Reinforcement Learning: The Trial and Error Method

Reinforcement learning (RL) feels the closest to how we actually learn as humans. It’s all about an agent interacting with an environment, getting rewards or punishments, and trying to maximize its score. I tried training a simple agent to play a game once. For the first few hours, it just sat there doing nothing. Then it slowly started moving. Then it crashed into walls. But eventually—after hundreds of thousands of tries—it got good. Watching that progress is incredibly satisfying, though it does test your patience.