Teach the Machine

Big idea: A machine learns by looking at the examples we show it. It looks for what the examples have in common — and then it can guess about something new.

A brand-new machine knows nothing. If we want it to tell a cat from a dog, we show it pictures: "this is a cat… this is a cat… this is a dog…" The machine looks for what the cat pictures have in common — maybe pointy ears, maybe whiskers. Then we show it a brand-new picture and ask: cat or dog?

That is why the examples we pick really matter. A machine can only learn from what we show it. If every cat we showed was tiny, the machine learns "cats are always tiny" — so when a big cat comes along, it guesses wrong. It is not being silly. It is doing exactly what our examples taught it.

The fix is simple: show it the kinds of things it will really see. Big cats and small cats. Cats sitting and cats running. Good examples make a good machine.

Essential question: How does a machine learn something new?

Teacher Guide — Teach the Machine

Sessions: 2 × 20 min (see grade-banded pacing) · Format: 100% unplugged — no devices, no student logins, no student data collected.

At a glance

• Big idea: A machine learns by looking at the examples we show it. It looks for what the examples have in common — so the examples we pick decide what it learns. Show it only tiny cats, and it learns "cats are tiny."
• Essential question: How does a machine learn something new?
• You do not need to be a computer scientist to teach this. Every move is scripted below; every material is a common classroom item or a printable in this unit's Printable Materials. The single most important moment — the "confused machine" demo — is fully scripted in Step 5.

Learning targets (kid language)

1. I can give a machine good examples to teach it a group (all cats, all dogs).
2. I can guess what a trained machine will answer for a new card.
3. I can tell why the machine guessed wrong — "we only showed it tiny cats."
4. I can sort cards into two groups and say what they have in common.

Standards (what each target proves)

Target	VA CS SOL (2024)	Plain gloss
Sort examples into one group by a shared attribute	K.AP.1	identify patterns; sort by an attribute
Compare two groups and say what each has in common	2.AP.1	compare/contrast objects by attributes
Use the examples shown to predict the next answer	K.DA.2	use data to make a prediction / draw a conclusion
Decide from the example set why a guess went wrong	2.DA.1	analyze data to make a decision (no device)

VDOE AI Guidance: Do No Harm (the confused-machine demo), Empower Student Success.

Materials (all common, all reusable)

• The example cards (cats & dogs), the "tiny-cats-only" bad-example set, the new-card test pictures, and the two-box teaching mat — print/cut from this unit's Printable Materials (one set per small group; one demo set for you).
• Two bins, hoops, or taped boxes on the floor labeled with the two group cards.
• Chart paper for the class "What Good Examples Look Like" list.

Pacing by grade

	Active time	Adjust
K	2 × 10-12 min	Teach ONE group (all cats); do the confused-machine demo longer; sort 4 cards
1	2 × 15 min	Teach two groups; add the "guess the new card" beat; sort 6 cards
2	2 × 20 min	Teach two groups; add "what do the two groups each have in common?"; sort 8 cards

If you only have 10 minutes: do the Hook (teach-me-a-cat) + the confused-machine demo (Step 5). That alone carries the unit's crown-jewel idea and meets K.DA.2.

---

Lesson plan

1. Hook — "I am a brand-new machine" (3-4 min)

Sit down, go still, and announce: "I am a brand-new machine. I know nothing about cats. Teach me — show me cat pictures." As children hold up cat cards, think aloud about what you notice them having in common: "Hmm… pointy ears… pointy ears again… whiskers… I am looking for what these all have in common." This is the whole unit in one move.

Why this works: it makes "learn from examples" a thing children do to you, not a thing they watch. Note the exact language: the machine looks for what the examples have in common — never say it "finds the rule" or "isn't memorizing." That overstates what it does.

2. I do — teach two groups (4-5 min)

Place the two-box teaching mat ("Cats" / "Dogs"). Sort a few cards into each, narrating: "These go in Cats — they have pointy ears and whiskers. These go in Dogs — floppy ears, big nose." For Grade 2, name what each group has in common out loud — that naming is the 2.AP.1 beat.

Keep it to two groups only at K-2. Two choices is enough to teach the whole idea.

