AI Literacy: The 5 Things Every HK Child Must Understand Before Secondary School

AI literacy is becoming as fundamental as reading literacy and numerical literacy. I don't say this to be dramatic — I say it because children who finish primary school without a working mental model of what AI is and how it shapes their world are going to find secondary school, and the world beyond it, more confusing than it needs to be.

Here are the five things I believe every child should understand before they start S1. None of these require technical knowledge. All of them can be developed through conversation at home.

1. AI learns from examples, not rules

Most people — including many adults — have a mental model of computers as rule-following machines. You tell it what to do; it does exactly that. This was largely true of computing for decades and produced a mental model that is now misleading.

Modern AI systems don't work this way. They learn patterns from large numbers of examples. A face recognition system doesn't have a rule for "what a face looks like." It has been shown millions of faces and has learned to detect features that correlate with "face." A text generation system doesn't have a rule for "how sentences work." It has processed enormous amounts of text and learned patterns of what word tends to follow what other word in what context.

This matters because it explains both the strengths and the weaknesses of AI. It's very good at pattern recognition in domains where it has seen many examples. It can fail in surprising ways when it encounters situations that differ from its training data. And it can inherit biases present in the examples it learned from.

Children who understand this framework can make sense of why AI facial recognition works better for some groups than others, why AI translation occasionally produces absurd results, and why an AI system confident that a sentence is "normal" can be confidently wrong.

2. AI systems are designed to optimise for something — and that something may not be your wellbeing

Every AI system is optimised towards a goal. The AI that curates your news feed is optimised to maximise engagement. The AI that recommends your next YouTube video is optimised to keep you watching. The AI that shows you advertisements is optimised to produce purchases.

These goals are not the same as your wellbeing. A news feed optimised for engagement will surface the most emotionally provocative content. A video recommendation system optimised for watch time will surface content you'll watch all the way through — not necessarily content that is accurate, healthy, or intellectually enriching.

This doesn't make these systems malicious. They're doing what they were designed to do. But a child who understands what a system is optimised for can ask: "Is what this system is showing me serving me, or serving its purpose?" That question is protective.

3. AI-generated content is not automatically true

Language models produce plausible text. Images generated by AI look real. Video generation is becoming convincingly realistic. None of this is the same as being accurate.

Children need to understand at a gut level — not just intellectually — that content produced by AI can be entirely fabricated and entirely convincing. The confidence with which an AI system expresses something has no relationship to whether it is true.

The practical implication: any specific factual claim, any cited source, any statistic from an AI system needs independent verification. "ChatGPT said so" is not a source. This habit of verification is learnable and children who develop it before secondary school will navigate the information environment much more safely.

4. Their data is valuable and they are constantly generating it

Every interaction with a digital system generates data. Every search, click, purchase, location check-in, heart rate reading, and social media scroll creates a data point that is stored, aggregated, analysed, and used to make decisions and predictions.

Children who are aware of this can make more deliberate choices. Not necessarily refusing all data-generating activities — that's neither practical nor necessarily desirable — but understanding what they're exchanging and whether they're comfortable with that exchange.

Ask a primary school child: "When you play this game for free, do you know how the company makes money?" Most won't have thought about it. The answer — that their attention data and behaviour data have commercial value — is a useful foundation for more sophisticated thinking about privacy and consent as they get older.

5. Being critical doesn't mean being cynical

I want to end with this because it's the imbalance that concerns me most. Some well-educated children, having absorbed messages about AI risk and fake news and algorithmic manipulation, become reflexively sceptical about everything digital — a kind of learned helplessness dressed as sophistication.

That's not the goal. Critical AI literacy means having the tools to evaluate rather than the reflex to dismiss. It means asking questions, checking sources, understanding incentives — and also being able to use and appreciate AI systems that genuinely serve you well.

A student who uses Khan Academy to understand a maths concept, and understands that Khan Academy is designed to help them learn, is using AI well. A student who refuses to use any AI tool because "you can't trust AI" has gone too far in the other direction. The skill is discernment, not avoidance.

Developing these understandings at home

None of these require formal instruction. They develop through repeated small conversations.

When a recommendation appears on a streaming service: "Why do you think it's suggesting that? What does the app know about you that made it choose that?"

When your child reads something surprising online: "How would we check if this is true?"

When a game makes them feel urgently that they need to purchase something: "Let's look at how this game is designed to make you feel that way."

These conversations are not burdensome. They take two minutes and they build mental habits that compound over years. By S1, a child who has had these conversations regularly will have a conceptual vocabulary for the AI-shaped world they're entering — and that vocabulary will serve them far better than a course in programming.