Learning by trial and error from a score for actions, often delayed, where actions change what the learner sees next.

ExampleA system learning the game Breakout discovers a tunnel strategy nobody taught it.