Reinforcement learning(RL)
A way of training AI by trial and error, rewarding good outcomes and penalising bad ones so the system gradually learns which actions work best. It is how AI masters games, controls robots, and, increasingly, learns to reason.
Reinforcement learning takes its cue from how animals learn. Rather than being shown the correct answer for every situation, the system tries things, sees what happens, and receives a reward when the outcome is good or a penalty when it is not. Over many attempts it works out a strategy that earns the most reward. There is no teacher marking each step, only the consequences of its own actions.
How reinforcement learning works
The pieces have standard names. An agent (the AI) takes an action in an environment, which responds with a new situation (the state) and a reward signal. The agent’s goal is to learn a policy — a rule for choosing actions — that maximises its total reward over time, not just the next step. The hard part is that rewards are often delayed: a move in a game of Go might only prove good or bad twenty moves later, so the agent has to learn to credit the right earlier decisions.
How it differs from ordinary machine learning
Most familiar machine learning is supervised: you show the model thousands of labelled examples (“this image is a cat”) and it learns to copy the labels. Reinforcement learning has no labels. It suits problems where the right move depends on a long chain of decisions rather than a single correct answer, and where the only feedback is how things turn out. That makes it powerful, but also finicky: a poorly designed reward can teach the agent to do exactly the wrong thing, very effectively.
Where it is used
Reinforcement learning is how Google DeepMind’s systems learned to play Go and chess beyond human level, and it underpins much of the work on robotics and self-driving cars — including British efforts like Wayve, whose vehicles learn driving behaviour from experience rather than hand-written rules. More recently it has become central to language models: it is the mechanism behind reinforcement learning from human feedback (RLHF), which shapes how a model behaves, and behind the newer reasoning models that are rewarded for working carefully through a problem before answering.