What supplies the feedback

Put the two arrangements side by side. The difference is entirely in where the feedback comes from.

Supervised learning. The feedback is a target supplied with a training example. The picture came with school bus written on it. The machine guesses, the guess is compared with that target, and the gap guides the correction. A dataset may have missing, delayed, noisy, or disputed labels, but the training arrangement still treats the supplied target as the answer for that example.

Reinforcement learning. The feedback is a score, produced by the world after the fact. Nobody says what should have happened. The machine finds out that something went well or badly, sometimes long after the action that caused it, and has to work out for itself which of its choices deserved the credit.

One tells you the right answer. The other tells you how it went.

The two arrangements can also be combined. A system might first learn from human demonstrations, then improve against a reward. Reinforcement learning is useful when outcomes provide information that a fixed answer key does not.