YanSi Lesson Planner

Moment Library

Ready-made 5–15 minute AI-literacy moments, written for Hong Kong classrooms. No AI, no login — just open one and teach it tomorrow.

Strand
Level
Devices

15 moments

Know AI10 min · Board only

The Next-Word Game

Students play the role of an AI, guessing the most likely next word for sentence starters — until a trap question shows that the most popular word is not always the true one.

P4 · P5 · P6 · Chinese Language · English Language

Why this matters: AI chatbots don't look up facts — they predict the most likely next word, which is exactly why a fluent, confident answer can still be wrong.

Steps

  1. 1. (1 min) Announce: "Today the whole class is the AI. Your only job: say the MOST LIKELY next word — not the cleverest word, the most common one." Write sentence 1 on the board.
  2. 2. (3 min) Run sentences 1–3. For each: students write their word in their exercise book, then hands up; tally the top word on the board. Point out that almost everyone picks the same word — that is what an AI does: it has 'read' millions of sentences and picks the most common continuation.
  3. 3. (2 min) Run the trap: "The capital of Australia is ___." Tally, then reveal: most predicted Sydney; the answer is Canberra. Ask why so many 'predicted' Sydney — because Sydney appears with 'Australia' far more often in what we hear and read.
  4. 4. (3 min) Pairs discuss: "If an AI works this way, WHEN will it give wrong answers?" Take 2–3 responses; steer to: whenever the most common way of saying something is not the true fact.
  5. 5. (1 min) Exit check, cold-call two students: "Does the AI KNOW the answer is Canberra, or does it PREDICT Sydney? What's the difference?"

Put on the board

YOU are the AI. Say the MOST LIKELY next word. 1. Once upon a time, there was a ___ 2. For breakfast this morning I ate ___ 3. The opposite of hot is ___ ★ 4. The capital of Australia is ___ The AI's rule: pick the word that most often comes next — NOT the word that is true.

Make it yours: Humanities or science twist: use content starters — "Water boils at ___ degrees" — with the trap "The largest desert in the world is the ___" (most predict the Sahara; the answer is Antarctica).

Know AI10 min · Board only

Teach the Machine Wrong

The whole class becomes a 'learning machine' that may only copy patterns from its training data — including a batch of homework data containing a classic fraction mistake.

P5 · P6 · S1 · Mathematics

Why this matters: An AI can only be as right as the data it learned from: it copies patterns faithfully, and has no understanding it could use to notice that its data taught it wrong.

Steps

  1. 1. (2 min) Announce: "You are now one learning machine. One rule: you may ONLY copy the pattern in the training data — no using the maths you already know." Show Machine 1's data (2→4, 5→10, 7→14). Ask: input 9 — what does the machine output? Hands up, tally.
  2. 2. (2 min) Machine 2: 1→1, 2→4, 3→9. Ask for inputs 5 and 10. Then tell the class: "Nobody told you the rule — you worked it out from examples. That is exactly what 'training an AI' means."
  3. 3. (3 min) Machine 3: reveal the training data 'taken from someone's homework' (1/2 + 1/4 → 2/6, etc.). Ask the machine: 1/2 + 1/3 → ? Insist on machine rules until the class answers 2/5. Then: "Hands up if that answer is wrong. Now hands up if the MACHINE could ever notice."
  4. 4. (2 min) Pair talk: "The machine got it wrong — is it stupid, or is something else going on?" Draw out: it learned its bad data perfectly; it has no idea what a fraction IS, so it has nothing to check against.
  5. 5. (1 min) Close: "AI chatbots were trained on the whole internet. What does that tell you about trusting their answers?" Each student says one sentence to a partner; cold-call one pair.

Put on the board

THE LEARNING MACHINE — you may ONLY copy the pattern. Machine 1 training data: 2→4 5→10 7→14 Input: 9 → ? Machine 2 training data: 1→1 2→4 3→9 Input: 5 → ? 10 → ? Machine 3 training data (from someone's homework): 1/2 + 1/4 → 2/6 1/3 + 1/3 → 2/6 2/5 + 1/5 → 3/10 Input: 1/2 + 1/3 → ? Big question: Can the machine ever KNOW its answer is wrong?

