Interesting. One of the common complaints levelled against some machine learning techniques (artificial neural networks and genetic algorithms spring to mind) is that their solutions are usually so bizarre that it's impossible to learn anything about the problem domain from them. You've gone the other way, which is pretty cool.