PART 2 · MODULE 4Learning by doing
Reinforcement Learning
Learning through trial and error
Loading module...
Learning through trial and error
Reinforcement learning (RL) is the trial-and-error approach. There are no labeled examples. Instead, an agent takes actions in an environment, receives rewards or penalties, and gradually learns which actions lead to the best outcomes.
Think of training a dog. You don't explain the rules in English. You give a treat when the dog sits on command and withhold it when the dog ignores you. Over time, the dog learns what behavior earns rewards.
Supervised learning is studying with an answer key. Unsupervised learning is exploring without a map. Reinforcement learning is learning to ride a bike by falling down and getting back up until you stop falling.
Every RL system follows the same cycle: the agent observes the current state, takes an action, receives a reward, and updates its strategy. Then it repeats.
THE REINFORCEMENT LEARNING LOOP
The learner or decision-maker (the AI itself)
The world the agent interacts with (a game, a road, a market)
A snapshot of the current situation (where am I now?)
A choice the agent can make (move left, buy, wait)
Feedback signal — positive for good moves, negative for bad ones
The agent's strategy for choosing actions in each state
One of the most interesting dilemmas in RL. Should the agent stick with what it knows works (exploitation) or try something new that might be even better (exploration)?
Going to your favorite restaurant every Friday. You know it's good. Safe choice, reliable reward.
Trying a brand-new restaurant. It might be amazing or terrible, but you'll never know unless you try.
Reinforcement learning shines when there's no "right answer" dataset, just a goal to achieve. Instead of memorizing examples, the agent discovers strategies through experience. It's the closest thing to how humans learn many skills in real life.