Make it yours: Language-class twist: the training data becomes five sentences that all repeat the same grammar slip ('He go to school', 'She like apples'…); the machine must complete 'My brother ___ (watch) TV' the same wrong way — same insight, no numbers.

Know AI12 min · Projector

The Species That Never Was

Students read two equally fluent AI-written species profiles — one a real Hong Kong amphibian, one completely invented — vote on which is real, then unpick why the AI 'made up' a species so confidently.

S1 · S2 · S3 · Science

Why this matters: An AI never checks whether something exists — it predicts what a species description usually sounds like, so a made-up answer can look exactly like a true one.

Steps

  1. 1. (2 min) Project the board text. "I asked an AI to introduce Hong Kong amphibians and it wrote both of these. Read them silently — one minute."
  2. 2. (3 min) Vote by hands: only A real / only B real / both real / neither. Tally on the board, then ask two or three students to defend their vote using clues in the text.
  3. 3. (1 min) Reveal: A is the real Hong Kong newt. B is entirely invented — the species, the Latin name, the golden stripes and the 'fewer than 500' figure exist nowhere.
  4. 4. (4 min) Pairs: "The AI wasn't lying — why is 'lying' the wrong word?" Collect answers; steer to: lying requires knowing the truth. The AI only predicts what a species profile usually sounds like — real profiles mention habitat, diet and numbers, so the fake one does too.
  5. 5. (2 min) Exit: "Next time an AI gives you a scientific fact complete with a Latin name and a statistic — what does today tell you to do?" Cold-call two students.

Put on the board

An AI wrote BOTH of these. Which species is real? A. The Hong Kong newt (Paramesotriton hongkongensis) is the only salamander species found in Hong Kong. It lives in clean, rocky hill streams and grows to about 11–15 cm. It feeds on small creatures such as worms and tadpoles, and moves to stream pools to breed in the cooler months. It is protected by law in Hong Kong. B. The Hong Kong golden-lined salamander (Paramesotriton auripictus) is found only in the upper streams of Tai Mo Shan. It is easily recognised by the two golden stripes along its back. It feeds mainly on mosquito larvae and small snails, and shelters under leaf litter in winter. Scientists estimate fewer than 500 remain in the wild. VOTE: Only A real? Only B real? Both? Neither?

Make it yours: No projector? Read both profiles aloud twice at dictation pace and write only the two species names and key numbers on the board — the vote and reveal work the same. Humanities twist: swap in two AI-written biographies, one real historical figure, one invented.

Use AI well12 min · Projector

Prompt Detectives

The class reads three AI answers on the same festival topic, plays detective to reconstruct the prompt behind each one, then upgrades the weakest prompt themselves.

P5 · P6 · S1 · Chinese Language · English Language

Why this matters: The quality of AI's answer is set by the prompt — naming the audience, the format and the concrete details is what turns a bland answer into a useful one.

Steps

  1. 1. (3 min) Project the three answers only (keep the prompts covered). Say: "The same AI wrote all three of these, on the same topic. Which one would get the best mark in our class, and why?" Take two or three hands.
  2. 2. (3 min) In pairs, students play detective: for each answer, guess in one sentence what instruction the writer must have typed.
  3. 3. (2 min) Reveal Prompts A–C. The class matches each prompt to its answer by show of hands; confirm the matches.
  4. 4. (2 min) Ask: "In the winning prompt, which words did the work?" Circle them together: who it is for, what format, which details (past tense, one smell, one sound).
  5. 5. (2 min) Quick check: everyone upgrades Prompt B by adding one 'who', one 'format' and one 'detail'. Pick a few students to read theirs aloud.

