What should students learn around advanced AI?
Follow language, mathematics, science and judgement through a complete fictional decision, then use the evidence to evaluate a realistic curriculum.

Students should learn how to make sense of a claim, work out its quantities, investigate what the evidence can establish, and decide what deserves to be done. Language, mathematics, science and judgement contribute different parts of that work. An advanced AI system may help with every part, but a curriculum still has to decide which knowledge and practices students are developing, where assistance belongs, and what students should be able to explain for themselves.
This guide turns that curriculum question into a complete interdisciplinary task. A fictional garden proposal supplies a short persuasive text, measured quantities, an imperfect comparison, alternative designs and firm resource limits. We work through the language, mathematics, science and judgement separately, bring them together, and repair a curriculum that spends too much time producing an impressive presentation. A second packet about a library workshop lets readers try the reasoning again with different evidence and constraints.
Super Intelligence is the wider series title. Artificial superintelligence, or ASI, refers here to a hypothetical broadly superhuman capability; the scenarios do not establish that present AI systems are ASI. The classroom materials, institutions, prices, measurements and responses below are invented teaching examples. They are not reports about real students, forecasts of attainment or an official Singapore syllabus. No real student account, photograph, learning record or personal information is needed to use the paper exercises.
Choose your reading route
Choose the learning priorities · Work through the Canopy packet · Repair the curriculum · Try the independent library task
What each discipline contributes
- 1. Choose capabilities before choosing fashionable topics
- 2. Language teaches students to interpret the exact claim
- 3. Mathematics makes the relationships visible
- 4. Science connects explanations to evidence
- 5. Judgement decides what the numbers cannot decide alone
- 6. Use frameworks and learning research for bounded purposes
The complete Canopy packet and worked reasoning
- 7. The complete Canopy packet: brief, claim and records
- 8. The complete Canopy packet: options and limits
- 9. Work the language check before calculating
- 10. Reproduce the water and growth calculations
- 11. Reproduce costs, workload and conditional payback
- 12. Identify the scientific question the first comparison cannot answer
- 13. Read a stronger comparison without exaggerating it
- 14. Make a recommendation with an explicit decision boundary
- 15. Combine the disciplines in one worked response
- 16. Repair a plausible wrong student answer
A curriculum that fits its limits
- 17. Audit a curriculum that looks impressive but fails the brief
- 18. Build a feasible six-lesson alternative
- 19. Count preparation, materials and marking honestly
- 20. Look for each discipline in the learner’s response
Independent transfer and the next curriculum decision
- 21. Independent transfer: the complete Juniper Library packet
- 22. Transfer questions to attempt before reading the answers
- 23. Transfer answers: quantities and interpretation
- 24. Transfer answers: science, options and curriculum repair
- 25. Adapt depth without pretending one project is the whole curriculum
- 26. Judge future curriculum proposals by what they preserve
1. Choose capabilities before choosing fashionable topics
A curriculum proposal can sound modern while leaving its educational purpose unclear. It may promise AI literacy, creativity, future readiness and problem solving, yet allocate almost every lesson to generating slides. Before considering the software, ask what the learner will know or be able to do afterwards. “Produce a persuasive video” describes a deliverable. “Distinguish a measured reduction from a claimed cause, calculate the relevant rate and justify a bounded recommendation” describes several capabilities that can be examined in the work.
The distinction matters when outputs become inexpensive. If a system supplies a polished explanation, the class still needs a reason for studying that explanation. Perhaps students are learning how a percentage uses a denominator. Perhaps they are comparing interpretations of a passage. Perhaps the explanation is a worked example before a different problem. Each purpose suggests a different lesson. Showing the same generated answer to everyone does not, by itself, specify which of those purposes has been served.
For this guide, the curriculum has four connected commitments. Language develops precise interpretation and accountable communication. Mathematics develops explicit relationships, quantities and constraints. Science develops explanations that remain answerable to observation and investigation. Judgement develops defensible choices among goals, evidence, duties and limits. These are proposed organising commitments for the examples, not a claim that four labels exhaust education. History, the arts, physical education, languages, practical craft and social learning have their own important content and purposes.
The useful question is therefore not which subject defeats AI. It is which disciplinary contribution disappears when a subject is reduced to checking a machine’s output. If mathematics becomes only reading an answer, students may never meet the model behind it. If language becomes only rewriting a prompt, they may miss perspective, ambiguity and argument. If science becomes only accepting an explanation, they may miss what would count against it. A curriculum should make those contributions visible rather than treating every activity as generic digital competence.
