When AI is confidently wrong
AI can sound very sure and still be wrong. Learn why, and how to catch its mistakes.
In 60 seconds
When AI is confidently wrong
AI can sound very sure and still be wrong. Learn why, and how to catch its mistakes.
Common mistakes
- Made-up facts: a wrong date, a wrong capital city, a person who never existed
- Made-up sources: a book, website or quote that doesn't exist
- Maths slips: small calculation mistakes, especially in long sums
- Old information: things that changed recently
How to catch mistakes
- 1
Ask "How do you know?"
Ask for sources. Then check that those sources really exist. - 2
Check your textbook
Your textbook and class notes are made for your syllabus. - 3
Do the maths yourself
Or check with a calculator. - 4
Ask your teacher
Especially if the AI and your book disagree.
- "It sounds sure, so it's right"
- Copying a source without checking it exists
- Believing it over your textbook
- "Sounding sure is not proof"
- Searching for the source yourself
- Checking with your textbook or teacher
Watch and read more
Lab
A "catch the mistake" record comparing AI with your textbook.
The problem
Do this with a parent or teacher (many chatbots are only for ages 13 and up).
- Pick 5 facts from your textbook (dates, definitions, a sum).
- Ask a chatbot the same 5 questions.
- Make a table: Question, Textbook answer, AI answer, Same or different?
- For any difference, check with your teacher or another trusted source. Who was right?
No chatbot? Do it on paper: an adult writes 5 "AI answers", some with sneaky mistakes, and you catch them.
You are done when
Hard questions
Try to answer before you reveal. If you can answer these, you understood the lesson.
Q1Why can a made-up source be more dangerous than a made-up fact?Reveal
A source makes an answer look checked and trustworthy. People may copy it into projects without looking. If the source doesn't exist, the mistake spreads further and looks more official than it is.
Q2Why might AI be wrong about something that happened last week?Reveal
Chatbots learn from text collected up to a certain time. Unless the app searches the internet, it may not know about recent events, and it might guess instead. For fresh news, check a trusted news source.
Questions people ask
Why doesn't the AI just say "I don't know"?
Chatbots are built to give helpful-sounding answers, so they often fill gaps with a guess. Some newer ones say when they're unsure more often, but you still can't count on it.
Is AI more likely to be wrong about some things?
Often yes: very recent news, small local facts, exact numbers, long calculations and quotes. Be extra careful there.
If two chatbots say the same thing, is it true?
Not necessarily. They may have learned the same wrong idea. Check with a trusted source like your textbook, a teacher or an official website.
Lesson test
5 questions. Get 3 right (60%) to pass and complete this lesson.