AI's invented answers
Be able to explain why an AI answer can be fabricated even if it sounds confident.
Prerequisites
- BLanguage models and probabilitiesrequired
Everyday explanation
Ask an AI about a report or article that does not exist. It may respond with authors, dates, and a summary. It is all made up. This is called the model hallucinating.
Why? It guesses words that fit. “The report ... was published by ...” — a company name fits, so it writes one down. It does not know the difference between what it has read and what it is making up.
A confident tone does not mean the answer is true.
Intuition
How to check:
- Ask for the source — and check that the source exists and says what the AI claims.
- Ask the same thing again in a different way. Different answers = warning sign.
- Look it up in a database, on Wikipedia, or ask a knowledgeable colleague.
Fabrication is most common with: exact figures, years, quotes, names of people, and obscure topics. It is most often correct with: things found in thousands of places (capital cities, basic rules).
Mastery means
- Explains why an AI answer can be fabricated even if it sounds confident
- Verifies an AI claim against a source
Sign in to do the exercises and build your mastery up.
Sources
- Wikipedia — Hallucination (artificiell intelligens) (CC BY-SA 4.0) — CC BY-SA 4.0
- Skolverket — About AI in school (in Swedish) — Skolverket's open terms