Learning is a weird word to use for software. It implies something active, something that changes over time, something that gets better with experience. That's not how most software works. But it is, more or less, how AI works — and the interesting question is: better with experience doing what, exactly?
There are three fundamental answers to that question, and they correspond to three different ways AI systems can be trained.
The first is supervised learning. You give the system a large set of examples, each one labeled with the correct answer. Thousands of emails marked "spam" or "not spam." Thousands of images labeled "cat" or "not cat." The system learns to associate inputs with outputs by seeing enough examples that it can generalize to new ones. This is the most common approach in production AI, and it works well whenever you have a clear right answer and enough labeled data to train on.
The second is unsupervised learning. Here, you give the system data but no labels — no right answers, no guidance about what to look for. The system finds structure on its own: clusters, patterns, relationships that weren't explicitly defined. This is useful when you don't know in advance what you're looking for, or when labeling data at scale would be impractical. Customer segmentation, anomaly detection, and topic modeling are common applications.
The third is reinforcement learning. This one works differently from both. Instead of learning from a fixed dataset, an agent learns by taking actions in an environment and receiving feedback — rewards for good outcomes, penalties for bad ones. Over many iterations, it figures out which actions tend to produce which results. It's how AI learned to play chess and Go at superhuman levels, and it's a key component of how large language models get tuned to be more helpful and less harmful after initial training.
These three approaches aren't mutually exclusive. Real AI systems often combine them. But understanding each one separately is the foundation for understanding how any particular AI system was built — and why it behaves the way it does.
The articles that follow go deeper on each approach, starting with supervised learning and unsupervised learning, then reinforcement learning.
