How Machines Learn

From Last Lesson...

You learned that AI is software that learns from experience. But HOW does it learn? That's what this lesson is about.

Don't worry — no maths, no code. Just clear explanations with everyday examples.

---

Learning by Example

Imagine you're learning to cook. How would you do it?

1. Watch someone cook several dishes (these are your examples)
2. Notice patterns — "When they add salt, the food tastes better. When they add too much, it's ruined."
3. Try it yourself — Maybe your first dish isn't great
4. Get feedback — "Too salty!" So next time, you use less salt
5. Improve over time — After many dishes, you become a good cook

This is exactly how AI learns.

The fancy term for this is machine learning, but really it's just: show the computer lots of examples and let it figure out the pattern.

---

Three Types of Learning

Just like humans learn in different ways, machines do too:

1. 📚 Supervised Learning — "Learning with a Teacher"

This is like a teacher showing you flashcards:

• Card shows a picture of a cat → Teacher says "CAT"
• Card shows a picture of a dog → Teacher says "DOG"
• After hundreds of cards, you can recognise new cats and dogs on your own

In AI terms: You give the computer labelled examples (photo + correct answer), and it learns the pattern.

Real-world example: Email spam filters. Someone labelled thousands of emails as "spam" or "not spam", and the AI learned what spam looks like.

2. 🔍 Unsupervised Learning — "Learning by Exploring"

This is like sorting your wardrobe without anyone telling you how:

• You naturally group t-shirts together, jeans together, formal clothes together
• Nobody told you the categories — you figured them out yourself

In AI terms: The computer finds hidden patterns in data without being told what to look for.

Real-world example: Customer grouping. A shop's AI might discover that some customers always buy organic food, while others always look for discounts — without being told these groups exist.

3. 🎮 Reinforcement Learning — "Learning from Trial and Error"

This is like learning to ride a bicycle:

• You try, you fall, you adjust
• Nobody gives you step-by-step instructions
• You learn by doing and getting feedback (falling = bad, staying up = good!)

In AI terms: The computer tries things, gets a score (reward or punishment), and learns to maximise the reward.

Real-world example: Game-playing AI. AlphaGo learned to play the board game Go by playing millions of games against itself — and eventually beat the world champion.

---

Let's Walk Through a Real Example

Let's say we want to build an AI that recognises handwritten numbers (0-9).

Step 1: Collect Examples

We gather thousands of handwritten numbers from different people:

• Some write "7" with a line through it, others don't
• Some write "1" as just a line, others add a base
• "0" and "O" look similar but mean different things

Step 2: Label Them

A human looks at each image and labels it: "This is a 3", "This is a 7", etc.

Step 3: Train the AI

We show the AI all these labelled examples. It starts to notice patterns:

• "Numbers with a curve at the top and bottom are probably 8"
• "Numbers with one straight line are probably 1"
• "Numbers with a loop are probably 0, 6, or 9"

Step 4: Test It

We show the AI handwritten numbers it has never seen before. Can it guess correctly?

If it gets 95% right — great! If not, we give it more examples and try again.

---

What is "Training"?

When we say we "train" an AI, we mean:

1. Show it thousands (or millions) of examples
2. It makes a prediction
3. We tell it if it was right or wrong
4. It adjusts slightly to be better next time
5. Repeat steps 2-4 millions of times

After enough rounds, the AI has "learned" the pattern.

Think of it like practicing free throws in basketball. Each throw, you adjust your aim slightly. After thousands of throws, you're quite accurate — not because you memorised each throw, but because you learned the pattern of what works.

---

"Data" is the Key Ingredient

You've probably heard people say "Data is the new oil." Here's why:

• More data = better AI (usually)
• Bad data = bad AI (garbage in, garbage out)
• Biased data = biased AI (if you only train on photos of golden retrievers, it won't recognise poodles)

---

Quick Recap 🎯

1. Machine learning = showing AI lots of examples and letting it find patterns
2. Three types: Supervised (with a teacher), Unsupervised (exploring), Reinforcement (trial and error)
3. Training = the process of showing examples, getting predictions, and correcting mistakes — millions of times
4. Data quality matters — bad or biased data leads to bad or biased AI

---

What's Next? 🚀

Ready to try it yourself? In the next lesson, you'll train your very first AI model — right here in your browser. No installation needed!