An agent is learning if it improves its performance on future tasks after making observations about the world.

An agent learns, based on the gained knowledge base K with respect to some set of actions A and performance measure P after performing an action A if performance measure P is improving then knowledge base K is updated, because of the improved experience in performing that action.

Given:

A: Action

P: Performance Measure

K: knowledge base

Goal: Keep a generalized knowledge base which allows to improve performance on an action.