3. We do — guess the new card (5-6 min)

Hold up a new card the class has not sorted. Ask: "I learned from your examples. What will I guess — cat or dog? Why?" Take a vote with thumbs, then reveal. Model the sentence: "You guessed cat because it has the same things the cat examples had." Predicting the answer from the examples shown is the K.DA.2 beat.

4. You do — small-group teach-and-test (5-6 min)

Groups use their card set: they pick good examples to teach one group, then quiz a partner with a new card. Circulate and ask one child per group: "Why did you pick these examples to teach 'cat'?" (Accept a point, a gesture, or a home-language answer — see Differentiation.)

5. Closure / crown jewel — the "confused machine" demo (4-5 min) — do not skip

This is the most important moment in the entire K-2 band. Set it up secretly: this time, teach yourself using the "tiny-cats-only" bad-example set — every cat card is a tiny cat drawn in the corner. Narrate as you "learn": "Tiny cat… tiny cat… tiny cat. Got it — cats are tiny."

Now hold up the big-cat test card. Confidently guess wrong, with full commitment: "That's too big to be a cat. That must be a dog!" Let the children erupt and correct you. Then ask the key question:

"Wait — why did I get it wrong? I looked carefully at all my examples!"

Guide them to the answer in their own words: "You only showed it tiny cats." Land it: "My examples didn't cover the kinds of cats I would really see. To teach me right, you have to show me the kinds of things I will really see — big cats and small cats." Add one good example, re-test, and let the machine "get it right." Deciding from the example set why the guess failed is the 2.DA.1 beat and the Do No Harm beat.

Capture evidence here. This single demo gives you a look-for on every child at once (see Assessment). Add their fix to the class chart: "Good examples cover the kinds of things it will really see."

---

If you only have N minutes

• 10 min: Hook + Step 5 (confused-machine demo). Crown-jewel idea intact; meets K.DA.2.
• 15 min: add Step 3 (guess the new card).
• 20 min: full sequence including the small-group teach-and-test.

Common misconceptions

• "The machine already knows what a cat is." → No. It knew nothing until your examples; it only looks for what those examples have in common.
• "The machine finds the rule / it isn't really learning." → Avoid this phrasing yourself. Say it looks for what the examples have in common — that is honest about what it does and what it can get wrong.
• "The machine is just dumb when it guesses wrong." → No — it guessed exactly what the examples taught it. Blame the examples, not the machine. (This is the whole point of Step 5.)
• "More examples is always better." → Not quite. What matters is that the examples cover the kinds of things it will really see — big cats and small cats — not just a bigger pile of the same tiny cat.

Background (teacher notes)

This is supervised learning in child form: we show labeled examples, the model looks for what they share, and it generalizes to new cases. The confused-machine demo previews one of the most important ideas in all of AI: a model inherits the blind spots of its examples. Be precise in your own language — a model does not "understand" or "find the rule"; it looks for what the examples have in common, which is why unrepresentative examples (only tiny cats) produce confident, wrong answers. That precision is exactly what 6-8 ("training data," "data quality") and 9-12 ("dataset bias," "representativeness") build on. You are planting the seed, not the whole tree — keep it concrete and kind to the machine.

Safety & privacy note

This unit is fully unplugged: no devices, no logins, no student data. Keep all examples about animals or objects (cats, dogs, fruit) — never sort pictures of people. Teaching "the examples decide what it learns" with pictures of children would model exactly the harm this lesson warns against. If a child connects the idea to a real app at home, validate the thinking and defer specifics to families.

Spiral — where this goes next

• Unit 2 (before): children learned a machine guesses what comes next from a pattern. Here they learn where that pattern comes from — the examples we choose.
• Unit 4 (next): good examples → fair machine becomes fair and safe — what happens when the examples leave people out.
• 3-5 → 9-12: examples → training data → a labeled dataset; "we only showed it tiny cats" → data quality → dataset bias and representativeness.

Printable Materials — Teach the Machine

How to use: Print one copy per small group on cardstock if you can, cut along the lines, and reuse all year. Each card has a picture and a word so non-readers can play. Nothing here needs a device. (K: teach one group with the 4 cat cards. Grade 1: use 6 cards — 3 cat and 3 dog. Grade 2: use all 8 cat and dog cards, add the test cards and the compare beat.) Print one extra demo set for yourself — you will need the Material C bad-example set for the confused-machine demo in the Teacher Guide.

