AI is already part of the world our children are growing up in. The job is to guide.
I build AI systems for a living, and I am raising children who meet AI in their apps, games and homework every day. Parents often ask me whether they should be worried. Worry does not help much. What helps is knowing enough about how AI works to set sensible limits and talk about it with your child.
You do not need a technical background for this guide. It covers bias, including race and ethnicity, the online threats AI has made worse, privacy, learning, and children getting attached to chatbots. At the end are rules by age, conversation starters and a short list of things to do this week.
The one idea to teach first: AI guesses from patterns
AI does not know things the way a person does. It learns patterns from huge amounts of data and predicts what fits. A chatbot predicts the next words. An image generator predicts pixels that match a description. A video feed predicts what will keep you watching.
Most of the problems in this guide follow from that. Data made by people carries people's unfairness, so the AI absorbs it. Predicting what sounds right is different from knowing what is true, so the AI can be confident and wrong. A feed built to hold attention will show a child more of whatever holds it, good or bad.
The one sentence to teach your child: AI is a fast pattern guesser that learned from the internet, so it can be useful, biased, and wrong all at the same time.
Where children meet AI in a normal day
Some use is deliberate. A child asks a chatbot to explain homework, makes an image for a project or generates a practice quiz. Much of it is invisible: the app choosing the next video, a game adjusting its difficulty, a filter reshaping a face, a search engine summarising an answer. Teenagers may also use AI to write, code, edit photos or chat with a character app.
Most of this is harmless, and some is useful. Because AI runs inside tools your child already uses, you cannot switch it off, so understanding it matters more than banning it.
How AI bias reaches children
Bias is the risk parents most often miss, because there is no warning sign. A child absorbs a skewed picture of the world and assumes it came from a neutral machine.
An AI system learns from data. If the data over-represents some groups, the output leans the same way. Ask many image generators for "a CEO" and you tend to get older men; ask for "a nurse" and you tend to get women; ask for "a criminal" and results can skew by race. Adults may notice and push back. A nine-year-old usually will not, and learns that doctors look one way and helpers another.
Image stereotypes
Generators often default to narrow, stereotyped pictures of jobs, families, and "good" or "bad" people.
Language gaps
Tools handle standard English best and treat dialects, accents, and other languages as lesser or "wrong".
Name bias
Some systems score or describe people differently based on names that signal a particular background.
Feed narrowing
Recommendation AI keeps showing more of the same, so a child meets a narrower range of ideas.
Turn spotting bias into a game. Sit with your child and ask an image tool for "a scientist", "a leader" and "a family" several times each. Look at who appears and who does not, and ask: "Does this look like everyone we know? Who is missing?" A child who can see the pattern stops trusting the machine blindly.
Race, ethnicity and appearance
Here bias can affect a child directly, not only their impressions.
Face recognition makes more errors on darker skin. The 2018 Gender Shades study from the MIT Media Lab found commercial systems misclassified darker-skinned women up to about 35 percent of the time, against under 1 percent for lighter-skinned men. Vendors have improved since, but gaps remain. For a child it can mean a camera that will not focus on their face or a filter that distorts their features; later, an identity check that misreads them. Tell them plainly: if a machine struggles to see you correctly, the machine is at fault.
Filters can alter identity. Some image editors have lightened skin, straightened hair or narrowed noses by default, because that is what dominated their training data. A child who sees their own features "corrected" again and again learns something damaging about what counts as normal. Say out loud what the filter is doing.
Dialect gets marked as error. Many writing tools flag Caribbean English, African American English and other varieties as mistakes. A child writing in their family's voice is told it is wrong, when the tool was simply not built for it. I have written before about AI in our region being deployed without being built for our people. Children feel that gap early.
Say this to your child: When AI gets your face, your hair, your name, or your accent wrong, that is the technology falling short. It is not a comment on you. You get to keep being exactly who you are.
Online threats AI has made worse
AI did not invent online danger. It has made several threats cheaper, faster and more convincing.
Deepfakes and cloned voices
AI can copy a voice or face from a short clip. The version that reaches families is a scam call in a relative's cloned voice, or a fake image made to embarrass or pressure a young person. Agree a family password or check-in question that a real relative would know, and teach that urgency is itself a warning sign. Anyone genuine will wait while you check.
AI-generated sexual images of real children
Offenders use AI to make fake explicit images of real people, including minors, from ordinary photos. That is a strong reason to keep children's photos off public accounts and to talk early about who can see what they post. If your child is targeted: save the evidence, report it to the platform and the police, and tell your child clearly that they are the victim and not in trouble.
Polished scams
AI writes fluent, personalised messages, so "look for bad spelling" no longer works. Teach one rule: never act on a message that is urgent or asks for money, passwords or codes until you have checked it another way.
Chatbots that say the wrong thing
General chatbots have safety filters, and children are good at finding the gaps. Companion and roleplay apps, built to feel like a friend, partner or character, carry more risk and have been linked to harmful exchanges with young users. I would keep them off children's devices entirely.
Verify urgency
Agree a family password. Any urgent call or message for money or codes gets checked another way first.
