The video below shows and implementation of a neural network to make to balls learn to shoot one another. The idear is to test strategies, remember the results and use the ones that better reach the objective of hit the enemy and avoid being hit.
Each ball can see their adversary, the bullets and the filed of view of their enemies (that lines coming from each ball). Their movement options are to move forward, spin clock-wise, spin counter-clock-wise, shoot and adjust their field of view.
Despite the tendencious music, it is quite impressive how fast their evolution is and how easy it is to understand their learning as "intelligence", right?
If you wanna know more about the program and prepare to the eventual war against the machines click ehre.