AIQ
AIQ
Everything a teacher needs to deliver this lesson — pick your grade's script below once you've read the background.
This lesson (badge: Eco Defender, 40 XP) sits in World 5, "Is AI Fair?" — the unit where AIQ turns from "how AI works" to "what AI costs and who it affects." By the end of the lesson, across the full 5–18 range, a student should be able to:
You don't need a technical background to teach this lesson well — you need a clear, honest picture of where AI's environmental cost actually comes from, and where its environmental benefit is actually real (not just marketing). Here's the plain version.
Every time an AI system does something — recognizes a face, answers a question, recommends a video — a physical computer somewhere is doing the work. Almost always that computer sits in a data center: a warehouse-sized building packed with server racks. Those servers need electricity to run, and because dense racks of computer chips generate a lot of heat, the building also needs a cooling system, which very often uses water. "The cloud" is not a weightless metaphor — it's someone else's building, someone else's power bill, and someone else's water supply.
It helps to separate two different kinds of AI energy use, because the lesson mixes both together and a teacher who can tell them apart sounds more credible. Training is the one-time (or occasional) process of teaching a model from scratch, using enormous numbers of examples — this is the expensive, concentrated burst of computing the "power 1,000 homes for a year" figure in the lesson is describing. Inference is what happens every single time someone actually uses the finished model — asking it a question, having it caption a photo. Each individual inference is cheap, but it happens billions of times a day across the world, so the running total adds up too, sometimes even surpassing training's share over a popular model's lifetime.
How big are these numbers, really? The lesson's headline figures — a large model's training run using roughly as much electricity as 1,000 homes use in a year, and emitting CO₂ comparable to several cars over their operating lifetimes — are order-of-magnitude estimates published by researchers studying specific models, not a universal AI energy bill. The real number for any given model depends heavily on its size, the hardware it runs on, and critically, what the local electricity grid is made of — a data center running on a coal-heavy grid emits far more CO₂ per unit of computing than one running on a hydro- or solar-heavy grid, even doing the identical job. Treat the headline numbers as "this is a genuinely large amount," not as a number to defend to the decimal point.
Because of that concern, there's now a real, active research area often nicknamed Green AI. It has nothing to do with color or environmental branding — it means building models that reach similar performance while doing less computation. Two real techniques worth knowing the names of, for the older age bands: making a model physically smaller without losing much accuracy (compression/distillation), and building a large model so that it only "wakes up" a small fraction of itself for any single request instead of running its entire capacity every time (this family is called sparse or mixture-of-experts architectures — the Hacker and Architect quizzes both test this idea directly).
The lesson is equally insistent on the other side, and this is the part teachers sometimes under-sell: AI genuinely helps the environment in specific, real, non-hand-wavy ways. AI-assisted control systems optimize electrical grids, reducing waste. Precision-agriculture tools — drones and sensors that tell a farmer exactly where a field needs water or pesticide instead of spraying a whole field uniformly — measurably cut pesticide use. Image-recognition models scanning satellite photos help spot wildfires early, track deforestation, and map ocean plastic for cleanup crews. None of this cancels out the cost side; it just means the honest answer to "is AI bad for the environment?" is "it's a real tradeoff," not a yes or no.
There's a piece of the cost side the lesson doesn't dwell on but that's worth having in your back pocket if a curious student asks "what about the computers themselves?": the hardware that runs AI — specialized chips called GPUs and other accelerators — has its own environmental cost from manufacturing (mining the raw materials, the fabrication process itself) and from disposal once it's replaced by newer hardware, which happens often because the field moves fast. This is usually called the hardware lifecycle, and it sits alongside electricity and water as a third real cost, even though the lesson's numbers focus on energy because that's the piece with the clearest public data.
A device with a browser for each student (or pairs), or one shared screen and a projector for Explorer-mode read-alouds — that's the whole requirement. There is no printing, no lab kit, and nothing to prep in advance. The lesson's own comparisons (electricity, water, pesticide, CO₂) work perfectly well read straight from the screen or spoken aloud from this guide.
The only optional extras are for the wrap-up extension activities described in each age-band script below: scrap paper and crayons for Explorer's drawing activity, or a whiteboard for the older bands' tally/estimate activities. None of it is required to deliver the core 5–15 minute lesson.