OK, well we wish to model behaviors, so we start at the top with abstract behaviors such as searching, moving, running, attacking, etc. Petri nets allow us to change these "modes" of behavior through the occurence of events. What we wish to do next is attach a time-driven model to this event-driven Petri net model for each of the "behavior modes" Then we got something! We'll investigate different ways we can model running, attacking, etc. and put some adaptation, learning, etc. inside. While we do these we shall try to see if we can generate some metrics to tell us how "intelligent" our system is... :) Well, we'll try. Also, note, while Petri nets seem similar to the simple state machine. They are not, since Petri nets can be used to model concurrent systems.
What would be cool is if the computer could generate new Petri net places in our behavior model as we are playing the game. :)
Anyway, got any comments... good.. bad... indifferent... ?