📮 Homework · AI Energy and Environment: Where the Electricity Goes

You will run the class prompt audit at home.
1. Ask one family member how many times they think they used an AI tool today.
2. Add your own honest count.
3. Multiply the combined total by 3 watt-hours and divide by 100 Wh to work out how many kettle boils your household 'spent' on AI in a day.
4. Write one sentence describing what surprised you.
5. Bring the sentence back to the next lesson.

You will trim one prompt at home and count the saving.
1. Think of a task you would normally ask an AI to do this week (revision notes, meal ideas, homework help).
2. Write down the prompt you would usually type.
3. Now rewrite it so the answer needs to be one sentence or one bullet list, not an essay.
4. If you have access to an AI tool, run both versions and note the word count of each reply.
5. Work out how many watt-hours you saved (each 100 words is roughly one watt-hour).
6. Bring both prompts and the saving figure to the next lesson.

You will do a plug-hunt at home with a grown-up. Walk round together and count how many plug sockets you can find. Ask the grown-up which ones are used most days and which are hardly ever used. Bring back one number: how many plugs you counted in total. Draw one plug on the back of your sheet if you like. The grown-up helps you count safely and reads the instructions with you.

You will do this with a grown-up. Ask them to switch on a lamp for exactly one minute while you count to sixty. That is how much light one chatbot question buys. Talk about it together: does that feel like a lot or a little? Now think of one question you would ask a chatbot this week, and only that one. Bring back your one question written on paper. The grown-up helps you time the minute and writes with you if you need help.

You will make a list of five things you or a grown-up asked a chatbot, phone assistant or computer today. Next to each, mark it as a sip, a full kettle-boil, or a whole pot. Guess based on how long the answer was. Bring back your list. Talk to a grown-up about which one on the list you would skip next time.

You will ask a grown-up at home one fact question that you would like the answer to. Write down what they say and how sure they sound. Then, if a grown-up agrees, ask the same question to a chatbot they use, or imagine asking it. Write down the chatbot's answer or your best guess of it. Compare: which felt more honest? Which used more electricity? Bring back the two answers side by side on one sheet.

You will make a small table at home. In the first column, list four questions from a piece of homework or revision. In the second column, choose search or chatbot for each. In the third column, calculate the watt-hours: 0.3 per search, 3 per chatbot. Add up the total. Bring back the table and one sentence explaining which choice you would change if you were doing the homework again.

You will keep a short diary for one day at home. Every time a grown-up in your house uses a chatbot on their phone, note the length of the answer (short, medium, long). At the end of the day, ask the grown-up whether they noticed their battery going down faster than usual. Bring back your diary with three or more entries and one sentence about which entry probably cost the most battery.

You will build the full map of one AI prompt's journey at home. Draw every stop from your fingertip to a data centre and back. Label each stop with what it does and mark each one with a lightning bolt if it uses electricity. Add one paragraph explaining which stop surprised you the most and why. If a grown-up already uses a chatbot, ask them where they think it is answering from. Bring back your map to compare with the class.

You will sketch a data centre for homework. Ask a grown-up what they save to the cloud most often. Draw the real building where their photos and files probably live: hangar-sized, cooling towers, generators. Label each part. Under the sketch, write two sentences: one about what would happen to your family's files if the building's electricity failed, and one about which files you would protect first.

You will pick one of the five low-energy AI habits and try it for a week at home. Write the habit at the top of a page. Each day, note whether you actually used it and one sentence about what happened, like 'batched three homework asks into one'. If your family does not use AI, apply the same habit to search: 'asked once, clearly, instead of retrying'. Bring back your seven-day log and one sentence about whether the habit stuck.

You will audit a chatbot session at home. If a grown-up uses a chatbot, sit with them for one task and count the prompts they send. Write down each one. After the session, redraft the whole chain as a single prompt that specifies tone, length and audience. Bring back both versions and one sentence about how many data-centre trips your redraft would have saved. If nobody at home uses a chatbot, invent a session for a writing task and audit that.

You will keep a routing log for the next week. Every time you or a family member is about to open a chatbot, write down the task first. Then ask: is a calculator, dictionary or plain search enough? Choose the simpler tool for at least three tasks that would normally have gone to the chatbot. Bring back your log with at least ten routed tasks and one paragraph about whether the simpler tools got you the same answer.

