The question is whether humans can control the systems they build and whether we can trust humans to do so. This is related to the concept of reward function optimization and the value function of humans versus that of RL agents.