How to Build a Simulator AI?

When you hook up an agent is going to find loopholes. We don't think optimally at all, but we've been doing okay, I guess. And the funny thing about that is that sometimes, or there's like most of the time, and it just happened a couple of days ago to me too, you don't think of things and those AIs find them. But here's the very, very interesting point in reinforcement is, aren't you distracting your AI from the main goal, which is destroy that other aircraft over there by saying, Oh, and by the way, fly smoothly. Like I don't care. Are you going to destroy that guy or not? Right?