Reinforcement Learning 101

Reinforcement learning now sits behind the models and agents senior leaders are asked to approve.

Reinforcement learning is the branch of machine learning in which an agent interacts with an environment, takes actions, receives rewards or penalties, and gradually learns a policy that maximizes long-term reward. The vocabulary is compact and worth memorizing: agent, environment, state, action, reward, policy, value, and episode. The agent is the learner and decision-maker; everything outside it is the environment. A state is what the agent sees at a given moment, an action is what it does next, and the environment responds with a new state and a reward signal.

This matters operationally because reinforcement learning now sits behind the systems senior leaders are asked to approve. Modern language models are refined through reinforcement learning from human feedback, which trains a reward model on human rankings and then optimizes the language model against it. Humanoid robots, game-playing systems that beat world champions, and production LLM assistants all share the same learning loop.

After this module you will be able to interrogate that loop. You will know that policy is the strategy for selecting actions and value is the expected long-term reward of a state. You will understand the tension between exploration, trying unfamiliar actions to learn, and exploitation, using the best-known action, and why balancing the two is central to effective learning. Most importantly, you will know to ask what the reward function actually optimizes, because unintended behavior usually traces back to a badly defined reward.

The failure modes cluster around the reward signal and the deployment path. Sparse rewards, delayed rewards, and reward hacking each require careful design of the signal itself. Teams optimize a proxy metric that diverges from the real objective. They let an agent learn in production without an off-switch and a replay log. They ignore the compute cost of training until it has consumed the budget. Treat these projects as multi-month efforts with safety review at each stage, and keep simulation environments separate from production.

In one paragraph

Reinforcement learning is a branch of machine learning in which an agent interacts with an environment, takes actions, receives rewards or penalties, and gradually learns a policy that maximizes long-term reward. The vocabulary is compact: agent, environment, state, action, reward, policy, value, and episode. Modern large language models are refined through reinforcement learning from human feedback, which teaches them to prefer responses humans rate higher. Humanoid robots, game-playing systems that beat world champions, and production LLM assistants all share this learning loop.

This module is part of the Foundations track in the free AI Learning Hub. Source material: Reinforcement Learning 101.