Material A — Good example cards (cut apart)

These are the good examples — they cover the kinds of cats and dogs a machine would really see (big and small, sitting and running).

🐱 Cat<br/>small, sitting	🐶 Dog<br/>small, sitting
🐈 Cat<br/>big, standing	🐕 Dog<br/>big, standing
🐈‍⬛ Cat<br/>running	🦮 Dog<br/>running
😺 Cat<br/>pointy ears, whiskers	🐕‍🦺 Dog<br/>floppy ears, big nose

Material B — Two-box teaching mat (one per group)

🐱 Cats	🐶 Dogs
(what do they have in common?)	(what do they have in common?)

Material C — "Tiny-cats-only" bad-example set (TEACHER DEMO — Step 5)

Use only these to "teach yourself" in the confused-machine demo. Every cat is tiny and in a corner — on purpose. They do not cover the kinds of cats the machine will really see.

🐱 Cat<br/>tiny, in a corner	🐱 Cat<br/>tiny, in a corner	🐱 Cat<br/>tiny, in a corner

Material D — New test cards (the "catch me" cards)

Hold one up after training. Use the big cat for the confused-machine demo.

🦁 Big cat<br/>the surprise test	🐕 Big dog<br/>the exit-ticket card

Material E — Class chart sentence strip (optional)

✅ What good examples look like
They cover the kinds of things it will really see.
Big cats AND small cats. Sitting AND running.

---

Teacher answer key (do not print for students)

• Cats group (Material A): the four 🐱/🐈/🐈‍⬛/😺 cards. Shared features children may name: pointy ears, whiskers, thin tail.
• Dogs group (Material A): the four 🐶/🐕/🦮/🐕‍🦺 cards. Shared features: floppy ears, bigger nose, often a wagging tail.
• Material C is the trap. Teaching only tiny-corner cats makes the machine "learn" cats are tiny. When you show the 🦁 big-cat test card, guess dog with full confidence — then let the children catch you. The target answer to "why did it guess wrong?" is "you only showed it tiny cats" → the examples did not cover the kinds of cats it would really see.
• Fix it live: add one big-cat good example, re-test, and let the machine get it right.
• Accept reasoned disagreement ("a hairless cat has no whiskers showing") — a child arguing from the examples is exactly the skill this unit assesses (K.AP.1, 2.AP.1, 2.DA.1).

Vocabulary — Teach the Machine

Five child-sized words. Post them on the word wall; they spiral forward — good examples (this unit) becomes fair and safe (Unit 4), and training data in the 6-8 band.

Word	Kid-friendly meaning	Say it in a sentence
example	One thing we show the machine to teach it.	"This cat picture is an example."
teach / train	To show a machine many examples so it can guess.	"We teach it by showing lots of cats."
learn	When the machine looks for what the examples have in common.	"Now it has learned what cats have in common."
good examples	Examples that cover the kinds of things it will really see.	"Big cats and small cats are good examples."
wrong guess	When bad examples make the machine choose wrong.	"Only tiny cats led to a wrong guess on a big cat."

Anchor pair

The engine of this unit is examples → learn. And the warning that powers Unit 4: bad examples → wrong guesses. Say both pairs together every day.

Spiral note (for teachers)

Examples becomes "training data" in 6-8, and "a labeled dataset" in 9-12. Good examples — examples that cover the kinds of things it will really see — becomes "data quality," "representativeness," and "bias." Keep your own language precise: the machine looks for what the examples have in common — it does not "find the rule" or "understand." You are planting the single most important vocabulary seed of the band: a machine inherits the blind spots of the examples it was given. Carry good examples straight into Unit 4.

Differentiation & Access — Teach the Machine

Designed against Universal Design for Learning: multiple means of representation, expression, and engagement. The activity is hands-on and language-light by default — children can show full mastery by handing you cards, with no speech required.

Support (emerging learners & IEP)

• Keep groups binary and concrete (cat vs. dog; apple vs. banana).
• Teach one group only (all cats) before adding a second.
• Use real photos instead of drawings if abstraction is hard.
• Let the child physically hand you example cards to "teach" — action over explanation.
• For the confused-machine demo, narrate the cause aloud for them: "We only showed it tiny cats, so it learned 'cats are tiny' — that's why it guessed wrong."