Back to contents · Next: 2. Language teaches students to interpret the exact claim
2. Language teaches students to interpret the exact claim
Language is the first working instrument in the garden task. Students must identify who is speaking, what is being claimed, which words narrow or enlarge that claim, and which evidence is actually offered. A sentence that says a device “could reduce recorded watering in this trial” commits its writer to something different from a sentence saying it “will save water everywhere.” The difference is not decorative caution. It changes the promise a reader is being asked to accept.
Vocabulary also carries conceptual boundaries. Recorded watering, total water received, plant growth and plant health are not interchangeable expressions. A reader who treats them as synonyms can make a scientifically weak inference before doing any arithmetic. Language teaching can therefore ask students to define terms in context, paraphrase without strengthening a claim, and preserve a distinction when shortening a passage. A concise summary that removes the decisive qualification is a worse summary, even if its grammar is elegant.
Students also need practice recognising the relationship between sentences. A writer may place a large saving beside a photograph of thriving plants without explicitly stating causation. The reader may nevertheless infer that the intervention produced both outcomes. The curriculum can examine that implied argument: which step is stated, which is suggested, and which still needs evidence? This work belongs to reading and composition as well as to scientific reasoning. Persuasion can operate through order, selection and omission, not only through false sentences.
Writing then makes the learner’s own reasoning inspectable. A useful paragraph identifies the recommendation, cites the relevant record, acknowledges the important uncertainty and states the next decision. It does not need impressive vocabulary. Oral discussion contributes another practice: hearing an objection accurately enough to answer the objection actually made. In this article’s tasks, students should be able to explain a peer’s strongest concern before responding. That is an original teaching choice for the proposed unit, not a measured claim about the effect of a particular discussion routine.
Back to contents · Next: 3. Mathematics makes the relationships visible
3. Mathematics makes the relationships visible
Mathematics contributes more than correct arithmetic. It asks what is being counted, which quantities are comparable, what the units mean and how a conclusion depends on assumptions. A percentage can be calculated perfectly while answering the wrong question. A total can fall because fewer people attended. An average can conceal differences between groups. A forecast can multiply a small observation into a large promise without explaining why the observed relationship should continue.
In the garden task, students calculate totals, means, differences, percentage changes and a conditional projection. They also distinguish a physical quantity from its financial value. Seventy litres of water and seventy dollars are not the same saving. A cheap resource can have an ecological or educational importance that is not captured by a small bill reduction. Mathematics clarifies that distinction; it does not settle how much the community should value the resource. Clear units help prevent a numerical answer from quietly becoming a value judgement.
Representation is another curriculum choice. A table is useful for checking individual measurements. A bar chart can make totals easier to compare. A written equation shows the operation and denominator. A sentence explains what the result refers to. Students should encounter the translation between these forms, because a chart alone may hide the calculation and an equation alone may hide the context. In a short unit, one well-chosen conversion between representations is more useful than requiring several decorative charts of the same numbers.
A suitable mathematical explanation should also say where it stops. “If each of twenty planters saved the observed average of 3.5 litres, the projected reduction would be 70 litres per week” is an inspectable conditional statement. “The new system saves 70 litres” silently removes the assumption. The arithmetic is shared, but the claims differ. Curriculum selection should preserve enough number sense, proportional reasoning, algebra and data interpretation for students to recognise that difference, while adapting the specific operations to their prior learning.
Back to contents · Next: 4. Science connects explanations to evidence
4. Science connects explanations to evidence
Science contributes questions about observation, measurement, comparison and causal explanation. The garden packet presents lower recorded watering and greater average stem growth after a change. Those observations invite a hypothesis, but the packet also changes the location of the planters and includes rainfall that was not counted. Students have to decide whether the evidence isolates the intervention. Knowing the names of scientific variables is less useful here than being able to point to the particular variables that weaken the proposed explanation.
The National Research Council’s Framework for K–12 Science Education describes scientific and engineering practices including asking questions, planning investigations, interpreting data and constructing evidence-based explanations. Its account also distinguishes empirical questions from choices that depend on other forms of reasoning. This is a useful conceptual reference for the science component below. It is a United States framework, not certification that this invented unit meets a Singapore syllabus or a demonstration that an AI-supported version will improve learning. Science and engineering practices
For this unit, a student should be able to propose a comparison in which the disputed explanation has a fairer test. That involves defining what will be measured, keeping relevant conditions comparable, deciding how the intervention will be assigned, and recording departures from the plan. A sensible design also identifies what its evidence would still leave unresolved. A small, short study may answer a narrow question while saying little about seasonal performance, maintenance or long-term plant condition.
