Lesson 3 · AI Around Us
Data, patterns and training — and why AI can learn the wrong thing.
When AI recognises your face or writes a poem, it can feel like magic. It isn't.
Behind every AI are four down-to-earth ideas: data, patterns, training and prediction.
Imagine we train an AI to recognise mangoes — but every photo we show it has the mango hanging on a tree.
The AI might learn the wrong pattern: "mango = green leaves + branches." Then we show it a ripe mango on a plate… and it says "not a mango!"
The AI wasn't silly. It simply learned from limited data. An AI can only know what its examples taught it.
Do exactly what an AI does when it learns. We'll use one running example — teaching a machine to spot a samosa 🥟 — so you can see each step happen for real. Then try it on something of your own.
Gather 6–10 examples of the SAME thing, as different from each other as you can. For samosas: a fried one, a baked one, a big one, a tiny one, one on a plate, one in a paper bag. Write your list down.
This chapter is free — open this lesson in your portal to save your answers and track progress.
AI only knows what its examples taught it. Narrow or unfair data leads to wrong or unfair predictions. You'll hear this rule again and again in this course.
Pick something simple — say, "what makes a good samosa" 🥟. Give a friend three examples, ask them to guess your pattern, then test them with a fourth.
Did your three examples teach the right pattern, or a lazy one?
Describe one situation where showing an AI limited data would make it get things wrong — just like our mango on a plate.