does, look up "neural network dreams". These things are not pretty. But they are satisfying, because we can look at them as puzzles and fill in the details.