Put on the board

THE SAME AI, THE SAME TOPIC — THREE ANSWERS Answer 1: "The Mid-Autumn Festival is a traditional festival celebrated in many Asian countries. It usually falls in September or October. People eat mooncakes, admire the full moon and spend time with family. It is linked to old legends such as Chang'e flying to the moon." Answer 2: "The Mid-Autumn Festival is a happy time for families in Hong Kong. We eat mooncakes and play with lanterns. The full moon is big and bright that night." Answer 3: "Last night Grandma and I walked around Victoria Park under a huge yellow moon. The air smelled of candle wax and sweet lotus paste. Children laughed and rattled their plastic lanterns all around us. It was my best Mid-Autumn night ever!" — REVEAL LATER — Prompt A: "Write a four-sentence diary entry by a P5 pupil who spent Mid-Autumn night at Victoria Park with Grandma. Use the past tense. Include one smell and one sound." Prompt B: "Tell me about the Mid-Autumn Festival." Prompt C: "Write three simple sentences about the Mid-Autumn Festival for a Primary 5 student in Hong Kong." (Match: 1–B, 2–C, 3–A)

Make it yours: No projector? Read the three answers aloud in three different voices and write only the three prompts on the board for the matching step. Maths twist: use three AI explanations of fractions instead — one textbook-dense, one in three plain sentences, one as a pizza-sharing story.

Use AI well10 min · Projector

Send It Back!

The class rejects AI's jargon-heavy first answer on photosynthesis, competes to write the follow-up instruction that would fix it, then compares with the real improved second draft.

P6 · S1 · S2 · Science

Why this matters: AI's first answer is only a first draft — you get a good one by telling it exactly what is wrong and exactly what you need, like a teacher marking homework.

Steps

  1. 1. (2 min) Project Draft 1 only. Say: "This is AI's answer to 'Explain photosynthesis'. Thumbs up if you could revise for the test from this." Collect the complaints: too many hard words, no example, one giant sentence.
  2. 2. (3 min) Say: "Good news — AI never gets offended. In pairs, write ONE follow-up instruction. Rule: no lazy 'make it better' — say what is wrong AND what you want instead, like a teacher marking homework."
  3. 3. (2 min) Invite three or four pairs to read their instructions aloud; the class votes for the most specific one.
  4. 4. (2 min) Reveal the real follow-up and Draft 2. Ask: "Which pair came closest? What did this instruction spell out — audience, steps, analogy, word difficulty?"
  5. 5. (1 min) Quick check: everyone completes in their exercise book: "When AI gives me a weak answer, I will ___."

Put on the board

AI'S FIRST DRAFT — prompt: "Explain photosynthesis" "Photosynthesis is the biochemical process by which photoautotrophic organisms convert light energy into chemical energy, synthesising glucose from carbon dioxide and water within the chloroplasts, with oxygen released as a by-product via the light-dependent reactions and the Calvin cycle." YOUR JOB: write ONE follow-up instruction. It must say (a) what is wrong with the draft, (b) exactly what you want instead. — REVEAL LATER — The follow-up that was actually sent: "Too hard. Explain it again for a Primary 6 student, in three numbered steps, using a kitchen analogy. No word harder than 'energy'." AI'S SECOND DRAFT: 1. A leaf is like a tiny kitchen. Its ingredients are water from the roots and carbon dioxide from the air. 2. Sunlight is the cooker: it gives the energy to turn those ingredients into sugar — the plant's food. 3. The leftover, oxygen, is pushed out into the air — the same oxygen we breathe.

Make it yours: No projector? Write only the first line of Draft 1 on the board and read the rest aloud in a robot voice. Humanities twist: use a bland AI answer to "Why did Hong Kong's population grow after 1945?" and have students coach it to add one real example and one date.

Use AI well12 min · Board only

The Push-Up Robot

Students vote scenario by scenario on whether five ways of using AI for maths make their brain stronger or just do their push-ups for them, then derive the class rule for when NOT to use AI.