Scientific content remains essential. The garden exercise uses a deliberately limited set of stipulated facts so it can focus on reasoning from evidence. A fuller science course would need substantive knowledge of plants, water, environments, materials and measurement. General advice to “think critically” cannot supply that knowledge on demand. The curriculum decision is to connect selected content with disciplined investigation, rather than replace content with a universal checklist or turn one small project into evidence that all scientific understanding has been covered.
Back to contents · Next: 5. Judgement decides what the numbers cannot decide alone
5. Judgement decides what the numbers cannot decide alone
Judgement begins when more than one defensible goal matters. The garden committee may care about reducing resource use, keeping plants in acceptable condition, limiting maintenance work and creating a useful learning opportunity. A design can perform well on one goal and badly on another. A calculation can show the difference, but someone must still explain which trade-off is acceptable, who is entitled to decide and whose work or access is affected.
This does not make judgement arbitrary. A recommendation can be constrained by factual accuracy, the stated budget, the available staff time and the committee’s actual authority. It can also be judged by whether it notices excluded people, applies a principle consistently and offers reasons that others can challenge. A student is not entitled to invent a maintenance volunteer in order to make a favourite proposal fit. Nor should a majority preference be reported as scientific proof that the chosen design is best in every respect.
The exercise distinguishes three kinds of statement. “The kit costs 96 dollars” is a supplied fact. “The projected bill reduction is 14 cents per week under the stated assumptions” is a calculation. “A modest educational trial is worth funding even without financial payback” is a value-based proposal. The third can be reasonable, but it needs an authorised budget and an explicit educational purpose. Giving it a numerical score does not transform it into a discovered law of nature.
A curriculum that includes judgement should therefore let students defend different conclusions when the evidence and constraints allow them. Two students might prefer a low-cost manual trial and a paper-only investigation respectively. Both can be defensible if they account for the limits. A proposal that exceeds a firm resource cap cannot be rescued merely by confident advocacy. The task is to develop reasoning that is principled and responsive, while keeping adult responsibility for real safety, spending and school policy decisions intact.
Back to contents · Next: 6. Use frameworks and learning research for bounded purposes
6. Use frameworks and learning research for bounded purposes
UNESCO’s 2024 AI competency framework for students organises twelve competencies across a human-centred mindset, AI ethics, techniques and applications, and system design, with progression through understanding, applying and creating. It provides one way to examine whether AI education has become too narrow. A unit devoted entirely to obtaining fluent answers would leave important parts of that framework unexplored. The framework is a curriculum resource; it is not empirical proof that a particular lesson sequence, tool or hypothetical ASI system produces better learning. UNESCO student framework
The U.S. Institute of Education Sciences’ 2007 practice guide recommends spacing learning and alternating worked examples with problems students solve, among other practices. It assigns different evidence ratings to different recommendations. In this article, that guidance supports providing a worked packet followed by a new task and a later revisit. It does not supply a universal six-lesson prescription, establish the right amount of AI assistance or validate the fictional budget below. Those are explicit design choices that an educator would need to adapt. IES organising instruction guide
One primary study helps explain why immediate fluency deserves cautious interpretation. Roediger and Karpicke’s 2006 experiments compared repeated study with recall testing of prose passages and examined retention at different delays. Repeated study did better on the five-minute test, whereas prior testing did better on delayed tests. The result concerns those experimental tasks and conditions. It is not evidence that every independent quiz improves every subject, nor a study of today’s generative AI. It supports asking when and how learning is being observed. Original study record and abstract
The remaining proposals in this guide are original curriculum examples, not reported findings from those sources. Their usefulness rests on whether the objectives, materials, work and constraints fit together. A teacher can disagree with the allocation and propose a better one. The appropriate response is to show which learning requirement the alternative preserves or improves, what it costs in time, and what evidence would be examined afterwards. Referencing a recognised framework should make that reasoning more accountable rather than end the discussion.
Back to contents · Next: 7. The complete Canopy packet: brief, claim and records
7. The complete Canopy packet: brief, claim and records
Canopy is an invented learning workshop considering a four-planter garden demonstration. The decision in this packet is a paper recommendation only. The class must advise whether the evidence supports a twenty-planter rollout, whether a smaller investigation is feasible, and what a six-lesson interdisciplinary unit should teach. All prices are fictional Singapore-dollar teaching inputs, inclusive of any charges for this exercise. They are not local tariffs, supplier quotations or requests to purchase equipment.
The publicity draft reads: “Our new watering kit cuts water use by 35 per cent and makes plants healthier. It will save 70 litres every week across twenty planters, pays for itself through lower bills, and needs almost no extra work. Every class should switch immediately.” The draft was prepared for discussion; it has not been published. The named “new watering kit” refers to the Drip option specified later. Treat each clause as a separate claim requiring its own support.
