Reinforcement learning is based on occasional rewards. Reinforcement learning agent doesn’t have the exact output for given inputs, but it accepts feedback on the desirability of the outputs. This feedback can be provided by the environment or the agent itself.

Feedback generally occurs after a sequence of actions, so there can be a delay in getting respective improved action immediately.

Reinforcement learning agent knows that the results are higher, but it doesn’t know what action caused the results.