Only if you consider single games; in repeated games a lower risk engagement (choose to draw or retreat rather than take a high chance of loss) will make sense.

Only if you consider single games; in repeated games a lower risk engagement (choose to draw or retreat rather than take a high chance of loss) will make sense. The flight dynamics model won't have any data points for large parts of the flight envelope, so if your robo pilot exceeds tested limits we don't know what will happen, or even if control is recoverable. Experienced pilots certainly understand dynamic performance better than state space models -- the original learning platform! People learn jets in 100 hours, it remains to be seen how much training a machine takes. Pulling 12G all the time also imposes stresses on the platforms that they were not designed for, so you'd probably want to ration that. What I'm getting at is that sustainment of operations is as big a problem as individual engagements. You can loose a whole campaign by responding too heavily (or weakly!) to feints and probing.