What you're describing is a Markov chain. It could work to implement something like that, but I'd suggest that it should be associated with lower difficulty settings. Personally I like it how it is.