We have a corpus of "bad questions" with justification. Have we considered making a spam-filter-like pre-filter that can test the question for it's badness and suggest to the question writer "this question is likely a bad question for reason 'x' so please re-read the requirements and try again". Homework questions come to mind.
On some sites, the "ask question" pages will pop up just-in-time help when users seem to be getting ready to ask about certain topics. Of course, the specific topics and exact help text vary from community to community, but one example is regexes on SO:
It's not a particularly smart system, though. It's really just half a step up from keyword matching. Creating a full-blown machine learning system, on the other hand? Could be an interesting challenge, but there's no immediate urgent need for it, it would take a lot of effort and add a lot of complexity, and the amount of marginal benefit it would provide (over keyword matching or some other simpler system) is unclear.
So depending on what you're asking for, the answer might be anywhere from "post a more specific request" to "sure is a nifty idea, but not practical for the near future."