English learners

• The teach-the-teacher role-play carries the whole idea with very little language.
• Pair example / teach / learn with gestures (hold up a card / point to your head / thumbs up).
• Accept the wrong-guess explanation in the home language, then echo back "only tiny cats."
• Post photos + the word together (example, teach, learn, good examples) on the word wall.

Extension (advanced learners) — classroom-anchored

• Ask the child to design a good example set on purpose so the machine will not be fooled — big and small, sitting and running. (Links to 2.AP.1, comparing by attributes.)
• Pose: "What kinds of cats does our machine need to see so it won't be surprised?" — steering them to "cover the kinds of things it will really see," not just "more cards."
• Gather and chart real classroom data with no device: tally classroom pets or favorite fruits on chart paper, then ask, "If a machine only learned from our class, what kinds of pets might it miss?" (Links to K.DA.2 / 2.DA.1 — use data to predict and decide.)

Access & accommodations

• Vision / color: use high-contrast cards; read each card aloud; tell cards apart by the word + picture (never color alone). The "tiny vs. big cat" contrast must be obvious by size and label, not by color.
• Hearing: pair every spoken cue with a gesture or a held-up card; face the child when speaking; the teach-and-test can be done entirely by placing cards, so no step depends on hearing.
• Non-speaking / AAC: accept teaching and predicting by placement, pointing, eye-gaze, or an AAC device — a child shows full mastery by which cards they choose, with no speech.
• Fine-motor: use large cards, a pocket chart, or a floor sort with hoops instead of small manipulatives.
• Attention: the confused-machine demo is the priority — protect those minutes even if you trim the rest; keep the share-out short and offer a quiet "show me" option.

Teach the Machine — Assessment

K-2 assessment is observational and performance-based — no worksheets, no reading load. Use the look-fors during the teach-and-test activity and the confused-machine demo. Each look-for names the standard it gives evidence for, so the alignment is verifiable at a glance.

Learning targets

A child who has met this unit can:

1. Pick good examples to teach a machine one group, and say what they have in common. (K.AP.1)
2. Compare the two trained groups and say what each group has in common. (2.AP.1)
3. Predict what a trained machine will answer for a new card. (K.DA.2)
4. Explain why the machine guessed wrong by pointing to the examples — "we only showed it tiny cats." (2.DA.1; Do No Harm)

Observational checklist (circle one per child)

Look-for	Standard	Not yet	Developing	Got it
Sorts example cards into one group and names a shared feature (pointy ears)	K.AP.1	○	○	○
Says what each of the two groups has in common (cats vs. dogs)	2.AP.1	○	○	○
Predicts the machine's answer on a new card from the examples shown	K.DA.2	○	○	○
Explains a wrong guess by the example set ("only tiny cats"), not "it's dumb"	2.DA.1	○	○	○

Performance task — "Teach me, then catch me" (5-7 min, 1:1 or small group)

1. The child teaches you a group with example cards. (K.AP.1)
2. You show a new card; the child predicts your answer. (K.DA.2)
3. Run the bad-data twist: you re-teach yourself secretly on the tiny-cats-only set and guess wrong on a big cat. The child explains the failure. (2.DA.1; Do No Harm)

• Meets: picks examples that cover the kinds of cats it would really see (big and small) and explains the wrong guess by the examples — "you only showed it tiny cats."
• Approaching: teaches and predicts, but blames the machine ("it's silly"), not the examples.
• Reteach: thinks the machine "already knew" the answer → re-run the brand-new-machine Hook.

Exit ticket (whole class, 1 min)

"A machine only ever saw pictures of small dogs. Now it sees a big dog. What might it guess?" A strong answer connects the example set to the wrong guess — "it might say not-a-dog, because it only saw small ones." (2.DA.1; VDOE Do No Harm.) This is the most important idea in the K-2 band — log who gets it and reteach with Step 5 if many miss it.

Evidence to keep: one photo of the class "What Good Examples Look Like" chart documents targets 1, 2, and 4 at once. The exit-ticket tally documents target 3-4 reasoning.