P6 · S1 · S2 · S3 · Mathematics

Why this matters: AI is worth using only when it makes you think more, not less — if it does the thinking part, you skip the workout and stay weak.

Steps

  1. 1. (2 min) Hook: "Imagine I buy a robot that does your push-ups for you in every PE lesson. On fitness-test day, who's in trouble?" Take answers, then announce: today we test which ways of using AI are push-up robots for the brain.
  2. 2. (4 min) Read Scenarios A–E aloud one at a time. After each, the whole class votes: arms raised = my brain gets stronger; arms folded = the robot is doing my push-ups; hand wobbling = it depends. Jot each vote split on the board.
  3. 3. (3 min) Take the most divided scenario (usually E). One voice from each side gives a reason; keep pressing the same question: "In this scenario, who did the thinking?"
  4. 4. (1 min) Complete the class rule on the board together: "AI helps my learning only when the one doing the thinking is ME."
  5. 5. (2 min) Quick check: in their exercise books, students write one maths task this week where they will refuse to ask AI for the answer, and one way to make AI push them to think harder.

Put on the board

THE PUSH-UP TEST — is AI making you stronger, or lifting for you? A. The night before the test, you photograph all ten revision questions and copy AI's answers. B. Stuck for ten minutes, you write out your own working, then ask AI which step went wrong. C. You ask AI for the final answer first, then write the "working" backwards to match it. D. After finishing your homework, you ask AI to set you two similar questions to test yourself. E. AI checks your finished homework and finds three mistakes; you correct them and hand it in. Vote: arms raised = brain gets stronger | arms folded = robot push-ups | wobble = it depends CLASS RULE (finish it together): AI helps my learning only when the one doing the thinking is ______.

Make it yours: Language-class twist: swap in composition scenarios (copying a whole AI essay; asking AI to quiz you on this week's vocabulary; writing your own opening first, then asking AI for three stronger alternatives). If the class needs to move, run it as a four-corners walk instead of hand votes.

Question AI10 min · Board only

Mark the AI's Homework

Students play teacher: they mark a confident AI solution to a division word problem and hunt down the mistake it never noticed.

P4 · P5 · P6 · Mathematics

Why this matters: A confident tone is not evidence — AI never checks its answer against the real world, so the marker (you) must.

Steps

  1. 1. (1 min) Write the board text up and announce: "An AI did this maths question last night. Today YOU are the teacher. Full marks is 3 — decide what it deserves."
  2. 2. (3 min) In pairs, students check the working line by line — "Is this step actually true?" — then award a mark out of 3 and write one teacher-style comment.
  3. 3. (3 min) Collect marks by show of hands (3? 2? 1? 0?). Ask a pair who gave low marks to explain: "So how many students do 8 minibuses really carry?" (16 × 8 = 128 — two children are left standing at the school gate!)
  4. 4. (2 min) Ask the class: "Every line of working looked so neat — why is the answer still wrong?" Draw out: the AI followed a rounding rule without ever asking whether the answer makes sense for real people.
  5. 5. (1 min) Close with the wall rule: "Confident is not the same as correct — mark the working, not the tone."

Put on the board

Question: 130 students are going to Ocean Park by minibus. Each minibus carries 16 passengers. How many minibuses does the school need? AI's answer (mark it out of 3): Divide the total number of students by the seats per minibus: 130 ÷ 16 = 8.125 Rounding to the nearest whole number gives 8. The school needs 8 minibuses. Final answer: 8 ✓

Make it yours: Reusable with any topic: take one question from this week's exercise, write your own confident 'AI answer' with a single hidden slip (wrong unit, rounded the wrong way, misread number), and let pairs mark it.

Question AI10 min · Board only

Two Truths and a Hallucination

The class reads three AI-written 'facts' about Hong Kong, votes on which one is invented, and discovers the fake carries the most convincing number.

P5 · P6 · S1 · Humanities

Why this matters: Hallucinations don't look vague — precise details prove nothing, so the only real test is checking against a source.

