Mission 4 of 7 · 8 min
AI can get it wrong – and make things up
Why a chatbot sounds sure even when it doesn't know, how to spot a made-up number or source – and how to use AI for school wisely.
Why AI gets things wrong
A chatbot usually doesn't look up the answer in an encyclopedia. It puts together text from words that probably belong together – based on the huge amount of text it learned from. Most of the time that works well. But when it doesn't know something, it may not realise. Instead of “I don't know”, it can write something that only sounds right: a name, a number, a quote, a book title or a link.
This is called a hallucination. And the chatbot writes it just as confidently as the truth.
The companies that make chatbots admit this themselves. Their terms say that answers may not be accurate and need to be checked. One of them warns that an answer can be wrong even if it looks accurate because it's full of details. A chatbot also may not know the latest information – such as today's opening hours or yesterday's match result.
Game: Where is the AI making things up?
Correct: 0 of 3Read the chatbot's answer. Tap the parts that are made up or can't be checked. One answer is fine, though.
If you can't see anything suspicious, tap “I see nothing suspicious”.
Round 1 of 3 · Found: 0
Question: Write me a few sentences about Charles Bridge for my report.
Charles Bridge in Prague was founded by Emperor Charles IV in 1357. The bridge is exactly 2 kilometres long and stands on 100 arches. All historians agree on this. Source: the book “The Stone Secrets of Prague” (2016).
The breakdown will appear here once you finish.
How to use AI wisely
- What it's good for: explaining a topic in other words, coming up with practice questions, helping with an outline or checking your own writing.
- What to always check: names, dates, numbers, quotes and sources. That's where mistakes matter most and are hardest to spot.
- You have to find the source yourself. If a chatbot mentions a book or an article, try to track it down in the library or online. If you can't find it, it doesn't belong in your work.
- Don't type personal things into the chat: your address, phone number, passwords or photos of friends. One big provider warns that some conversations are read by people who work on improving the chatbot, and advises not to type confidential information into the chat.
- Ask your teacher whether and how you can use AI for an assignment. And if you use it, say so.
Find out more about the “Fake proof” trick in the Manipulation atlas.
Quiz: Smart with AI
Quiz · 0/31. A chatbot gave you a source for your report: a book with a title and a year. What do you do?
2. Why does a chatbot sound sure even when it doesn't know something?
3. What is AI most useful for at school?
Talk together
- Do you use a chatbot for school? What does it help you with – and where has it got things wrong?
- Shall we ask it together about something we know for sure – like our town or a castle we visited on a trip? Does it get it right?
- What do you never type into a chatbot, and why?
Tip for adults: Chatbots' terms differ: some allow them from 13 with parental permission, others only from 18. Agree with your child (and the school) on what they may use one for. And try out together how easily a chatbot makes something up about a place you know well.
Found a mistake or inaccuracy in this lesson?Report a mistake
Finished the mission?
Only this browser remembers the progress. We don't send it anywhere and store nothing about your child.