AIQ
AIQ
Part of the Finding Patterns lesson guide. Teaching a different grade? 🌈 Explorer (5–7) · 💻 Hacker (11–14) · ⚡ Architect (15–18)
Start by writing a list of 15–20 random numbers on the board (or project a spreadsheet with a column of random numbers) with no headers, no explanation, nothing. Ask: "If I told you 'find something interesting in this list' — no other instructions — could you do it?"
"What if AI doesn't have labels? What if you just dump a pile of data on it and say 'find something interesting'? 🔮 Every AI lesson so far has given AI the answers ahead of time. Today we're taking the answers away — and it turns out AI can still find patterns on its own. That's called unsupervised learning, and it's surprisingly powerful."
Give the class thirty seconds with the number list and let a few volunteers point out anything they notice — odd/even split, a repeated value, numbers clustering around a range. Whatever they find, point out: "Nobody told you what to look for, but you still found a pattern. That's basically what we're about to watch AI do — except with way more data than one list on a whiteboard."
Ask a quick contrast question to link back to the previous lesson: "In 'Learning from Examples,' did the AI ever get told the right answer ahead of time?" (Yes — every photo was labeled cat or dog, every email was labeled spam or not.) "Today, there's no answer key at all. So what do you think AI can even do without one?" Let a few guesses stand without correcting them yet — the app is about to answer the question directly.
Open the lesson and read the hook screen's four icons together — 🔮 Discover, 🗂️ Sort, 🔍 Patterns, 🧩 Groups — before moving into the scenes.
Set expectations for the lesson's earned badge before diving in: finishing this lesson unlocks "Pattern Finder" 🔮, and it's worth telling students directly that this badge is really named after the skill they're about to practice, not just a random reward — spotting patterns and similarity in data with no answer key is the actual, transferable skill underneath everything the app is about to show them.
Work through the app's three scenes as a class, reading each fact and pausing to unpack it.
Read the clustering fact: give AI 1,000 customer profiles with no labels, and it will find groups on its own — "these people are similar!" Explain: "This is called clustering. Real companies use it — the app mentions Amazon groups its shoppers into things like 'budget buyers,' 'tech enthusiasts,' and 'gift shoppers,' completely automatically, just from what people buy." Connect the color-sorting fact: "Same idea, but with colors instead of people — give AI a huge photo and it can group all the similar-looking pixels together to find the main colors, without anyone telling it 'this is red' or 'this is blue.'" Mention the star-clusters fact briefly: astronomers use the exact same clustering idea on millions of stars to spot new patterns in space.
Pause here for a comprehension check: "If Amazon never told the AI what 'budget buyer' or 'tech enthusiast' means, where did those group names come from?" Guide the class toward the right idea — the AI only found that certain purchase patterns cluster together; a person at Amazon looked at what was actually inside each cluster afterward and gave it a human-friendly name. The AI found the groups; a person named them.
Ask: "Have you ever noticed some songs on the radio just feel similar, even if you don't know the genre?" Read the music fact: AI can listen to thousands of songs and group them by style, without ever being told what "style" means. Then walk through the fraud-detection fact carefully, since it's the one most likely to show up on the quiz: "A bank's AI learns what YOUR normal spending looks like — where you usually shop, how much you usually spend. If a transaction shows up that looks very different from your normal pattern, the AI flags it as possibly not really you." Ask: "Why would a bank want AI to notice unusual patterns instead of just checking every single purchase against a list of rules?" (There are too many possible spending patterns for a person to write rules for all of them — the AI learns what "normal" looks like from the data itself.)
Briefly cover the gene-patterns and trends facts as quick extra examples of the same idea in different fields: scientists cluster similar genes in DNA data to discover links to diseases, and stores use AI to spot trends humans would likely miss on their own — like certain products selling better specifically on rainy Tuesdays. Ask: "What do fraud detection, gene research, and rainy-day sales trends all have in common?" (In every case, nobody handed the AI a rule ahead of time — it found the pattern by looking at the data itself.)
This is the mechanics scene — go slower here. Read the "measure distance" fact: AI treats things like points that can be plotted on a map, and points close together count as similar. Read "find centers": AI places some center points, puts each data point with its nearest center, then moves the centers and repeats until things settle down. You can act this out physically: pick 2–3 students to stand as "centers" in the front of the room, then have five more students each walk to whichever "center" student they think they're most similar to (by height, by favorite subject — pick one rule secretly and don't tell them the rule, just have them guess). Finally, read "Choose K": a person tells the AI how many groups to look for — this number is called K — and read "No Right Answer": unlike the labeled-data lessons before this one, there's no single correct grouping; the AI's answer might genuinely surprise you.
Make the "choosing K" idea concrete with a quick example: "If I told a clustering AI to find K=2 groups among our whole school's students, it might split into 'younger grades' and 'older grades.' If I told it to find K=6 groups instead, it might split much more finely — maybe closer to one group per grade level. Same students, same data, but a different choice of K gives a really different-looking answer." Ask: "Is one of those two answers more 'correct' than the other?" (No — they're just different levels of detail, both valid depending on what the groups are actually going to be used for.)
Close by reading the app's own summary together: "Unsupervised learning means AI finds patterns WITHOUT labels. It clusters similar things together!"
Close with: "Today AI got no labels and no answer key, and it still found real patterns — just by measuring how similar things are. That's a completely different skill from the labeled examples we've seen in earlier lessons, and it's how companies group customers, spot fraud, and organize huge piles of data they could never sort by hand." Remind students that they earned the "Pattern Finder" badge for practicing exactly this skill themselves, both in the app and in the card-sorting activity if you ran it — noticing similarity in a pile of information with nobody telling you the rule.
Leave the class with one honest limitation to sit with: clustering finds that a pattern exists, but it never explains why. If a bank's AI groups two customers together, it can't tell you the real reason they're similar — that interpretation is still a human job. This isn't a flaw to hide from students; it's an accurate, useful thing to know about how this kind of AI actually works.
Extension activity (15–20 minutes): Give each small group of students a set of 10–15 index cards, each listing one fictional "customer" with 3 made-up traits (e.g., "buys mostly school supplies, spends ₱50–150, shops on weekends" vs. "buys mostly gadgets, spends ₱2,000+, shops late at night"). Do not tell them how many groups to make. Have each group sort the cards into clusters that make sense to them and give each cluster a name, exactly like a clustering algorithm would (except a human is doing the "measuring distance" step by eye). Compare results across groups at the end — different groups will likely choose a different K and land on slightly different clusters, which is a very concrete way to show there's no single correct grouping.
To stretch the activity further, add a second round where every group must fit the same cards into exactly 3 clusters, no matter what they chose the first time. Some groups will find this harder, since a fixed K might force them to lump together customers they'd rather have kept separate — that friction is worth naming out loud: it's the same real trade-off a data analyst faces choosing K for an actual dataset with real consequences riding on the outcome.