[Jubloz]

I haven't had time to delve too deep into this, but I read a short article in the new Discover Magazine (Jan '08 issue) about a project aimed at developing AI through evolving robots. Basically, the researchers had a group of robots, as well as "poison" and "food" to encourage a sort of natural selection. The "poison" drained the robots batteries, effectively killing them, whereas the "food" boosted their batteries. The programs of the robots who took the "poison" and were drained of batteries were scraped, whereas the survivors' programs were entered into other robots, some of which underwent a "random mutation" in their neural circuitry of 30 programmed "genes". By the 50th generation, the robots were actually communicating with each other; they were using their lights to indicate to other robots which batteries were poison and which were food. More amazingly--and this is the part that must've freaked the researchers out--one of the robot groups actually learned to lie. This particular robot group would tell other robots that poison was actually food, thus draining the batteries of the competition and allowing them exclusive access to the food. These robots had quite literally adapted to trick other robots for their own gain.I imagine that, if we see AI developed to any level that's substantial, it will be through a system like this that mimics evolution.