In plain English
Reinforcement learning is a type of machine learning where an agent learns to make decisions by trying actions and receiving rewards or penalties, gradually favouring behaviour that earns more reward. There are no labelled right answers, just feedback on outcomes. It is commonly used in robotics, game-playing, and recommendation, and it underpins techniques that align language models to human preferences.
A worked example
A game-playing system that wins more often by being rewarded for good moves and penalised for losing ones is learning through reinforcement.
Common confusion
Reinforcement learning differs from supervised learning. It learns from rewards on outcomes, not from a dataset of correct input-output pairs.