supervised vs reinforcement learning

In supervised learning, a teacher provides the right answer for every example: this image is a cat, that one a dog. The model just learns to copy the labels. In reinforcement learning there is no answer key—only a reward saying how good an outcome was, never which action was correct. The agent must discover good actions itself, by trying them.

Two more gaps follow. Supervised data is usually independent and fixed, whereas an RL agent's data is a sequence it generates, and its own choices decide what it sees next—act timidly and you never learn what bold moves do. And supervised feedback is instructive ("the answer was X"), while reward is merely evaluative ("that scored 3"). These differences are why RL needs exploration, credit assignment, and a long-term view.