Reinforcement Learning

Definitions

1. A machine learning paradigm where an agent learns to make decisions by performing actions in an environment and receiving rewards or penalties. The technique behind AlphaGo, robotics control, and RLHF (the process that makes ChatGPT polite).

   In plain English: Teaching an AI by giving it rewards for good behavior and penalties for bad behavior — like training a dog with treats, except the dog is a neural network.

Etymology

1950s

Arthur Samuel's checkers program at IBM learns by playing against itself — early reinforcement learning before the name exists

1989

Chris Watkins formalizes Q-learning, establishing the mathematical foundation of modern reinforcement learning

2016

DeepMind's AlphaGo defeats world champion Lee Sedol using deep reinforcement learning, stunning the AI community

Related Terms

• Fine-Tuning
• Neural Network
• RLHF
• Machine Learning