How about a difficulty rating for questions?
How about being able to sort questions by the OPs ranking? That way I could for example find easy questions to answer or difficult ones to chanllenge myself...
I think it would be difficult to automate this:
Consider a user who has earned a lot of rep on tag
<T> and then asks a question on tag
<T'> are not related, the OP's rep earned from
<T> has no effect in automating the classification of this question as easy/difficult.
So then we would consider using only the rep of the OP earned from questions tagged with
<T'>. Well, questions can have multiple tags, so we might need to take the average (or some other function) of the difficulty rating generated for each of those tags. Then, we would need to recalculate this difficulty every time a re-tag is performed.
Note however, that this approach does not guard against a new or inactive expert. For example, I could be Guido van Rossum (the creator of the python programming language) and just be not very active on SO (therefore, I have very low rep on all tags). Does this mean that if I ask a python question, that it should be tagged as a beginner? Absolutely not. Also, what should we do when the OP's rep increases over time? Should we go back and re-classify all their questions?
So the next thing that I can think of is that the difficulty of questions increases with the the number of views the questions get, while remaining unanswered. That is, if more people have seen the question and it is still unanswered, then all those people who have seen the question are unable to answer it. Therefore, this must be a difficult question. Note that there are assumptions built into this method: no other activity (flag, vote to close, etc) is performed on this question. But then, what happens if it's an overly simple question that nobody wants to answer? What if it's a bad question and we're waiting for clarification from the OP?
Ultimately, I think that with some (or a lot of) heuristics and some AI classification methods, this might be doable, but not without a high computational cost, which is why I don't think it will be implemented. Still, I think this is an interesting idea