A good starting point might be Perceptron, then Hopfield and then FeedForward. And in the mean time you will learn different training and error calculation algorithm like- hebb's rule, delta rule and after that, Backpropagation.

Good Luck.

Maybe you should get some actual experience with ANNs before you get too excited about their possibilities. Start by writing ANNs that learn basic things, like a linear combination of things to predict another (an ANN with a single neuron, which is like linear regression), then XOR... Then you can try to reproduce the results of other people with things like character recognition. By now you'll realize that backpropagation has lots of issues and doesn't get you too far. You can then learn something about support vector machines, deep belief networks, etc. Then go back and read your own post above, and see if it still makes sense.

Why do I feel like this is another example of "magic word" syndrome? That the OP is fascinated with "ANN" and keeps saying it over and over because he thinks it sounds cool despite having very little knowledge about what it really does or the (more suitable?) alternatives?

what about natural motion? that exists does it not?

Not sure what you are getting at there.

That's not the point. Your argument is very much like, "well, you can use a hammer on a nail, so why can't you use a hammer on a screw as well?" Right tool for the job, sir.

Is your goal to accomplish a gameplay trick or is your goal to use an ANN to do... something?