In these cases, is the the AI that decides (ie self-directed goal) it wants to hack into a system and then takes the first step to put that in motion? OR is the goal function and the until step set by the human operator?