Pashavo/Explainer

AI makes things up: how to protect yourself

A false answer looks exactly like a true one. These are the five places where the mistakes cost you.

The technical word is "hallucination". In plain terms: AI makes things up. It is not lying on purpose — it is guessing, and when it guesses wrong it carries on with exactly the same confidence.

Five dangerous places

1. Health. Do not ask AI for a drug dose, and do not use its answer in place of a doctor. It can name a medicine that does not exist, or a dangerous dose.

2. Law and government. AI can invent a section of an act that sounds completely real. For land, tax, licences or court matters, confirm at the relevant office.

3. Money. Prices, interest rates and loan terms change monthly. The training data does not contain today.

4. References and quotes. Ask for books or studies and the AI may invent author names and titles that do not exist. This has cost many students dearly.

5. Recent news. What happened this week is not inside it, unless the system can search the web.

Three simple defences

  • Ask for the source. "Where did you get this?" If it cannot name something you can check, do not use it.
  • Ask twice, differently. When the two answers disagree sharply, that is your warning.
  • The names-and-numbers rule. Any name, any date, any number, any price — verify it outside the AI before you act on it.

You will go far with AI if you treat it as a knowledgeable friend who is sometimes wrong, not as a book of facts.

More stories