You will estimate your household's AI usage for one day. Ask each family member roughly how many chatbot queries or AI-assisted searches they ran today. Total the household. Multiply by one watt-hour per query. Convert into kettle-boils, phone charges, or fractions of a washing-machine cycle. Bring back a short paragraph naming your total and one habit your household could try to cut it, like a shared batched prompt for the day.

You will audit AI features on the devices at home. List every device (phones, laptops, tablets, smart speakers) and note which ones can run on-device AI: Apple Intelligence, Gemini Nano on Pixel, Copilot+ on Windows. For each, note two tasks it could handle without the cloud. Bring back your table and one paragraph about which device you would default to for private or quick tasks and why that saves grid electricity.

You will scan your local area for data centre siting risk. Look up whether any hyperscale data centres are in or near your county (Slough, Docklands, Newham, Manchester, Cardiff are the main clusters). Ask a grown-up whether they have heard of local water or electricity constraints. Draft a one-page brief with three stakeholders (resident, utility, planner) and one sentence each on what they would want if a new hyperscale site were proposed near you.

You will draft a short letter to your local MP about a proposed hyperscale AI data centre near your town. Take a clear position (approve, refuse, or approve with conditions). Ground your argument in at least three things you have learned this term: prompt-cost, community stakeholders, water use. Bring back the letter as one side of A4. You do not need to send it. If your family already has opinions on data centres, listen to them first and note what changed your mind.

You will run one paper-only school day. Pick a day this week. For that day, do all schoolwork with notebook, textbook, dictionary and search only, no chatbot, no AI image, no AI summary. Keep a short log of tasks completed and how long each took. Bring back a one-page reflection: what did paper handle just as well? What genuinely needed AI, and how much electricity did that save? Estimate the total watt-hours saved using one watt-hour per query as a rough figure.

You will write a short siting brief for a fictional UK frontier lab planning its next hyperscale build. Cover: two possible sites (one hydro, one baseload), the carbon per training run for each, the latency implications for the UK user base, and one political consideration. End with a clear recommendation and one caveat. Keep it under five sides of A4. Use the current UK grid intensity of roughly 180 gCO2e per kilowatt-hour as your baseline.

You will apply Priya's rule to your own household energy use. Pick one day. Every time someone in the household makes a call that changes electricity use (turning the heating up, running the dishwasher on eco vs quick, opening a hot chatbot session), log the time, decision, rough energy impact and reason in one line. Bring back at least six log entries and one paragraph about whether logging the calls made you think differently. Compare, evaluate and explain the biggest gap you found.

You will draft a formal planning statement for the fictional Peterborough proposal, from the position of a stakeholder you choose (tenant farmer, resident, planner, AI-firm customer, sustainability lead). Two sides of A4. Cover: your position (approve or refuse), your three strongest arguments, the trade-off you accept in taking that position, and one honest concession to the other side. If your position is 'approve with conditions', list the three conditions clearly.

You will produce a one-page cluster brief on Slough, Docklands, Manchester, Cardiff or Newham. Cover: rough gigawatt demand, any known water or grid constraints, latency role for the UK internet, and one honest limitation of the cluster. End with one sentence naming which AI services you know run from your chosen cluster. Use search or a chatbot to find current figures, but cite the source in a footnote.

You will produce a two-sided FAQ card for a public audience (parents, school newsletter, community meeting). Choose four AI-energy questions the public actually asks. For each, write a fifty-word answer grounded in the current numbers: UK grid intensity 180 gCO2e/kWh, data centre global share 1 to 1.5 per cent, IEA projection roughly doubling by 2030. End with one line naming the biggest thing an individual can do that is not already covered elsewhere.

You will draft a one-page policy brief for the UK Department for Energy Security. Recommend a strategy on AI compute: pure efficiency, binding caps, or a defined hybrid. Cover: your recommendation, the historical Jevons pattern, the Irish precedent (Dublin freeze since 2022), one significant risk of your recommendation, and one measurable indicator that would trigger a policy review. Under 500 words. End with one line naming the trade-off you are consciously accepting.

