A closed-loop control system in AI does not automatically guarantee an exponential "takeoff" in capability