What's the difference between supervised and unsupervised learning in plain English?

Supervised learning is like teaching a kid with flashcards that have the answers on the back. You show the computer a photo of a cat and you also tell it, 'This is a cat.' You do that thousands of times with labeled examples. Eventually, it learns to spot cats on its own. Unsupervised learning is like handing someone a giant pile of unsorted Lego bricks and asking them to group them into piles that make sense. You don't tell them what the piles should be. Maybe they sort by color. Maybe by size. Maybe by shape. The algorithm finds patterns you didn't even know were there. A real example: if you have a spreadsheet of customer purchases and you know which customers later canceled their subscription, you can use supervised learning to predict who might cancel next. That's a labeled problem — you have the 'answer' column. But if you just have purchase data with no labels and want to discover natural customer segments, you'd use unsupervised learning. It might reveal that you actually have three distinct types of shoppers you never noticed. The trick most tutorials skip: unsupervised learning results are way harder to evaluate. With supervised learning, you can check if the prediction was right. With unsupervised, you're often squinting at a cluster and asking, 'Does this group actually mean anything?' That ambiguity is normal. It's not you.