Reinforcement learning is a type of machine learning in which an agent learns by acting in an environment and receiving feedback in the form of reward. Instead of learning only from fixed labeled examples, the system improves by trying actions, observing consequences, and adjusting behavior to do better over time.
Why RL Is Different
In supervised learning, the correct answer is already known for each example. In reinforcement learning, the system must discover which actions lead to better long-term outcomes. This makes RL especially useful for sequential decision-making problems such as robotics, game playing, control, resource allocation, and adaptive optimization.
A core challenge in RL is balancing exploration and exploitation. The agent has to use what it already knows while still trying new actions that may lead to better outcomes. That trade-off is one reason RL can be both powerful and difficult to tune in practice.
Where RL Shows Up Today
Reinforcement learning has been used in robotics, games, recommendation systems, and parts of model post-training. In language model workflows, ideas related to RL are one piece of methods such as RLHF, where human feedback helps steer behavior after pretraining.
RL is not the right tool for every problem, but it is an important concept because many real-world tasks depend on decisions that unfold over time rather than on one-step predictions alone.
Related concepts: AI Agent, Machine Learning, RLHF, and Responsible AI.