Reinforcement Learning
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
A machine learning approach where an agent learns by trial and error, acting in an environment and adjusting its behavior based on rewards and penalties.
What is Reinforcement Learning?
Reinforcement learning (RL) is a machine learning approach in which an agent learns by trial and error: it acts in an environment, receives rewards or penalties, and adjusts its behavior to maximize cumulative reward.
Nobody supplies correct answers. Unlike supervised learning, which learns from labeled examples, RL learns from consequences – and a reward often arrives many moves after the action that earned it, which is what makes the problem hard.
How Reinforcement Learning Works
- Agent: the learner and decision-maker.
- Environment: everything the agent interacts with; it answers each action with a new state and a reward.
- Actions: the moves available to the agent in a given state.
- Rewards: the numeric feedback that scores each outcome.
- Policy: the agent’s current strategy, mapping each state to the action it will take there.
- Value function: the agent’s estimate of how much future reward a state or action leads to.
Training is the loop that improves the strategy: act, observe the reward, update, repeat – millions of times.
Applications of Reinforcement Learning
- Game playing: RL trains agents that master chess, Go, and video games through self-play; our post on training bots to beat the best covers how.
- Robotics: robots learn to walk, grasp, and navigate by being rewarded for progress rather than programmed move by move.
- Fine-tuning language models: reinforcement learning from human feedback (RLHF) teaches chat assistants which responses people prefer.
Example of Reinforcement Learning
AlphaGo, the system that defeated Go world champion Lee Sedol in 2016, was trained largely through reinforcement learning. It played millions of games against itself, treating a win as reward and a loss as penalty, and refined its strategy game by game.
Self-play produced moves that surprised professional players, most famously move 37 of the second game, because the system was never limited to imitating human play. It searched for whatever maximized its chance of winning, and reward alone shaped the result.
Related AI terms: Q-learning · AI Agent · Machine Learning · Supervised Learning
Did you like the Reinforcement Learning gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About