AIQ
AIQ
Part of the AI Is Everywhere! lesson guide. Teaching a different grade? 🌈 Explorer (5–7) · 🔧 Builder (8–10) · 💻 Hacker (11–14)
Architect mode drops the mascot entirely and talks to students as near-adults heading toward careers — the whole age band assumes students are within a few years of choosing a college major or a first job, and the framing throughout treats AI literacy as career-relevant rather than merely interesting. Open with the app's own framing, delivered plainly:
"AI has become so embedded in daily technology that most interactions with it are invisible. Let's audit your daily tech stack — you'll be surprised how many AI inference calls happen before you finish breakfast."
Frame the warm-up as a literal audit: have students individually list, in two minutes, every digital interaction from waking up to arriving at school (alarm, phone unlock, checking notifications, music, transit app, etc.). Then ask them to mark which ones they think involve an actual trained model making a prediction, versus which are just data display or fixed automation. Don't correct yet — tell them this lesson gives them the precise vocabulary to check their own answers.
At this age, treat the class less like students receiving new information and more like junior analysts who already have working intuitions worth testing rigorously. Most will already know AI is "everywhere" in the loose sense; the value this lesson adds for them specifically is a shared, precise vocabulary — narrow AI, inference, training data — that turns a vague sense of "AI does a lot of stuff now" into something they can reason about and argue with precision.
Write the app's own age-band summary on the board before working the scenes, since it's the most technically precise framing in the whole lesson and worth having students read closely: "Modern AI systems leverage machine learning models trained on massive datasets to perform inference — making predictions or classifications in real-time. The distinction between AI and conventional software is adaptability: traditional programs execute predetermined logic, while ML models generalize from training data to handle novel inputs. Most consumer AI today is narrow AI — optimized for specific tasks like NLP, computer vision, or recommendation systems."
Then work the three scenes as case studies in that framework, naming the underlying task category explicitly:
Morning: Smart Alarm — a speech-recognition / NLP task (converting audio to intent). Face Unlock — a computer-vision classification task (does this camera frame match a stored face embedding). Shower and Light Switch are conventional, deterministic mechanisms with no model in the loop.
School: Spell/grammar check — an NLP task, now typically a language model predicting likely words/corrections from context rather than a static dictionary lookup. Adaptive game difficulty — a recommendation/optimization task tracking player performance to adjust parameters, structurally similar to a recommendation system even though its output is a difficulty setting rather than a piece of content. Paper book and scissors involve no computation at all.
Evening: Song and video recommendations — classic recommendation-system tasks (collaborative filtering / learned-preference models) built on large-scale behavioral datasets. GPS traffic prediction — a time-series prediction task layered on top of (non-AI) satellite positioning; the positioning itself is precise physics, the congestion forecasting is the learned part. A bicycle involves no model whatsoever. The GPS example is worth dwelling on longer than the others: it's the clearest case in the whole lesson of a system that is genuinely part-AI and part-not, layered together, rather than a single clean "AI or not AI" object — and most real consumer products are built this way, as a stack of AI and non-AI components rather than one or the other.
Have students place each example into one of the three named categories from the summary — NLP, computer vision, or recommendation/optimization — as a quick worksheet or verbal round-robin. This is the piece that separates Architect mode from Hacker mode: not just "AI or not," but "which narrow-AI category, specifically."
Run the in-app practice round (ten more items) as a fast written check rather than a class discussion at this point — Siri/Alexa and Auto-Translate are both NLP tasks, Photo Filters is computer vision, Smart Thermostat and Netflix Picks are recommendation/optimization, and Spam Filter is a classification task, a category adjacent to but distinct from straightforward recommendation (it's sorting into two labeled categories from patterns in prior labeled examples, rather than ranking a continuous set of options by predicted preference). Broom, Paper Clip, Metal Key, and Ruler remain non-AI. Having students commit their category answers to paper individually, then reveal, surfaces genuine gaps faster than open discussion does at this depth of material.
These questions don't have a single correct answer, and that's the point at this age — treat them as a seminar-style discussion rather than a Q&A with a key. Fifteen minutes is a reasonable target; the disclosure question in particular tends to run long once students start comparing consumer apps to higher-stakes domains (lending, hiring, healthcare) where similar systems already operate with far less visibility than a music app's "why am I seeing this" button.
Close with: "You now have real vocabulary — narrow AI, inference, training data, adaptability — for a conversation that most adults have only informally. That vocabulary is directly useful whether you go into a technical field or not, because AI-driven decisions increasingly show up in hiring, lending, college admissions, and healthcare, not just apps."
Extension activity — Personal AI Footprint Report: Have students write a short (half-page to one-page) audit of their own daily AI exposure, structured around the three categories used in the Main Activity (NLP, computer vision, recommendation/optimization). For each category, they should name at least one real system they interact with, what data it likely trains on, and one plausible way that system's optimization target (e.g., "maximize engagement," "maximize accuracy of face match") could work against their own interests. This stretches the 10–15 minute core lesson into a full period plus a short homework deliverable, and previews the fairness/ethics threads the curriculum develops later (privacy, bias, "when AI isn't fair").
For students heading toward a technical or policy career, add an optional stretch question to the report: "Pick one system from your audit and describe, in a sentence or two, what a reasonable person might disagree about regarding how it should be built or regulated." This isn't asking for a right answer — it's practice at identifying that a technical design choice (what data to train on, what metric to optimize) is very often also a values choice, which is a genuinely useful habit of mind heading into any field that touches AI, technical or not.