AI learning to play games

Question

I have been fascinated by Ai for a long time now and then I found and article here by Matthew Cochran. That helped me a lot in understanding an AI and how it works.

 

I set a personal goal where I would like to have an AI that I can "set it in front" of a game and with time it would learn how to play the basics of a game, like if I threw it in front of Super Mario it could complete levels (not necessary all levels) or if I started a mission in The Crew it could do the race without driving too much off road.

 

Sure this aren't an easy task but I don't get better at it of I don't challenge myself.

 

How I could see this work is by taking a section of the game convert it to the three colors red, blue, green. Then feed the colors (as an array) to the input neurons they would then pulse it to the Hidden layers where there would be 1.5x more neurons then there are input neurons (if we split the screen in 4 so we had 4 input neurons we would have 6 neurons in the hidden layer) and in the last layer the output layer there would be a neuron for each control in the game as needed to play it with basics.

 

I am for sure forgeting an importent part of this, that is why I'm asking here.

 

Thanks in advance.