Family Letter — Teach the Machine

Dear family,

This week your child learned that a machine learns from the examples we show it — it looks for what the examples have in common. That means the examples really matter: if a machine only ever sees tiny cats, it learns "cats are tiny" and guesses wrong on a big cat. The fix is to show it the kinds of things it will really see.

Try this together — no device needed: play "teach a grown-up." Have your child pick a group — say, red foods — and "teach" you by naming a few examples (apple, strawberry, tomato). Then test you with a tricky new one (a green apple, or a banana). Ask your child: "Why might a machine guess wrong here?" A wonderful answer is "It didn't see enough good examples." No items handy? Just talk it through, or play it on a walk pointing at things you pass. Your child can play this with a trusted adult, including their teacher — there is no wrong way to do it.

One habit to keep: when something gets a guess wrong, ask "what examples did it learn from?" — a kind, curious question that works for machines and for people.

Thank you for learning alongside us.

---

This letter is available in other languages — just ask your child's teacher. Esta carta está disponible en otros idiomas — pregunte al maestro de su hijo/a.

Notice, Predict, Decide

A home or classroom scavenger hunt for machines.

This activity uses the same observation skill a scientist uses: look closely, notice a detail, and use it to predict what happens next.

What to do:

1. Walk around your home, classroom, or school with a grown-up or a partner. Find five machines — anything from a light switch to a tablet to an automatic door.
2. For each one, sort it into a pile: "Does the same thing every time" or "Notices something and decides."
3. For the machines in the "notices and decides" pile, name the part that notices — is it a camera, a microphone, a button, or something else? That part is called a sensor.
4. Pick one "notices and decides" machine. Before it acts, predict what it will do next — then watch and check your prediction.

Essential question: What makes a machine seem smart?

Sorting chart

Machine	Does the same thing every time	Notices and decides
Example: light switch	X
Example: tablet that knows your face		X
1.
2.
3.

Talk about it: Ask a grown-up, "Which machine surprised you the most?" A machine that notices and decides is doing the very beginning of what we call artificial intelligence — and every machine on your list is a real, physical thing you can touch, built by real engineers.

Smart and Not-So-Smart

Big idea: Some machines just do one thing the same way every time. Other machines seem to notice things and decide — we call those smart machines.

A light switch is a machine, but it isn't smart: you flip it up, the light turns on. Every time. It never decides anything.

Now think about a tablet that knows your face, a speaker that answers when you talk, or a car that beeps when something is behind it. Those machines notice something about the world — with a part called a sensor, like a camera to see or a microphone to hear — and then they choose what to do. That noticing-and-choosing is the start of artificial intelligence — machines that act a little bit clever.

Look around your day. Which machines just do the same thing every time? Which ones notice and decide? Smart machines are everywhere once you start looking.

Essential question: What makes a machine seem smart?

Patterns Everywhere

Big idea: A pattern is something that repeats. When you find the pattern, you can predict what comes next.

Clap with me: clap–clap–stomp, clap–clap–stomp, clap–clap… what comes next? You knew it was stomp because you found the pattern.

Patterns are everywhere: red–blue–red–blue beads, day–night–day–night, the steps of your morning. When you guess what comes next, you are doing the same job a smart machine does. A smart machine looks at lots of examples, finds the pattern, and predicts what comes next.

That is why a tablet can finish your word before you do, or a music app can guess a song you might like. It found a pattern in what came before. Smart machines are good at spotting patterns — but a pattern can be new or broken, so a machine can guess wrong too.

Essential question: How does finding a pattern help us guess what comes next?

Being Fair and Safe with Smart Machines

Big idea: A smart machine is a tool. People decide how to use it. We use it in ways that are fair, safe, and honest — and people decide what is right.

We learned that a machine only knows what people show it (Unit 3). If people show it unfair examples, it can make unfair guesses. So people work hard to give a machine good, fair examples — so the machine works well for everyone.

We also keep ourselves safe. A smart machine is not a person and not your friend. We keep private things private — like a name, an address, or a password. If a machine ever asks for something that feels wrong, the rule is the same for everyone: stop and ask a trusted adult, including your teacher.

And we are honest. If a machine helps make something — like helping draw a picture — we say so. A smart machine is a helper. People are the ones who decide what is right.

Essential question: How do we use smart machines in a way that is fair, safe, and honest?