Steps

  1. 1. (1 min) Write the three 'facts' on the board (or project them). Say: "I asked an AI for three facts about Hong Kong. Exactly one is wrong. No shouting out — just read."
  2. 2. (2 min) Solo and in silence: each student writes 1, 2 or 3 in their exercise book plus a one-line reason.
  3. 3. (3 min) Groups of four must agree on ONE answer and name exactly where they could prove it — a textbook page, the library atlas, an official website.
  4. 4. (2 min) Vote with fingers, then reveal: No. 2. The Peak really is about 552 m — but Hong Kong's highest mountain is Tai Mo Shan (957 m). The fake sentence had the most precise number in the whole list.
  5. 5. (2 min) Ask: "So how DO you spot an AI hallucination?" Draw out the insight: you can't spot one by looking — confident details prove nothing; only checking a source settles it.

Put on the board

I asked an AI: 'Tell me three facts about Hong Kong.' It replied: 1. The Tsing Ma Bridge opened in 1997 and carries both cars and trains. 2. Victoria Peak is the highest mountain in Hong Kong, at 552 metres. 3. Hong Kong has more than 260 islands. Exactly ONE is wrong. Which one — and where could you prove it?

Make it yours: Subject twist: write your own set — two true statements and one plausible fake about this week's text, author or science topic — and make sure the fake can be exposed on a specific textbook page.

Question AI12 min · Projector

AI vs AI: Who's Right?

Students compare two confident but contradictory AI explanations of why hot air rises, pinpoint the exact clash, and design a way to settle it without asking the AI again.

S1 · S2 · S3 · Science

Why this matters: When two AI answers contradict each other, at least one is confidently wrong — asking the AI again is not verification; an independent source is.

Steps

  1. 1. (2 min) Project the board text. Say: "I asked the same question twice and got these two answers. Read both silently. Do they agree?"
  2. 2. (3 min) Pairs underline the exact phrases that clash. (A says heated molecules 'become lighter'; B says the same molecules just spread out.)
  3. 3. (2 min) Ask: "Can both be right? Can both be wrong?" Establish with the class: they contradict, so at least one AI is confidently mistaken — yet both sounded equally sure.
  4. 4. (3 min) Groups list ways to settle it WITHOUT asking the AI a third time: the particle-model pages of the textbook, a teacher demonstration… Then ask: "Why doesn't asking the AI again count?"
  5. 5. (2 min) Open the textbook and verify together: B matches the particle model — mass unchanged, volume increases, density decreases. Wall rule: "When answers clash, don't vote for the smarter-sounding one — verify."

Put on the board

I asked an AI the same question twice: 'Why does hot air rise?' Answer A: 'Hot air rises because heating makes air molecules lighter. The lighter molecules naturally float upwards, so hot air always moves to the top of the room.' Answer B: 'Hot air rises because heating makes air expand — the same molecules spread out and take up more space. The hot air is now less dense, so the denser cool air sinks and pushes it upward.' They can't both be right. Underline exactly where they disagree.

Make it yours: No projector? Read Answer A aloud slowly while students jot its key claim, then read Answer B — pairs reconstruct the clash from their notes before the discussion.

Fair & safe AI12 min · Board only

Breakfast Bias

The class 'trains' a pretend breakfast AI using data from one group only, then discovers how badly it describes everyone else.

P4 · P5 · P6 · Mathematics

Why this matters: An AI's answers can only be as broad as its training data — when a group is missing from the data, the AI confidently gets that group wrong.

Steps

  1. 1. (1 min) Introduce: "We are going to train a pretend AI called NOM-3 to answer one question — what do Hong Kong students eat for breakfast? An AI only knows its training data, and today its training data is Group 1 only."
  2. 2. (3 min) Ask only Group 1 (about 6–8 students) what they ate this morning; tally their answers on the board under the heading 'NOM-3's training data'.
  3. 3. (2 min) Announce NOM-3's confident answer from that tally alone, e.g. "Hong Kong students eat bread for breakfast." Ask the rest of the class: hands up if NOM-3 just got YOU wrong. Count the hands.
  4. 4. (4 min) Now tally the whole class and put the two tables side by side. Ask: "What did NOM-3 miss? Was its maths wrong?" Draw out the key discovery: the calculation was fine — the sample was too narrow.
  5. 5. (2 min) Connect to real life: real AI is trained on internet text and photos, so whoever posts most gets believed most. Close on the board question: "Who is missing from the data?"