You will write a compute-carbon memo for a hypothetical UK university research group. Include: the four-term formula, the group's assumed workload, the annual UK footprint, the same footprint moved to Iceland, and a recommendation on whether to move. Cover one honest downside of moving (latency to collaborators, jurisdictional data restrictions, political dependence). Under 500 words. Cite grid intensity sources in a footnote. Address it to the group's principal investigator.

You will write a position paper on the UK compute cap question. Under 800 words. Take a clear position: for, against, or for a defined hybrid. Cover: your position, the strongest counterargument you can construct, why you still hold your position after it, the specific workloads your position protects and the ones it lets fail, and the measurable indicator that would make you change your mind. Address it to a Commons select committee. Cite at least two current figures with sources.

You will draft an amendment to the UK's net-zero pathway that honestly includes AI compute growth. Under 600 words. Cover: the current pathway's implicit assumption about industrial demand, the National Grid ESO revised data-centre projection (roughly 500 per cent by 2050), what your amendment specifically requires (mandatory per-model reporting, grid-share ceiling, both), and one honest political cost of the amendment. Cite three sources including one non-AI source, and one international comparator.

You will draft a briefing note on UK AI compute regulation for a shadow front-bench MP. Under 700 words. Cover: your recommended toolkit (two priority tools, one you would not use, one you would defer), the historical Jevons pattern, the Irish precedent, one arbitrage risk with the EU AI Act, and one measurable indicator triggering a policy review. Cite three sources including at least one from outside the AI industry. End with one honest concession to the opposing position.

You will draft a one-page vendor briefing memo for your care group's procurement lead. Cover: three current AI tools in use, whether each vendor publishes per-transaction kilowatt-hours and hosting location, one gap in the group's sustainability reporting, and one recommendation, either drop the tool, request the data, or renegotiate the contract. Ground everything in the per-resident-day KPI. Under 500 words.

You will produce a monthly energy estimate for AI admin across your home. Estimate the daily kilowatt-hours from AI summaries, drafts and rota checks based on typical shift patterns. Multiply out to one month. Convert into recognisable units: households' worth, EV miles, kettle boils. Draft one process change that would cut the total by 50 per cent (typically a batched call). Address the estimate to your registered manager, under 500 words.

You will produce a one-page carbon-honest design log for a live project. List every design decision on the project, mark which were AI-assisted, note the tonnes CO2e each decision moved (material savings) versus the kilos CO2e it cost in AI compute. Total both columns. Under 500 words. End with one line naming the design decision that gave the strongest carbon return per unit of AI compute.

You will draft a hybrid inspection workflow for your next live project. Cover: which steps AI triage handles, which steps the site manager keeps hands-on, the estimated time saving per inspection, the estimated cloud energy cost per inspection, and one line about how the workflow preserves the site manager's ability to spot the snag that fits no obvious category. Under 500 words. Address it to the project's construction manager.

You will draft a one-page pilot brief for AI quality-control on one production line at your plant. Cover: the line chosen, the baseline KPIs (kilowatt-hours per unit, scrap rate, units per shift), the pilot duration, the pass/fail thresholds, the estimated AI inference cost per unit, and one honest risk (partial success masking throughput loss). Under 500 words. Address it to the plant's continuous-improvement lead.

You will draft an inference architecture memo for your team lead. Cover: one live AI feature, its current inference route (cloud or on-device), the four-factor score, the recommended change if any, and the estimated annual kilowatt-hour and cost saving. Ground everything in KSB B3. Under 600 words. Reference at least one internal SLA or KPI. Address it to your line manager and copy the sustainability lead.

You will produce a KSB B3 evidence entry for your portfolio. Pick a pipeline you own. Cover: the current monthly kilowatt-hours (with your working), the single change you would make to cut it (delta re-embed, cache, on-device fallback), the projected new monthly figure, the projected annual cost saving, and the measurement approach for production. Under 700 words. Attach a screenshot of the source code change if you can. Cite KSB B3 explicitly.

You will write the memo the scenario calls for. If pushing back: name the KSB B3 responsibility, quantify the energy and cost delta, propose the fine-tune alternative, and state the decision review point. If approving: quantify the customer benefit, own the extra kilowatt-hours in reporting, name the mitigation (offset, other efficiency saving), and state how you would defend the decision against B3. Under 600 words. Address it to your PM and copy the sustainability lead.