A powerful form of learning, also known as gradient descent learning, which suggests that following a particular action, the difference between the planned outcome and the actual outcome (the error) is calculated, and the system adjusted to reduce the error next time.