Put on the board

NOM-3 the Breakfast AI Training data (Group 1 ONLY): bread/sandwich | congee/macaroni | instant noodles | nothing | other: ___ NOM-3's answer: 'Hong Kong students eat ______ for breakfast.' Whole-class data: bread/sandwich | congee/macaroni | instant noodles | nothing | other: ___ Think: Who was missing from NOM-3's training data? What did it get wrong about them? Big idea: An AI only 'knows' what is in its training data.

Make it yours: Humanities twist: train NOM-3 on 'how does your family celebrate Lunar New Year?' using one group only — then discuss what an AI trained mostly on Western internet data leaves out when it describes Hong Kong festivals.

Fair & safe AI10 min · Projector

The Privacy Gatekeepers

Students act as 'privacy gatekeepers', hunting down every piece of personal data in a classmate's draft chatbot message and rewriting it into the shortest safe version.

P4 · P5 · P6 · S1 · Any subject

Why this matters: A chatbot never needs to know who you are to help you — anything you type may be stored on company servers, so share the problem, not the person.

Steps

  1. 1. (2 min) Project Ka Ming's draft message (see board text) and have one student read it aloud. Set the role: "Ka Ming wants homework help. Before he presses send — you are his privacy gatekeepers."
  2. 2. (2 min) In pairs, students count every piece of personal information in the message. Race to find them all.
  3. 3. (3 min) Collect answers; strike through each item on screen or board as it is found (there are 8). For each one, ask the class: "Does the AI need this to explain the fractions question?" — every single answer is no.
  4. 4. (2 min) Push deeper: "Where do the words you type actually go?" Draw out: they leave your device and may be stored on company servers or used for training. Offer the test: "Would you say it to a stranger on the MTR who happens to be good at maths?"
  5. 5. (1 min) Class agrees the safe rewrite and teacher writes it up: "Please show me, step by step, how to work out 1/4 + 2/3."

Put on the board

Ka Ming's draft message to the chatbot: 'Hi AI! I'm Chan Ka Ming from Class 5B at Cheerful Road Primary School in Kwun Tong. My phone number is 9123 4567 and I live in Block 2, On Wah Estate. My teacher Miss Wong gave us maths homework and I don't get Q3. Can you show me how to work out 1/4 + 2/3? If you can't reach me, my mum's number is 6234 5678.' Gatekeeper questions: 1. How many pieces of personal information can you find? 2. Which of them does the AI need to answer the maths question? 3. Rewrite it: what is the shortest SAFE version? Rule: Share the PROBLEM, not the PERSON.

Make it yours: No-projector fallback: read the message aloud twice at normal speed; pairs count the personal-data items on their fingers, then call them out for the teacher to list and strike through on the board one by one.

Fair & safe AI13 min · Board only

Before You Forward

Students decide whether to forward a late-night voice clip of 'the principal' suspending classes, build a Stop–Check–Ask routine, then re-vote after learning the voice could be cloned from 15 seconds of audio.

S1 · S2 · S3 · Humanities · ICT / Computer Literacy

Why this matters: AI can clone anyone's voice from a few seconds of audio, so 'I heard it myself' is no longer evidence — verify through official channels before you forward.

