Okay, so I was bored in math class. I came up with this idea for AI. What if you had 2 picaxes. One controlling sensors and locomotion, and one that had certain objectives (battery level stays above critical, don’t let your gyro tilt too far, etc) for basic survival implanted in its code. Now what if, as the robot zooms around the yard, it flips. The gyro objective has been violated. the robot rights itself, and the second, objective picaxe starts comprising a new code to burn into the first one. It does so, and then the robot is off again, with a new code. The theory is that over time, the robot will get better and better at achieving its objectives, by figuring out which code works better: the original or the revised one. if it is the original, the robot will revert to the original and try a new code. If the revised is better, it sets the revised as the original and carries on. Eventually, it will achieve sentience, given enough objectives and time. Skynet, here I come!
Ok so you want learning.
Ok so you want learning. I’ll use my softwre background to poke you into posible reading material:
neural networks
multi agent systems
That will keep you reading for a while so I’ll stop the list already =) neural nets should be bang on the type of thing you’re thinking and hopefully a neuron is simple enough to replicate even in a circuit.