> if lessons are learned This is the key statement. Even in reinforcement learning exploring intelligently is an important problem. There’s no point exploring

> if lessons are learned This is the key statement. Even in reinforcement learning exploring intelligently is an important problem. There’s no point exploring a new state if your world model is confident about what’s going to happen there: you won’t learn much.