Once I can get the framework for an expert system built (the rule-interpreter thing) something to play around with might be the three-brain concept I found in a Larry Niven story a while back: Three independant interpreters moving across the same set of not-deterministic rules, and then a consensus of two would be required. Possible problem: In a system where lots of "right" answers could be found, a two-brain consensus could be nearly impossible to reach. Perhaps a fuzzy set of the three independant answers would be better? Or a quick game-matrix analysis of the outcomes of the suggested actions?
Here endeth the bizarre thought.