what is reinforcement learning

Imagine teaching a dog a trick. You don't hand it a manual; you let it try things, and when it does something right you give a treat. Over time the dog learns which actions earn treats. Reinforcement learning (RL) trains software the same way: an agent acts in some world, receives a number called reward, and gradually discovers behavior that earns more reward over the long run.

More precisely, RL is the branch of machine learning where an agent learns a policy—a rule for choosing actions from situations—purely by interacting and observing consequences, not from a dataset of correct answers. The aim is to maximize expected cumulative reward, not the immediate payoff. This makes RL the natural framework for problems that are a sequence of decisions whose effects unfold over time: games, robotics, control, and trained language-model assistants.