Record A covers five school days before the kit was introduced. Four planters received recorded watering volumes of 10, 12, 8 and 10 litres respectively. Their measured stem-growth increments were 20, 25, 15 and 20 millimetres. Record B covers the following five school days after the kit was introduced to the same four planters. Recorded watering volumes were 6, 8, 5 and 7 litres; stem-growth increments were 24, 28, 19 and 25 millimetres. Each growth figure describes change over that record’s five-day period, not total plant height.
The observation note says that the planters moved to a shadier position at the start of Record B. Rain reached them during B, but rainwater volume was not measured or included in the watering figures. A and B therefore count deliberately added water only. The note supplies no measurement of leaf condition, root condition, survival, disease, harvest or long-term health. It does not identify which change caused any observed difference. These omissions are part of the packet, not facts the student is expected to fill in from imagination.
Back to contents · Next: 8. The complete Canopy packet: options and limits
8. The complete Canopy packet: options and limits
The garden committee has a separate one-time equipment ceiling of 140 dollars, an incremental maintenance allowance of 5 dollars per week, and at most 40 staff minutes per week for the whole twenty-planter arrangement. These are hard limits for this decision. The stated existing routine takes 25 staff minutes weekly. No unpaid student labour, family contribution, extra device or hidden budget may be assumed. A recommendation may request reconsideration of a limit, but cannot claim that the changed limit has already been approved.
Three options are supplied. Status quo keeps the existing arrangement, with no setup purchase, no incremental maintenance expense and the stated 25 staff minutes per week. Drip costs 96 dollars to establish, adds 4 dollars in maintenance each week and requires 55 staff minutes per week in total. Manual costs 28 dollars to establish, adds 1 dollar weekly and requires 35 staff minutes per week in total. The packet provides no measured water-saving result for Manual. Drip is the kit used during Record B, but the observational limitations remain.
For arithmetic practice only, the fictional water price is 0.002 dollars per litre. Financial comparisons count the supplied setup and incremental maintenance costs and the projected water-bill reduction. They do not put a monetary price on staff time or ecological benefit. Do not silently treat omitted values as worthless. Their omission means that the financial calculation has a restricted scope, while the separate staff-time limit still applies.
The learning unit has its own constraints: thirty learners, six groups of five, six shared devices already available, six lessons of fifty minutes, a materials ceiling of 180 dollars, no required personal AI accounts or home device use, up to ninety minutes of teacher preparation and up to 120 minutes of marking outside class. The garden budget and teaching budget are separate envelopes. The class may use printed fictional records throughout, and all AI examples may be supplied as fixed text. Neither live AI access nor a real growing trial is necessary to complete the unit.
Back to contents · Next: 9. Work the language check before calculating
9. Work the language check before calculating
Start by marking the publicity draft’s commitments. “Cuts water use” attributes a reduction to the kit and describes water use broadly. Record B measures only deliberately added water, excludes rain and coincides with a move. The language therefore exceeds the record in two ways: it implies a cause that has not been isolated and a total-water measure that was not taken. A good reader should identify both, rather than correcting only the word “cuts” while leaving the measurement claim too broad.
“Makes plants healthier” is another unsupported enlargement. The packet contains stem-growth increments, not a defined health assessment. Students can say that the average recorded growth increment was larger in B after calculating it. They cannot treat a single growth measure as a complete health verdict. Nor can they attribute the difference to the kit under the supplied design. This is a useful vocabulary exercise because the error is in the relationship between the label and the observation, not in an obviously impossible number.
“It will save 70 litres every week” changes a conditional projection into a guarantee. “Pays for itself” makes a financial claim that requires both benefits and costs. “Almost no extra work” is vague, but the separate option record gives enough information to challenge the impression. “Every class should switch immediately” generalises from a tiny, limited comparison and adds an action recommendation without checking the committee’s limits. Students should keep those failures separate so a repair addresses every unsupported step.
A corrected short version is: “During the second five-day period, the four planters received 14 fewer litres of recorded watering. Rain and a change of position prevent us from attributing the difference to the kit. A twenty-planter projection would be 70 litres per week only if the same average difference applied. The supplied maintenance and staffing figures do not support an immediate rollout.” This preserves the useful observation, identifies the limit and avoids pretending that uncertainty means nothing at all was learned.
Back to contents · Next: 10. Reproduce the water and growth calculations
10. Reproduce the water and growth calculations
Record A’s total watering is 10 + 12 + 8 + 10 = 40 litres. Record B’s total is 6 + 8 + 5 + 7 = 26 litres. The difference is 40 − 26 = 14 litres across the four planters over the two equal-length five-day recording windows. Relative to A, the reduction in recorded watering is 14 ÷ 40 × 100 = 35 per cent. The denominator is the original recorded amount, because the question asks how far B fell from A.
