For decades, the dominant paradigm in Reinforcement Learning (RL) was 'Model-Free.' In this approach, an AI agent is essentially a reactive machine. It observes a state, takes an action, and receives a reward.1 If the action was good, it statistically reinforces that behavior. It does not understand why the action was good, nor does it possess an internal map of the environment. It is comparable to a reflex: touching a hot stove and pulling away requires no understanding of thermodynamics, only a nervous system that registers pain.
learn more
