Glossary term

Reinforcement learning

Learning by trial and error, where an agent takes actions and is rewarded or penalized for the results.

How it works

An agent takes actions in an environment and receives rewards or penalties. Over many tries it learns a policy: a strategy that earns the most reward in the long run.

It has produced game-playing systems and robot control, and a form of it, learning from human feedback, is used to tune chatbots.

The reinforcement learning loop

In code

The agent-environment loop (Gymnasium API)

obs, info = env.reset()
for _ in range(1000):
    action = agent.act(obs)                       # choose action
    obs, reward, done, truncated, info = env.step(action)
    agent.learn(reward)                           # learn from reward
    if done or truncated:
        obs, info = env.reset()

Why it matters

Reinforcement learning handles decisions over time, where the right action depends on what happens next.

Related terms

Policy Reward function RLHF

Recent signals mentioning it

ResearchOct 12

StartLux-Decision: AutoThink Decision Models with Reinforcement Learning for Calibrated Deliberative Decisions

Upvotes55Authors6