A one-pass attempt probably wouldn't be that impressive, but i wonder about having games run split-second "play tests" of spontaneously generated content using

A one-pass attempt probably wouldn't be that impressive, but i wonder about having games run split-second "play tests" of spontaneously generated content using some kind of adversarial network to estimate the player's reaction. Then questlines could be iteratively fine-tuned while in progress, more like what a human DM does.