Reinforcement Learning (RL) is a type of machine learning where an agent learns by trial and error, taking actions in an environment and receiving rewards or penalties, gradually learning the behaviour that maximises reward.
What it means in plain English
Reinforcement learning is closer to how animals learn than to studying labelled examples. The agent tries actions, sees the results, and is “rewarded” for good outcomes and “penalised” for bad ones. Over many attempts, it learns a strategy that earns the most reward. There is no answer key — only feedback on how well it is doing.
It excels at problems involving a sequence of decisions over time, where each choice affects future options.
A simple example
An AI learning to play a game uses reinforcement learning: it makes moves, wins or loses, and over thousands of games learns which strategies lead to victory — without ever being shown the “correct” move for each situation.
Why it matters
Reinforcement learning powers game-playing AI, robotics, and control systems. It is also central to how modern chatbots are refined: human feedback is turned into rewards that shape the model toward helpful, safe responses.
Related terms
- RLHF — reinforcement learning from human feedback, used to train assistants.
- Supervised Learning — learning from labelled answers instead of rewards.
- Model — the agent that learns the strategy.
Frequently asked questions
What is reinforcement learning?
It is a type of machine learning where an agent learns by trial and error, taking actions in an environment and receiving rewards or penalties, to maximise total reward.
Where is reinforcement learning used?
In game-playing AI, robotics, recommendation, and — via RLHF — aligning large language models to human preferences.
