Teaching the Seven Questions
The split itself is easy to teach. What is hard — and what this page is mostly about — is delivering a correction about scale to a room that cares, without being heard to say the problem isn’t real.
Why this is hard to teach.
When you tell a room that water is a local issue rather than a global one, a portion of the room will hear “this doesn’t matter.”Not because they weren’t listening. Because in most public conversation, insisting on precision is a move people have learned to associate with defending an industry.
The distinction between this is smaller than you think and this is different from what you thinkis easy to state and hard to hear — especially from someone with a lectern, and especially when the listener’s concern is moral rather than technical. This is not a failure of your audience. It is a structural feature of the topic, and every educator who teaches this material will hit it.
So plan for it rather than being surprised by it. The rest of this page is mostly craft: a named sequence you can practise, and two activities that let the room discover the split rather than be told it.
The three-beat rule.
Never let a scale claim be the last thing you say. Teach this sequence explicitly, and have facilitators rehearse it out loud — it is much harder to do under pressure than it looks on a page.
The correction
“Water isn't a global resource. It's a watershed, in a season, on a peak day.”
Alone, this reads as dismissal. Never stop here.
What the correction buys
“Which means there's a specific water system, a specific drought plan, and a specific disclosure requirement that either exists or doesn't. That's a thing you can call.”
Beats one and two together read as technocratic — correct, and cold.
The stake
“And in most of the country nobody is required to produce that number at all. That's the actual problem, and it's a bigger one than the version you'd heard.”
All three together read as someone who cares enough to be exact.
The companion move: say your concern out loud before you correct.
“I think this matters a great deal, and I think the most common version of the argument gets it wrong in a way that makes it harder to fix.” Facilitators often skip this because it feels like grandstanding. It isn’t — it is the context that makes the correction legible. Without it, the room has to infer your motives from your content, and they will infer wrong.
Target misconceptions.
“Being precise about this means minimizing it.”
Precision is what converts a feeling into an action. The imprecise version of the claim has no addressee — there is nobody to call about “the planet.” The precise version names a water system, a commission, and a permit.
“These impacts are basically one problem.”
Two of them trade off against each other. Closed-loop cooling cuts water use and raises electricity demand, and therefore emissions. Things that move in opposite directions were never one thing.
“The environmental case is settled; the only question is whether you care.”
Two of the seven questions are unresolved, one has no CPAI evidence base at all, and one of the most widely repeated figures in the literature was retired by its own authors. Caring is not the variable in dispute.
“My personal AI use is the lever.”
The per-response footprint is small. Siting, cooling design, disclosure, and rate design are the levers — and all four are decided in proceedings that accept public comment and are attended almost entirely by parties with a financial interest.
Two classroom-ready activities.
Split the Sentence
25 min · no techWrite one statement where everyone can see it: “AI is bad for the environment.” In pairs, participants list every separate question hiding inside it. Reconvene and build the class list on the board.
Then, for each question the room found, ask one follow-up: “Who would you have to call to get the answer?”
That second question is what does the teaching. “The planet” is not an addressee. “My municipal water utility” is. Debrief on which questions had an addressee and which didn’t — and note that the ones without an addressee are usually the ones people argue about longest.
Materials: board or flip chart. Optional worksheet: seven blank rows with columns for the question · the scale · who has the answer.
Steelman Both Sides
30 min · group workSplit the room in three. Group A builds the strongest overstated case. Group B builds the strongest dismissive case. Group C builds the case the evidence supports. Present in that order.
Groups A and B both discover that their case requires either a retired statistic or a scale switch. Group C discovers that theirs is harder to deliver and less satisfying — which is the honest finding, and worth naming out loud rather than smoothing over.
Facilitator note: assign groups randomly and say so, so nobody reads their assignment as a statement about their views. Materials: the guide, printed or on screen.
Discussion prompts.
Ordered from easy to charged.
- 1.Which of the seven questions had you actually been thinking about?
- 2.What would have to be true for you to change your mind about one of them?
- 3.Every figure that drifted as it circulated drifted upward and toward AI causation. What does that tell you — about AI, or about the conversation?
- 4.Is it possible to be too careful about a problem that's real?
- 5.If you had to pick one of the seven to work on, which has the most leverage where you live?
- 6.When someone corrects a fact you cited in an argument you believe in, what do you feel first — and what does that tell you about how to deliver a correction to someone else?
Seeing whether it landed.
The rewrite test. Give students a fresh vague sentence and ask them to produce two or three specific questions plus who would answer each. If the addressee column is filled in, the lesson landed.
The scale-tag test. Hand out five statistics from any source and ask students to tag each with the scale at which it is true. Wrong tags are more interesting than right ones — ask what made it ambiguous.
The disagreement test. Have two students argue opposite sides for three minutes, then stop them and ask the room which question each one was answering. In most cases they were answering different ones, and the room will see it before either debater does.
When someone asks “is this proven?”
In teacher voice
“Some of it, yes — the federal electricity figures and the health modeling are solid, with the caveat that the health numbers are scenarios and you have to say which one you mean. Some of it is contested — whether data centers are raising retail rates is genuinely being argued by serious people on both sides. Some of it we don’t know: what actually clears at the margin when new load shows up. And one of the seven questions, materials and mining, we haven’t researched at all. Telling you which is which is the whole point of the lesson.”
The other four educator guides.
Where this leads
Teaching this is a different skill than knowing it.
Teaching AI Well is the facilitator library behind CPAI's Certified Applied AI Trainer Program — ten lessons on how to teach AI honestly, including the material on this page.