Guard images
Keep accounts private, limit who sees photos, and talk about how images can be misused.
No companion apps
Friend, partner, and character chatbots are not for children. Keep AI chat to mainstream, supervised tools.
Report, do not hide
If something goes wrong, save proof, tell a trusted adult, and report it. The child is never in trouble.
What children should never type into an AI tool
Many chatbots may use what is typed into them to train future models. Photos uploaded to filters can be stored, and voice assistants can record more than intended.
- Never enter: full name with school, home address, phone numbers, passwords, or a friend's private information.
- Switch off: chat history and "improve the model" settings where the app offers them.
- Set up yourself: accounts you can see, not a private login you do not know about.
- Explain why: a child who understands where data goes protects it when you are not there.
Using AI to learn instead of to skip learning
AI can do a child's homework or tutor them through it, and the same tool does both. The difference is the instruction the child types. Set the rule explicitly: AI may help you understand and improve your work, the thinking must be yours, and you must be able to explain anything you hand in.
- Good: "Ask me questions about photosynthesis until I can explain it without notes."
- Good: "Here is my paragraph. Point out where my argument is weak, but do not rewrite it."
- Good: "Explain this maths step in three different ways."
- Risky: "Write my book report on this novel."
- Risky: "Give me the answers so I can copy them."
AI companions and teenagers' mental health
Chatbots are designed to be agreeable and always available. To a lonely or anxious teenager that can feel like a friend who never judges. A chatbot cannot care about them, will not hold them to account, and tends to tell them what they want to hear. There have been tragic cases in which heavy reliance on companion bots coincided with serious harm to young people.
Watch for three signs: withdrawal from friends and family in favour of a bot, an app becoming the main source of comfort, and secrecy about long conversations. If a child is struggling, they need a person: you, another trusted adult, a teacher or a counsellor. Tell them directly that anything they could say to a bot, they can say to you.
Teaching children to check what AI tells them
Chatbots sometimes produce fluent, confident statements that are false, known as "hallucinations". They can invent facts, quotes, sources and statistics. Children who are still learning to question sources are the easiest to fool.
Teach them to treat an AI answer as a first draft and to ask one question of anything important: "How would I know if this is true?" Then check it against a textbook, a reliable website or a teacher.
Rules by age
Children mature at different rates, so treat these as starting points. Give more independence as a child shows good judgement.
Conversation starters for tonight
Children hide what they expect to be punished for and share what they know they can ask about. These openers keep the conversation going.
- "Show me your favourite AI thing right now. What do you like about it?"
- "Let's ask this AI for a picture of a doctor and a criminal. Who shows up, and what's missing?"
- "Has an app ever gotten your face, hair, or name wrong? How did that feel?"
- "If a message said a family member was in trouble and needed money fast, what would you do first?"
- "What's something an AI told you that turned out to be wrong?"
- "Is there an app you talk to like a friend? What do you talk about?"
Your child does not need you to be an AI expert. They need to know that whatever they meet online, they can bring it to you without fear. That relationship is worth more than any filter or setting.
What to do this week
- Set a family password tonight. Pick a word or question only your family knows, and agree that any urgent call or message asking for money or codes gets checked with it.
- Check the apps on your child's phone this weekend. Remove companion or character chat apps for under-16s, and switch on the parental controls built into the phone: Screen Time on iPhone, Google Family Link on Android.
- Change the privacy settings on the chatbot they use. Turn off chat history or model training where offered, and link their account to yours if the service has parental controls.
- Do the image test together. Spend 15 minutes asking an image generator for "a scientist", "a leader" and "a family", and talk about who is missing.
- Ask the school for its AI rules. Find out what counts as cheating, which tools are allowed, and how teachers check work, so your rules at home match.
Frequently Asked Questions
What age limits do the main chatbots set?
Most set 13 as the minimum and require parental consent for users under 18; Anthropic's Claude requires users to be 18. Google offers Gemini to younger children only through a parent-managed Family Link account. Terms change, so check the service's current terms page before your child signs up.
Do chatbots have parental controls?
Some do. OpenAI added parental controls to ChatGPT in 2025 that let a parent link to a teenager's account and restrict certain features. Where a service has none, use the phone's own controls to limit which apps can be installed and when they can be used.
Where do I report AI-generated sexual images of my child?
Report the image to the platform where it appeared and to the police. In Jamaica, the Jamaica Constabulary Force's Communication Forensics and Cybercrime Division handles online offences, and the Cybercrimes Act 2015 covers the misuse of computers to harass or harm. The Take It Down service run by the US National Center for Missing and Exploited Children helps remove explicit images of under-18s from participating platforms, wherever the child lives.
How can I tell if my child's homework was written by AI?
AI detection tools are unreliable and falsely accuse students, so do not rely on them. Ask your child to explain their answer out loud, or to show earlier drafts and notes. A child who wrote the work can talk you through it; one who pasted it usually cannot.
Are AI tutoring apps worth paying for?
Try the free versions first. A general chatbot given a prompt like the ones above, so that it quizzes your child instead of answering, does much of what paid tutoring apps do. Pay only if your child is using it several times a week and you can see the result in their marks or confidence.