DOORS · Digital Skills
Lesson 3

The wrong books

Sometimes the AI's shelf is full. Full of the same false story, copied a million times. Here is how a made-up story becomes a confident answer.

Nobody checked the books

An AI learns to be a chatbot by reading a giant pile of text from the internet. Books, magazines, newspaper articles, blog posts, and posts on discussion forums such as Reddit. That big, messy pile of ideas is its training data. Here is the catch: nobody read every page of that pile before the AI did. Nobody could. It is far too big.

So the pile holds true things and false things, side by side. And when a false story gets repeated online thousands of times, the AI reads it thousands of times. Repetition is how it learns. The AI cannot tell a popular fact from a popular fib. Both just look like something people say a lot.

How a joke gets on the shelf

Meet Kofi. He saw a sticker on a lamppost that said Birds Aren't Real, and he wants to know what that is about. Step through the story and watch the shelf.

Step 1 of 5
Kofi
AI Chat

Fluent does not mean true

The AI retold the whole bird story smoothly, with dates and details, because it read that story thousands of times. Smooth retelling proves one thing only: lots of people wrote it down. It says nothing about whether it is true.

The AI was not lying to Kofi. A lie needs a liar who knows the truth. The AI simply repeated the most common story on that shelf. In the last lesson the shelf was empty and the AI guessed. This time the shelf was packed, and every book on it was the same wrong book.

Why did it know this one was a joke?

Birds Aren't Real became so famous that newspapers wrote about the joke itself. Those articles went into the training data too, like a warning sticker on the books. The AI learned the story and the sticker together.

Brand-new false stories have no sticker yet. Neither do false stories that only spread in small corners of the internet. Yesterday's famous hoax gets caught. Today's quiet one sails right through, told in the same smooth, confident voice.

What you can do about it

Ask "who says so?"

When the AI states something surprising, ask where that idea comes from. If the honest answer is "a lot of posts online," you have measured its popularity and nothing else. Popular and true are different measurements.

Run the boring source test

Exciting claims deserve a boring check. A government health page, an encyclopedia, a librarian. If a claim is real, a dull and careful source will say so too. If only thrilling websites say it, that is your answer.

Be extra careful with new stories

The newest false stories are the most dangerous ones, because no warning sticker exists yet. If a claim appeared this week and the AI repeats it confidently, treat it as unconfirmed until careful sources catch up.

Watch how a claim makes you feel

False stories spread because they are exciting, scary, or satisfying to repeat. When an answer gives you a jolt of "wow" or "I knew it," slow down. That feeling is exactly what helped the wrong books fill the shelf.

Check yourself

These questions cover this lesson and the last one. For each situation, decide which shelf problem you are looking at.

Coming up next

Both shelf problems end the same way: a wrong answer arriving in a warm, confident, friendly voice. The last lesson explains where that voice comes from, and why the AI would rather agree with you than argue. Continue to The yes machine.

This mini-module