Comment by hb-robo

Layman question here since this isn't my field: how do you achieve success on closed-system tasks without supervision? Surely at some point along the way, the system must understand whether their answers and reasoning are correct.