Using 26 as the denominator instead answers a different comparison: 14 ÷ 26 × 100 is approximately 53.85 per cent, describing how much larger A was than B relative to B. The two percentages are not competing rounding choices. They express different reference points. A student who obtains 53.85 per cent should repair the question-denominator match, not merely replace the final answer with 35. This is why showing the operation matters when a tool can calculate either expression instantly.
The mean recorded watering is 40 ÷ 4 = 10 litres per planter in A and 26 ÷ 4 = 6.5 litres in B. The mean difference is 3.5 litres. Multiplying by twenty gives a conditional 70-litre weekly reduction. The five-day period is being treated as the workshop’s school week for this projection. Nothing in the packet establishes that weekends, holidays, other species or other locations follow the same relationship. Those conditions belong in the interpretation, rather than being concealed by a clean integer.
Growth increments total 80 millimetres in A and 96 millimetres in B, giving means of 20 and 24 millimetres per planter. The mean difference is 4 millimetres, or 20 per cent relative to A’s mean. Do not add the watering and growth percentages into a “55 per cent improvement” score. They describe different quantities, with different units and meanings. If a later decision uses a combined score, it must explicitly define and justify the weighting rather than disguise an arbitrary sum as a scientific result.
Back to contents · Next: 11. Reproduce costs, workload and conditional payback
11. Reproduce costs, workload and conditional payback
At the stipulated price, a 70-litre weekly reduction would lower the water bill by 70 × 0.002 = 0.14 dollars per week, or fourteen cents. Even granting the unproven projection for calculation, Drip adds 4 dollars in weekly maintenance. The bill reduction minus this expense is 0.14 − 4 = −3.86 dollars per week. That is a net recurring financial cost within the limited accounting model. The 96-dollar setup expense is additional.
Dividing 96 by 0.14 gives about 685.71 weeks, but that figure ignores maintenance and therefore does not establish payback under the packet’s terms. With a negative recurring net saving, this simple model has no positive payback period for the setup cost. A more sophisticated system could still be chosen for other reasons, but the publicity claim about paying for itself through lower bills is unsupported here. Students should be rewarded for correcting the model before performing a division, not for producing an impressive-looking duration.
The staffing comparison is equally important. Drip requires 55 minutes per week, which exceeds the 40-minute ceiling by 15 minutes and exceeds the existing routine by 30 minutes. Its 96-dollar setup and 4-dollar maintenance do fit their respective monetary ceilings. Passing two limits does not cancel failure on the third. Manual’s 28-dollar setup, 1-dollar weekly maintenance and 35-minute total workload fit all three ceilings, leaving 112 dollars, 4 dollars per week and 5 minutes per week of headroom respectively.
Manual’s feasibility is not evidence of water savings. The packet gives it no measured performance result. A careful recommendation can identify Manual as a financially and operationally feasible candidate for a bounded investigation, while refusing to assign it Drip’s observed difference. Status quo also fits the limits, leaving the whole equipment allowance unspent. The mathematical work narrows the feasible options; the judgement work must then consider whether the educational or environmental purpose justifies an investigation and what evidence would be needed before scaling.
Back to contents · Next: 12. Identify the scientific question the first comparison cannot answer
12. Identify the scientific question the first comparison cannot answer
A narrow descriptive question can be answered: how much deliberately added water was recorded in each period? A causal question cannot be settled from these records alone: how much of the difference was caused by the kit? The move and unmeasured rain provide alternative explanations or contributing factors. The same-planter design preserves some continuity, but it does not hold the environment or time constant. Labelling A a “control” after the event would not remove those differences.
A second causal claim concerns growth. The later period has a larger mean recorded increment, but the packet does not show whether that difference came from the kit, the position, rain, plant development over time or some combination. Students do not need to prove which alternative is correct to show that the publicity conclusion is too strong. They need to identify plausible unresolved differences supplied by the packet and explain why those differences matter to the attribution being attempted.
Measurement deserves separate attention. Because total received water is unmeasured, “35 per cent less recorded watering” is safer than “35 per cent less water used by the plants.” Stem length is one observable measure, not a complete account of plant condition. The packet also lacks repeated independent runs, a stated instrument precision and an account of recording errors. These absences should limit the conclusion. They should not be converted into invented numerical uncertainty intervals or claims that a hidden error definitely occurred.
An appropriate question for a redesigned investigation is: “Under comparable conditions in this demonstration, how does the kit affect the amount of deliberately added water while maintaining the specified observation measures over one week?” Even this needs operational definitions before data collection. Which conditions are comparable? What constitutes an acceptable observation? Who records a deviation? The curriculum contribution is the student’s ability to sharpen the question and connect it to a design, while recognising that a classroom-scale investigation remains narrower than a universal recommendation.
