← All questionsYou Noah Lab

How It Works · AI & Robots

Why does AI sometimes get things wrong?

A language AI builds its answer by predicting words that are likely to fit. It does not look facts up unless it has a tool for that, so it can sound sure and still be wrong.

How it works

  1. It learned patterns from a huge amount of writing, including writing that had mistakes.
  2. When it does not know something, it may still produce an answer that merely sounds right.
  3. People call a confident wrong answer a hallucination.
  4. That is why important facts should be checked against a trusted source.

How do we know?

In the AI & Agents lab, an agent with no library tool says Mars has 4 moons. Mars has 2. Researchers test AI systems on thousands of questions with known answers to measure how often this happens.

Quick quiz

What should you do with an important fact from an AI?

Show the answer

Check it with a trusted source

Try it yourself

🔬 Open AI & Agents Lab📖 Ask NOAH in the lab

More AI & Robots questions

Places grown-ups can check

NIST · NASA

Written for kids by You Noah Lab with AI help, using standard science references. Spotted a mistake? Science gets better when people check it, so please tell a grown-up. You Noah Lab · All questions · Privacy