Hands on AI education for kidsSmart isn’t the same as right.
Children ask AI questions, get homework help from it and see pictures it made, often before they can tell what it is. Ethical AI means teaching them to use it honestly, think for themselves, and notice when a machine gets something wrong.
84
sets of AI rules written by governments, companies and scientists around the world
The grown-up rules match the kid-sized ones.
Researchers at ETH Zurich read every major AI ethics guideline they could find and counted which ideas kept coming up. Five appeared in more than half of them.
Put plainly, they are the same habits we want children to have: be open, be fair, don’t hurt anyone, take responsibility, and respect privacy.
- Be open about how it works transparency73of 84
- Be fair to everyone justice & fairness68of 84
- Don’t cause harm safety60of 84
- Someone is responsible accountability60of 84
- Protect people’s privacy privacy47of 84
What “ethical AI” means for a kid
AI isn’t a person and it isn’t magic. It learned by looking at huge amounts of things people made, and it makes its best guess from that. Sometimes the guess is brilliant. Sometimes it’s confidently wrong, unfair to someone, or not real at all.
Ethics is simply the question of what the right thing to do is. With AI, that comes down to a handful of habits children can start building from their very first conversation with a machine.
Think for yourself
A machine can sound sure and still be wrong. It’s always okay to question it.
Do your own work
Getting help is fine. Handing in work you can’t explain isn’t.
Check what’s real
AI can make pictures, voices and stories that look true but never happened.
Be fair to everyone
Ask who the AI might not work well for, and who got left out.
Keep private things private
Names, addresses, photos and secrets don’t belong in a chatbot.
Why it matters: what researchers found
Four peer-reviewed studies about children, students and families. Each one shows why good habits with AI need to start early.
Kids follow robots, even when the robots are wrong
Children did a simple matching task next to small robots that sometimes gave the wrong answer on purpose. Their scores dropped, and 74% of their mistakes were the exact same wrong answer the robots gave. Adults in the same test ignored the robots.
It’s okay to disagree with a machine.
Vollmer et al., Science Robotics, 2018
Kids learn to trust voice assistants for facts
Children aged 4–5 and 7–8 heard a voice assistant and a person answer questions. The older the children, the more they trusted the voice assistant for facts, and the more they turned to the person for personal information. That trust is useful, but only if kids also learn the assistant can be wrong.
Know what AI is good for, and check it.
Girouard-Hallam & Danovitch, Developmental Psychology, 2022
Facts, for example
“How far away is the moon?”
Personal things, for example
“What did you have for breakfast?”
When AI does the work, learning drops
Nearly a thousand high school maths students were given AI help. Those with an ordinary chatbot did much better on practice, then worse on the test than students who never had AI. A version designed to give hints instead of answers avoided the drop.
Use AI to help you learn, not to do the learning.
Bastani et al., PNAS, 2025
Compared with students who never had AI.
Most people can’t tell an AI face from a real one
People were shown real photos mixed with faces made by AI. They guessed right less than half the time, and even rated the AI-made faces as slightly more trustworthy than real ones.
Seeing isn’t always believing.
Nightingale & Farid, PNAS, 2022
48.2%correct at telling real faces from AI-made ones. A coin flip gets 50%.
What it looks like at every age
The ideas stay the same from 5 to 14. What changes is how deep children can take them.
A computer is not a person
Young children often talk to AI as if it were a friend who knows everything. The first step is knowing it’s a tool that can make mistakes.
Try saying
“It’s a clever computer. Clever computers still get things wrong sometimes.”
Help versus doing it for me
This is when AI starts showing up in schoolwork. Children can learn the difference between getting a hint and getting the answer.
Try asking
“Could you explain this to me without the AI next to you?”
Real, fair and private
Older kids meet AI-made images, videos and chatbots on their own. They’re ready to ask whether something is real, who it’s fair to, and what they’re sharing.
Try asking
“How would you know if that picture was made by AI?”
Five questions to ask any AI
Short enough for a six-year-old to remember, useful enough for a teenager to keep using.
Could it be wrong?
Sounding sure isn’t the same as being right.
Did I do the thinking?
If you can’t explain it, it isn’t really yours yet.
Is this real?
Pictures, voices and videos can all be made up.
Who might it not work for?
Something that works for most people can still leave someone out.
Should I share this?
Private things stay private, even with a friendly chatbot.
The apps will change. Good habits last.
Children will use tools we can’t imagine yet. What carries over is the habit of pausing to ask: is this true, is this fair, is this mine, and is this safe?
Sources
- 1.Vollmer, A.-L., Read, R., Trippas, D. & Belpaeme, T. (2018). Children conform, adults resist: A robot group induced peer pressure on normative social conformity. Science Robotics, 3(21), eaat7111. doi.org/10.1126/scirobotics.aat7111
- 2.Girouard-Hallam, L. N. & Danovitch, J. H. (2022). Children’s trust in and learning from voice assistants. Developmental Psychology, 58(4), 646–661. doi.org/10.1037/dev0001318
- 3.Bastani, H. et al. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. PNAS, 122(26), e2422633122. doi.org/10.1073/pnas.2422633122
- 4.Nightingale, S. J. & Farid, H. (2022). AI-synthesized faces are indistinguishable from real faces and more trustworthy. PNAS, 119(8), e2120481119. doi.org/10.1073/pnas.2120481119
- 5.Jobin, A., Ienca, M. & Vayena, E. (2019). The global landscape of AI ethics guidelines. Nature Machine Intelligence, 1, 389–399. doi.org/10.1038/s42256-019-0088-2