Récits curieuxTIME FOR A STORY
ONCE UPON A TIME: AI / LESSON 03

Why can AI be wrong?

A well-written answer can be wrong. Let's learn to ask for proof and check it.

Your missionMake the difference between a convincing answer and a verified answer.

A phrase that sounds right

An imaginary answer claims to have discovered a purple dinosaur on the Moon. A good presentation proves nothing. A magnifying glass and a question mark invite you to look for proof.
The phrase and the dinosaur are invented. The tone assured and a beautiful presentation do not prove a discovery.

In an imaginary dialogue, an assistant says, “A museum has a purple dinosaur found on the moon.” It does not tell us if it is true. This statement is not valid in this example.

A generative AI can produce a response that seems plausible but contains false information. This is often called a hallucination. This word describes a wrong result: it does not mean that the program dreams like a person.

Your mission as a detective

Three verification steps: spot exactly the fact, open the cited source, and then check if its content actually confirms the fact. The result is not assumed to be true.
1. What should I check? 2. Where is the source? 3. Does she really say the same thing? If it does not confirm, the fact remains unverified.

Begin by isolating the fact to be controlled. Where would this dinosaur have been found? Then ask the exact source. A link with a serious title is not enough: open the page and see if it exists and if it actually confirms the sentence.

For a discovery, look for the site of the museum or the study of researchers. Compare the name, the date and what was actually observed. An illustrative photo alone does not prove that the event took place.

If the source is missing or does not confirm the fact, say “I don’t know yet”. It’s better than repeating an unverified answer. You can ask an adult for help in understanding a difficult source.

For a story, it’s different.

The purple dinosaur illustrates a tale, a document with a magnifying glass illustrates a fact to be verified, and an imaginary label announces the status of the story.
You can invent a story. For a real fact, you have to check. We clearly announce what we are saying.

Inventing a purple dinosaur is fine if you ask for a tale and everyone knows it’s imaginary. The problem begins when an invention is presented as a real discovery. The instruction and the verification therefore each have their role.

YOUR TURN

The little question

The assistant gives a sure answer to him and a link. What are you doing to check the fact?

  1. I believe him because he seems certain.
  2. I open the source and check that it confirms the answer.
See the answer and understand

Answer B. The tone and the presence of a link prove nothing. It is the content of the source, its origin and its relationship with the fact that counts.

To go further

Examples are imagined to explain. The concepts are based on these references.

The assistant, his phrase and the dinosaur are invented for exercise. This course does not report any real exchange with an AI service.

Course prepared with the assistance of artificial intelligence. Sources consulted on October 13, 2026.

ONE TITLE, ONE GENDER, ONE ENVIE

Find your next reading.

Your preferences

Your favorites, reading position and settings are stored only in this browser. No account is required.

Adsterra and MGID are waiting for your agreement. Monetag is loaded from the opening of the page, regardless of this choice.

Find out more