I hooked up the neural networks to a breeding program in a simple 2D environment: http://vortexcortex.com/thunc/
(HTML5 canvas; Up/Dn arrows = sim speed; click a being to see what it's thinking.)
The dots give energy, and bumping shares energy with lower energy beings. After a few generations of only energy based selection pressure the neural networks begin to emerge ever more complex survival strategies. Each page load gives a new random seed, so I'm always surprised at what evolves.
The most simple strategy is moving rather than sitting still. Some go forwards, others go backwards, some just spin from place to place slowly sweeping up all dots in their path. More complex strategies are soon to follow, such as: Steering towards dots, stealing energy from others, and avoiding other beings.
Such complexity arises from so simple a process. One batch of n.nets avoided turning at first until they've dispersed enough to give everyone some elbow room. It's amazing what even these simple brains with only 32 neurons can do. Imagine the feats a powerful distributed network can accomplish...
P.S. I'm glad to be able to get back to some game programming after some time being unplugged. That was a nice warm-up, now onto some more hardcore code. #machineintelligence #artificialintelligence #ai #html5 #html5canvas #gamedev