Back to contents · Next: 13. Read a stronger comparison without exaggerating it
13. Read a stronger comparison without exaggerating it
For a second paper exercise, suppose Canopy receives a new, entirely fictional controlled-comparison record. There are four matched pairs of planters, each pair chosen to be similar at the start. Within each pair, assignment to ordinary watering or the kit is random. Both members of each pair share the stipulated light conditions, container specification and starting-material protocol. Rain is excluded equally in this imagined setup. The measurement window is five school days, and only deliberately added water is counted in both groups.
The ordinary-watering volumes for pairs one to four are 9, 10, 11 and 10 litres. The corresponding kit volumes are 7, 8, 7 and 6 litres. Ordinary-watering growth increments are 18, 20, 22 and 20 millimetres; kit increments are 20, 19, 21 and 20 millimetres. These are supplied data for interpretation, not instructions to run an experiment with children. Any real investigation would require the educator’s normal suitability, care and safety decisions.
The ordinary group totals 40 litres and the kit group 28 litres, a difference of 12 litres or 30 per cent relative to the ordinary group. Pairwise differences are 2, 2, 4 and 4 litres, with a mean of 3 litres per planter. Both groups’ growth increments total 80 millimetres, giving the same 20-millimetre mean. The kit’s growth results are higher in one pair, lower in two and equal in one. The equal means do not mean every pair had an identical outcome.
This design makes the kit comparison more informative because assignment and conditions were specified to reduce the earlier confounding. The justified conclusion remains bounded to this supplied small, short experiment and these measured outcomes. It does not prove long-term safety, equal performance for every plant or a guaranteed thirty-per-cent saving elsewhere. The new thirty-per-cent result also does not require the earlier arithmetic to be wrong. Different designs and periods can produce different observations; the important task is to compare what each record can support.
Back to contents · Next: 14. Make a recommendation with an explicit decision boundary
14. Make a recommendation with an explicit decision boundary
The immediate twenty-planter Drip rollout should not be recommended under the supplied limits. It exceeds the staff-time ceiling, its financial-payback claim fails even under the favourable projection, and its general performance remains uncertain. That decision does not require claiming that the kit never works. It follows from the combination of an operational constraint and limited evidence. A recommendation becomes more useful when it identifies the actual barrier instead of using a broad label such as “bad technology.”
A defensible alternative is to keep the existing garden arrangement while using the paper packet for the learning unit. This option satisfies the educational brief without assuming equipment purchases or a real trial. Another defensible proposal is to seek approval for a bounded investigation of a feasible arrangement such as Manual, provided the scope, staff time and outcome measures are agreed. The packet permits proposing that next step; it does not supply approval or evidence of Manual’s water performance. The distinction should remain explicit in the student response.
If the committee values resource stewardship even when bill savings are small, students may explain that preference. They should state the purpose and acknowledge the cost. If another student gives higher priority to scarce staff time, that concern also has a place in the discussion. Neither side can present its priority as the only answer dictated by the numbers. A useful decision record states the chosen option, the constraints it meets, the benefit being sought, the uncertainty retained and the condition that would justify reconsideration.
For example: “Keep the current arrangement for now and teach from the supplied records. Do not approve the Drip rollout within a forty-minute staffing cap. Consider a separately authorised small investigation only after its workload and measurement plan fit the limits. Review the decision if a credible lower-workload design and comparable evidence become available.” This is a complete recommendation because it states both what follows now and what could change the conclusion, without assigning students authority over a real school’s spending or operations.
Back to contents · Next: 15. Combine the disciplines in one worked response
15. Combine the disciplines in one worked response
A strong integrated response can remain short. “The publicity draft overstates the records. Recorded watering fell from 40 to 26 litres, a 35 per cent reduction relative to the first period. The records exclude rainfall, and the planters moved, so they do not isolate the kit’s effect on total water use. Average stem-growth increments rose from 20 to 24 millimetres, but that measure does not establish overall plant health. A twenty-planter saving of 70 litres is a conditional projection, not a guarantee.”
The response continues: “At the fictional price, seventy litres corresponds to fourteen cents per week. Drip’s four-dollar weekly maintenance makes the limited net financial saving negative, before its ninety-six-dollar setup cost. Its fifty-five staff minutes exceed the forty-minute limit. We should keep the present arrangement while using the data for learning. A smaller investigation can be proposed separately if its full costs and workload fit the authorised limits, with no claim that Manual has already demonstrated a saving.”
