I've always been fascinated by MegaHAL. When learning haskell, a MegaHAL clone seemed like an ideal project. I think I managed to capture the essence of MegaHAL.
The way MicroGerty works is much like MegaHAL does. Its "brain" is a markov model of order 4 which learns which words are likely to follow others. Then, it uses this brain to chain words together, based on key words taken from the user's response.