Steps

  1. 1. (2 min) Set the scene from the board: 10:07 pm, the class group chat receives a forwarded 15-second voice clip — unmistakably the principal's voice — announcing school is suspended tomorrow. It is already in five other groups. Vote by show of hands: forward it, or not?
  2. 2. (3 min) In pairs, list every clue that could reveal whether the clip is real or fake. Prompt: think about the sender, the timing, the channel, and how your school actually makes announcements.
  3. 3. (4 min) Collect clues on the board and sort them into Stop–Check–Ask (see board text). Most pairs will fixate on the content; steer them towards the source and the channel — the strongest checks.
  4. 4. (3 min) The reveal: today's AI can clone a voice from about 15 seconds of audio — one speech-day video on the school's Facebook page is enough. Re-vote: forward or not? Ask two or three students what changed their mind.
  5. 5. (1 min) Exit line: every student completes, aloud or in their exercise book, "Before I forward, I will ______."

Put on the board

10:07 pm — Class group chat Forwarded ▶ voice clip (0:15): 'This is Principal Lam. Due to a burst water main, classes are suspended tomorrow. Students need not return to school.' 'my cousin's school got it too!!' 'is this REAL??' STOP – CHECK – ASK: STOP — do not forward until it is verified. CHECK — the official channels: school website, parent app, EDB announcements. Is the same message there? ASK — who recorded the ORIGINAL clip, and when? Does school ever announce suspensions by group-chat voice message? Warning: AI can clone a voice from about 15 seconds of audio. Hearing is not believing.

Make it yours: Language-class twist: after the Stop–Check–Ask step, students draft the two-sentence reply they would actually post in the group chat — stopping the rumour without embarrassing whoever forwarded it. Persuasive writing with a real audience.

AI & me10 min · Board only

Where Is the Line?

Students place five AI-homework scenarios on a line from '100% my work' to 'not my work' and argue out exactly where honest use of AI ends.

P6 · S1 · S2 · S3 · English Language · Chinese Language

Why this matters: Students walk away with a test for academic honesty they can apply themselves: work is yours when the thinking happened in your head — not measured by how many words you typed.

Steps

  1. 1. (1 min) Draw one long line across the board. Label the left end "100% my work" and the right end "Not my work at all". Say: "Five students all used AI on the same essay homework. Your job: decide where each one sits on this line."
  2. 2. (3 min) Reveal the five scenarios (board text) and read A to E aloud. In pairs, students place all five on the line — partners must agree on an order.
  3. 3. (3 min) Poll the class scenario by scenario: point at positions on the line, hands up, mark where most of the class lands. A and E are easy; expect real disagreement on C and D. Pick two students who disagree about D and let them argue it out in front of the class.
  4. 4. (2 min) Ask: "We can't settle this by feeling. What TEST could tell us whether it's still your work?" Collect ideas, then underline the test question on the board: could you do it again tomorrow, without AI?
  5. 5. (1 min) Close: the line is not about how many words you typed — it's about where the thinking happened. Each student tells their partner, in one sentence, which scenario is their personal line and why.

Put on the board

THE LINE: 100% my work ←——————————→ Not my work at all Five students, same essay homework. Where does each one sit? A. Kiu asks AI to explain what the question is asking, then plans and writes everything herself. B. Marco asks AI for five possible ideas, picks one, and writes the whole essay himself. C. Wing writes her essay first, then asks AI to fix her grammar and copies the corrections in. D. Jason writes one rough paragraph, asks AI to "make it longer and better", and hands in the result. E. Tracy pastes the question into AI, changes a few words of the answer, and hands it in. THE TEST QUESTION: Tomorrow, without AI — could you do it again?

Make it yours: Maths/Science twist: swap the essay for a problem set — A asks AI to explain the method, C finishes first then asks AI to check the answers, E photographs the question and copies the worked solution. The line lands in exactly the same place.

AI & me10 min · Board only

The Calculator Trap

A quick mental-maths race and one booby-trapped calculator answer let the class discover that only people who can estimate can tell when the machine is wrong.

P4 · P5 · P6 · Mathematics

Why this matters: Hand a skill entirely over to a tool and you also lose the power to spot when the tool is wrong — true for calculators, truer for AI.