Notice what each discipline contributes. Language narrows the publicity claim and preserves the difference between observation and promise. Mathematics reproduces the totals, rates, projection and resource comparison. Science identifies what the comparison can and cannot establish. Judgement combines the evidence with the committee’s constraints and explains a responsible next step. Remove any one contribution and the response loses something important. Accurate arithmetic alone would leave the causal overclaim untouched; a cautious scientific paragraph alone would leave the unaffordable workload unexamined.
The response is also open to revision. If the staffing figure changes, the feasibility conclusion should be recomputed. If a better controlled study is supplied, the scientific interpretation should be updated. If the committee explicitly adopts a different priority, the recommendation may change while the calculations stay the same. Curriculum learning is visible in those adjustments. A student who memorises the paragraph but cannot respond to a changed input has learned less of the intended reasoning than the polished final text might suggest.
Back to contents · Next: 16. Repair a plausible wrong student answer
16. Repair a plausible wrong student answer
Consider this invented answer: “AI analysed the data and proved a 35 per cent improvement. Plants grew 20 per cent more, so the garden improved by 55 per cent overall. The twenty-planter saving is 70 dollars each week, so the 96-dollar kit pays for itself in under two weeks. We should buy it because it is under the 140-dollar budget. Everyone can help with maintenance, and any uncertainty will disappear when we collect more data.” The answer is confident, connected and wrong in several distinguishable ways.
The first repair concerns the claim and units. Thirty-five per cent describes recorded watering relative to A, not a universal improvement score. The growth percentage cannot be added to it without an explicitly defined and justified model. Seventy is litres, not dollars; applying the supplied unit price gives fourteen cents. A corrected answer should show those operations in context. Telling the student only to “check the AI” would not identify the mathematical knowledge or language distinction that needs attention.
The second repair concerns evidence and constraints. The A–B comparison does not isolate causation. Passing the purchase ceiling does not pass the staffing limit. “Everyone can help” invents a resource that the brief explicitly excludes. More data collected under the same confounded design might make the description more detailed without answering the causal question. The revision therefore needs a better comparison, not simply a bigger spreadsheet. Each correction points to content the curriculum must actually teach.
The final repair concerns judgement. The student should replace immediate purchase with a decision that follows from the evidence and authorised resources. A useful revision might retain enthusiasm for investigating the kit while rejecting the rollout as presently specified. That allows the learner to correct the reasoning without being required to adopt a negative attitude towards technology. The target is disciplined enthusiasm: curiosity accompanied by defined questions, honest quantities and respect for limits, rather than either automatic acceptance or automatic dismissal.
Back to contents · Next: 17. Audit a curriculum that looks impressive but fails the brief
17. Audit a curriculum that looks impressive but fails the brief
Now evaluate an invented curriculum proposal called Bright Future. It allocates ninety minutes to prompting demonstrations, ninety minutes to generating a promotional video, sixty minutes to designing a dashboard, sixty minutes to group presentations and forty minutes to an AI-written reflection. The total is 340 minutes. The available teaching time is 300 minutes. The proposal does not identify which forty minutes would be removed or which learning objective would survive the removal.
Bright Future also assumes twenty-four simultaneous devices and personal AI accounts for every learner. The workshop has six shared devices and prohibits required personal accounts or home device use. Its materials cost is 190 dollars against a ceiling of 180. Teacher preparation is estimated at 140 minutes against ninety. Marking six videos at twenty-five minutes each requires 150 minutes against 120. These are supplied estimates for the exercise; the point is to test the plan against them rather than pretend that attractive resources appear for free.
The educational mismatch is more serious than the ten-dollar materials overrun. The proposal gives no protected practice in selecting the percentage denominator, distinguishing measured watering from total water, diagnosing the comparison or defending a recommendation under the staff limit. A dashboard could display the wrong model beautifully. A video could repeat the publicity draft persuasively. A generated reflection could describe understanding without demonstrating it. The plan’s deliverables therefore do not provide enough evidence that its stated interdisciplinary capabilities have been taught.
The appropriate repair is not to attach four subject labels to the same activities. It is to allocate time to the specific reasoning the packet demands, simplify production, preserve accessible participation and make room for a different problem. Some AI demonstration can remain if it serves one of those purposes. Novelty is neither a reason to reject the activity nor a reason to exempt it from the constraints. A credible curriculum proposal must work within its actual time and resources while preserving its essential learning content.
Back to contents · Next: 18. Build a feasible six-lesson alternative
18. Build a feasible six-lesson alternative
The replacement plan uses six fifty-minute lessons. Lesson one is language: ten minutes reading the brief, fifteen identifying and classifying claims, fifteen repairing a short passage, and ten comparing revisions. Lesson two is mathematics: ten minutes interpreting units and reference quantities, fifteen studying the water calculation, fifteen completing cost and workload calculations, and ten explaining a changed denominator. Each lesson totals fifty minutes, and each leaves a trace of the capability it is intended to develop.
