Nobody hands you the answer

Think about learning to ride a bicycle.

Nobody gave you the correct answer. Nobody could. There is no sheet of paper listing the exact angle to lean at each moment, and even if someone wrote one, reading it would not help you.

What you got instead was consequences. You leaned too far and the ground arrived. You corrected too hard and wobbled the other way. You found a stretch where nothing bad happened, and something in you noted how that had felt.

The fall was not a label saying which handlebar angle had been correct. It was a consequence that told you the attempt had gone badly.

This is an analogy, not a claim that a person learns balance through the same algorithm as a machine. The shared shape is that there is no answer key to compare against. An outcome arrives after the action and gives evidence about how the attempt went.

A robot-learning task can have this shape too. A designer may not supply the correct finger position for every unfamiliar object, but the system can measure whether the object was lifted or dropped.