Steps

  1. 1. (2 min) Round 1: uncover the three mental sums on the board. Brains and pencil only, 90 seconds, then reveal the answers: 200, 144, 594.
  2. 2. (2 min) Provoke: "A calculator does all three in one second. Hands up — who thinks schools should stop teaching mental maths?" Take two or three opinions from each side.
  3. 3. (3 min) Round 2: point at the board — "A student's calculator shows 62 × 48 = 3,976. You are NOT allowed to calculate. Is it right?" Give pairs one minute. Someone will estimate 60 × 50 = 3,000, so 3,976 must be wrong — a finger slipped on the keypad. Have that student explain the method to the class.
  4. 4. (2 min) Ask: "The calculator itself never miscalculates — so who caught the error, and what skill did they use?" Draw out the discovery: only people who can estimate know when the machine is wrong.
  5. 5. (1 min) Bridge: fill in the last line of the board together — "AI exists. Why still learn to write / think / work things out?" Land it: a tool makes a skilled person faster, and an unskilled person helpless.

Put on the board

ROUND 1 — brains only (90 seconds): 1) 25 × 8 = 2) 302 − 158 = 3) 6 × 99 = ROUND 2 — the calculator says: 62 × 48 = 3,976 No calculating allowed. Right or wrong? THE BIG QUESTION Calculators exist. Why still learn mental maths? AI exists. Why still learn to ______?

Make it yours: Language twist (English or Chinese lessons): replace the sums with the sentence "Their going two the park on Sunday." — spellcheck finds zero errors; only someone who knows the grammar can. Same trap, different subject.

AI & me12 min · Projector

The Paragraph From Nowhere

The class grades a fluent AI description of 'a market', discovers it could describe anywhere on Earth, then beats it with details only someone who has stood in a real Hong Kong street market could write.

P5 · P6 · S1 · S2 · Chinese Language · Humanities

Why this matters: AI writes the average of a million people's words and has never stood anywhere — your own observations and memories are the part of writing no machine can supply.

Steps

  1. 1. (2 min) Project the AI paragraph and read it aloud together. First question: "Grammar and structure — mark out of 10?" Quick show of fingers; the class will score it high.
  2. 2. (2 min) Second question: "Which market is this? Which district? Raise your hand if this paragraph could equally describe a market in Singapore, Taipei or Bangkok." Let the discomfort land: it fits everywhere because it has been nowhere.
  3. 3. (3 min) Challenge, in pairs: think of ONE real market, shop or street you actually know. Write one sentence with a detail the AI could never produce — a smell, a sound, a particular stall owner, water on the floor, the exact words someone shouts.
  4. 4. (3 min) Collect five or six sentences aloud and write the best details on the board under the AI paragraph. Ask: "Which version makes you SEE a place? Which writer would you rather read?"
  5. 5. (2 min) Land it: the AI paragraph scores high for grammar and zero for having lived. When you write, start from what only you know; if you use AI at all, it comes after your details — never instead of them.

Put on the board

AI wrote this: "The market is a lively and bustling place. There are many stalls selling fresh fruit, vegetables, meat and fish. Vendors call out to attract customers, and the air is filled with different smells. People bargain to get the best prices. The market is full of energy and shows the vibrant life of the city." Grammar: ___/10 Makes you SEE a real place: ___/10 YOUR CHALLENGE: one sentence, one detail the AI could never know.

Make it yours: No projector? Read the AI paragraph aloud twice, slowly, and copy only its first and last sentences onto the board — the blandness survives even out loud. Humanities twist: swap the market for 'a typical Hong Kong public housing estate'.

Print-and-go classroom kit

Two ready-made A4 sheets for the part of AI literacy that happens outside your lesson — homework and home. Both are bilingual on one page, so they work in any HK school.

AI rules for this assignment 本次功課AI使用規則

Tick what's allowed, staple to the worksheet. Includes a student declaration line.

A letter to parents 致家長的信

One page explaining how families can handle AI at home — conversation, not bans.