Learning without answers: how AI finds groups
Big idea: even when nobody gives the right answers, an AI can find order by noticing what is alike and what is different.
The story in short
In Dadi’s old wooden sandook (trunk), the children found a heap of little things with no names, and Dadi didn’t have her spectacles to tell them what was what. Mitthu sorted by shape: round or long, does it roll, does it have holes, is it heavy? Bunny sorted by colour. Their piles looked completely different, and Dadi said both were right. One flat, heavy thing fit in no pile at all. Dadi rubbed it on her saree and her eyes filled with tears: it was her father’s old one-anna coin, lost thirty years ago. Next morning she named the piles: kanche (glass marbles), buttons, imli seeds, kaudi (cowrie shells) and bits of glass bangles.
The idea
- Unlabelled data: examples that come with no right answers attached.
- Unsupervised learning: finding patterns on your own, without being told the answers.
- Clustering: putting similar things into groups.
- Choosing features: different clues give different groups. Neither is “the” answer.
- Anomaly (outlier): something very different from the rest. Worth a closer look.
- Human interpretation: people decide what the groups mean.
- Compare with supervised learning (Chapter 1’s mangoes), where every example came with an answer.
Where you see it
Shopping apps group customers who buy similar things. Banks flag a payment that looks nothing like your usual ones: that is anomaly detection, and it can stop fraud.
Remember: when something fits nowhere, look closer. It may be a mistake, a fraud, or a treasure.
Check yourself
Your bank texts you about a payment that is very different from your usual spending. What is the bank's AI doing?
What is the difference between supervised and unsupervised learning?
Mitthu sorted the heap by shape and Bunny sorted it by colour. Who was right?
More for you: 🧒 Kids · 🧑🏫 Parents and teachers