AI safety · 3 min read
Confidence is not evidence
The chatbot said “Absolutely.” Your child heard a grown-up voice. The voice sounded sure. Sure is not the same as true.
AI is good at the sound of an answer. It is less good at knowing when it has invented a book, a quote, or a neat number that never existed. UNESCO puts it plainly: these systems guess when they are unsure, and the guess often arrives with complete confidence. That is the trap.
- A confident AI answer can still be wrong.
- Teach one follow-up: “What source did you use?”
- Then open the source together — does it exist, is it reliable, does it actually support the claim?
- The habit we want: pause before believing, and look for evidence before repeating or sharing.
Why the sure voice fools us
Children (and adults) treat certainty as a clue. A teacher who says “I am sure” usually has a reason. A friend who says “definitely” has usually seen the thing. We borrow that habit when we talk to a chatbot.
The machine did not “see” anything. It predicted the next words. Common Sense Media’s parent guide names the result: a hallucination — something made up, presented as fact. A date that is wrong. A book that was never written. A statistic that looks tidy because tidy is a pattern the model likes.
That does not make AI useless. It makes tone a poor judge. The sentence that sounds most finished is not the one you should trust first.
The one follow-up question
You do not need a lecture on large language models. You need a sentence your child can say without rolling their eyes:
“What source did you use?”
Ask it out loud the first few times. Then let them ask it. If the tool names a page, a paper, a news story — good. That is a starting point, not a finish line. A citation is a pointer. It is not proof.
If the tool shrugs, changes the subject, or says “let’s not ruin the mood,” that is also an answer. No source means no sharing. No source means we do not put that sentence in the homework.
Open it together
The useful work happens after the chat. Open the source in a real browser.
Three small checks, said in kitchen words:
- Does it exist? Click the link. A title the AI invented will often vanish here.
- Is it reliable? Who wrote it? When? A random blog and a museum page are not the same kind of “source.”
- Does it actually support the answer? Read the bit that is supposed to prove the claim. Many times the page is real and the claim is still a stretch.
You will not do this for every joke and every football fact. Do it for the claims that matter: health, homework they will hand in, something they want to post, something they want to argue at the table.
UNESCO’s media-literacy advice is the same at family scale: pause before sharing, question the source, and do not treat a fluent paragraph as a finished thought.
The habit we want
The goal is not to make your child scared of AI. The goal is a pause.
Confidence is a tone. Evidence is something you can open. When those two come apart, believe the page — or the empty search — not the sure little voice in the chat.
If they bring you a confident answer and no page, do not turn it into a trial. Say, “Show me where that lives.” Sit down for two minutes. If the search comes up empty, name what happened in small words: the chat sounded sure, and the sure bit was made up. Then ask the original question again — or look it up the old way.
You will not catch every flourish. You do not need to. One empty search, seen together, teaches more than a warning poster.
Save the question on a sticky note near the computer if you like: What source did you use? Then look. Then decide.
The safe way for kids & teens to learn AI.
They build the thing they actually want with Pip — real coding skills grow on the way.
