I wonder if it would be possible to analyse many different stories from either a single author or many authors to produce similar stories automatically. I'm imagining mimicry of the writing style as well as common words and phrases. Generally, anything auto-generated falls by the wayside when you've explored all it's possibilities and begun to predict its results. For instance, if you had an automatic story teller that used the same five or six twists over and over, you probably wouldn't use it very long after you'd learned them. But if you can "borrow" twists, plots and styles from the vast soup of the Internet, things might not get so stale so quickly.
Then I expect the problem of feedback to enter the mix. Machines generating stories from web-scraped stories and posting them online are diluting the secret sauce, so to speak. We could try tagging machine-generated fiction and filtering it out of the raw input, but not all of it will be appropriately tagged. What we really need is an objective quality filter. You could say that Google does this already with their mysterious page rank algorithm, leaning on the wisdom of crowds, but I'm thinking of something in a more isolated sense, like a personal spam filter tuned to quality stories rather than spam.
Mokalus of Borg
PS - I don't think spam filtering techniques can be applied directly to this problem.
PPS - Still, it might be worth a shot.