I am a complete neophyte to the world of AI, but am working on a sim-earth-like game where players can plant gardens and raise critters which will "compete" for ecological space. ...all this is currently in the planning stages, with little snippets of functions being tweaked. What I was planning, was to have the object data be a string of characters/bits. (possibly genetic? Unsure if it's truly AI genetics). Example fields would be Stamina, Hunger, Reproduction, Aggressiveness, Productivity, etc. Each field would range from '#000000' to '#FFFFFF'. The offspring of each critter/plant would be subject to minor mutations (bit-swapping?) and as time progresses, interact with other players critters to determine what dominates a given cell/ecosystem. Those with more aggressive tendancies will try to damage rivals around them. Those with more Hunger would of course tap the location's food resources (which is determined by the _Plant.Productivity growing there). Additionally, since the world will be populated before players even show up (It's a web game), I hope to randomly seed the locations with example plants/critters in hopes they will stabalize and give the players something to work "against" when they join the game and no other players have setup things in that particular map-location. Is this a valid way of handling such interaction, or would the calculations involved get too skewed and all approach a finite situation where all plants have XYZ values and all critters would have ABC genetics? Ideally, I envision a dynamic ecosystem, in flux due to player activity against the background processes. And is this even Truly a genetics situation, a finite-state, or something that belongs in another thread altogether?

See if you can find some discussion on the "Creatures" series. They used GAs for everything from the look of the creatures to their base qualities. You could actually get crippled ones once in a while!