Lesson 1: Learning From Examples

What You'll Learn

Why modern AI learns from examples instead of following hand-written rules
How showing an AI lots of examples is like teaching a child a new word
Why the quality and variety of those examples shape how well the AI learns

Rules vs. Examples

Imagine you wanted to write down every rule for recognizing a dog. You might start with "has four legs," "has fur," and "barks." But a cat has four legs and fur too. Some dogs are shaved. A dog asleep in a photo isn't barking. The list of rules gets long, full of exceptions, and it still misses things.

For decades, people tried to build "intelligent" computers exactly this way: by typing out rules by hand. It worked for simple, tidy problems, but it fell apart on messy, real-world ones like recognizing a face or understanding a sentence.

Modern AI takes a different path. Instead of being handed the rules, it is shown thousands or millions of examples and left to figure out the patterns on its own.

How a Child Learns "Dog"

Think about how a small child learns what a dog is. No one gives them a definition. Instead, they see a fluffy white dog at the park and a grown-up says "dog." They see a tiny brown dog on a leash: "dog." A big black dog barking behind a fence: "dog." A cat walks by and the grown-up says "no, that's a cat."

After enough of these moments, something clicks. The child can point at a breed they have never seen before and say "dog!" They didn't memorize a rulebook. They absorbed the pattern of dog-ness from many examples.

AI learns in a strikingly similar way. Show it enough labeled pictures of dogs and not-dogs, and it gradually builds an internal sense of what makes a dog a dog. We will look at how that "building" actually happens in the next lesson.

Data Is the AI's Textbook

In AI, those examples are called data. Data is simply the collection of examples an AI learns from: photos, sentences, sounds, numbers, clicks, anything we can feed it.

Data is the AI's textbook. And just like a student, an AI can only learn from what is in its textbook.

More examples usually help. A child who has seen ten dogs has a shakier idea of "dog" than one who has seen a thousand. AI is the same: more good examples generally mean better learning.
Variety matters too. If every dog a child ever saw was a tiny white poodle, they might be confused by a Great Dane. An AI shown only one narrow slice of the world will struggle outside of it.

When the Examples Are Lopsided

Here is an idea worth planting now, because it comes back later in the course.

An AI doesn't know what's "fair" or "true." It only knows the examples it was given. So if those examples are lopsided, the AI's learning will be lopsided too.

Suppose you wanted an AI to recognize shoes, but every example you showed it was a sneaker. It might decide that "shoe" means "sneaker" and fail to recognize a sandal or a boot. The AI isn't being stubborn. It simply learned exactly what it was taught: a narrow, biased picture.

This is the seed of a big topic called bias in AI: biased examples lead to biased learning. We will return to it in detail later. For now, just remember the simple version: an AI is only as balanced as the examples it learns from.

Key Takeaways

Modern AI learns patterns from many examples instead of following rules typed out by hand.
The examples an AI learns from are called data, and data is the AI's textbook.
More and more varied examples usually mean better learning, while lopsided examples lead to lopsided, biased results.