Reward prediction error in learning-related behaviors