Lesson three is science: ten minutes separating the descriptive and causal questions, fifteen diagnosing the A–B comparison, fifteen interpreting the stronger paired record, and ten stating remaining limits. Lesson four is judgement: ten minutes distinguishing facts, calculations and priorities, fifteen checking option feasibility, fifteen comparing two defensible recommendations, and ten revising a decision after a changed constraint. A teacher may adjust these allocations for the class, but any adjustment should preserve the named reasoning rather than merely compress the tasks.
Lesson five integrates the work: ten minutes reviewing selected errors, twenty producing a one-page group recommendation, ten exchanging a focused challenge with another group, and ten revising the recommendation. Lesson six uses the independent transfer packet: five minutes orienting to the new brief, thirty completing individual answers, ten discussing the supplied answer reasoning, and five recording a specific next learning need. The answer discussion happens after independent work. This article provides the answers below so the task can also be used for self-study.
The sum is 6 × 50 = 300 minutes. The plan requires no live AI account. A teacher may show a fixed AI-style response as the wrong-answer object, or use a teacher-controlled demonstration if local policy and access permit. Six devices are sufficient because no more than one is needed per group, and all essential work also exists on paper. The plan does not claim to teach the whole of language, mathematics, science or ethics in six lessons; it demonstrates how a bounded interdisciplinary unit can preserve a contribution from each.
Back to contents · Next: 19. Count preparation, materials and marking honestly
19. Count preparation, materials and marking honestly
The replacement’s fictional preparation estimate is twenty-four minutes to assemble the packet, twelve to verify the mathematical key, eighteen to prepare the science comparison cards, twelve to select and check the fixed demonstration, twelve to prepare the transfer sheets and six to check access arrangements. This totals 84 minutes, leaving six within the ninety-minute ceiling. These are planning assumptions, not measured teacher performance. If local preparation actually takes longer, the plan needs more support or a simpler scope rather than an invisible claim on teacher time.
The materials estimate is 24 dollars for printing, 18 for reusable demonstration materials, 30 for accessible-format preparation and 12 for consumables, totalling 84 dollars. Devices are existing resources with no additional charge in this packet. The remaining materials headroom is 180 − 84 = 96 dollars. This headroom does not authorise spending it, and it cannot automatically be transferred to the separate garden equipment budget. The exercise deliberately keeps budget categories visible so a learner cannot solve one constraint by quietly borrowing from another.
The marking plan allocates two minutes to each of thirty individual transfer sheets, giving sixty minutes; six minutes to each of six group recommendations, giving thirty-six minutes; and eighteen minutes to review patterns and prepare feedback. The total is 114 minutes, leaving six within the 120-minute cap. The brief checks use a compact key and short responses. If the desired feedback needs longer essays or extensive individual conferences, the arithmetic should be revised openly. Calling marking “AI-assisted” is not evidence that the required time has vanished.
Access is a design requirement rather than an afterthought. The plan provides readable printed data, permits approved text-access supports and does not award marks for video polish, device speed or subscription features. A student can point to a calculation, speak an explanation through an agreed support or write a concise response, provided the intended reasoning remains visible. These are proposed accommodations to the task format, not diagnoses or claims about particular learners. Actual supports should follow the educator’s knowledge and responsibilities for the class.
Back to contents · Next: 20. Look for each discipline in the learner’s response
20. Look for each discipline in the learner’s response
A curriculum claim should be accompanied by a modest description of what evidence would count. For language, look for a revised sentence that preserves the observation while removing unsupported causation. For mathematics, look for the correct denominator, unit and constraint comparison. For science, look for a specific design limitation and a relevant improvement. For judgement, look for a recommendation that fits the authority and resource limits while making its priority explicit. These checks are part of evaluating curriculum coherence, not a complete assessment system.
The group recommendation and individual transfer answer serve different purposes. The group page shows how the disciplines can be integrated in a collaborative product. The individual answer gives a more direct view of each learner’s response to the new packet under the stated support conditions. Neither should be treated as a complete measure of intelligence, character or future potential. The work samples answer a narrower question: which parts of this unit’s intended reasoning are visible, and which require further teaching?
A planning key can use four two-point dimensions for the transfer response: language, mathematics, scientific interpretation and judgement. In each dimension, two points mean the essential reasoning is accurate and explained; one means there is a relevant partial step but an important gap; zero means the supplied response does not yet show that contribution. Eight is the maximum. This simple example helps allocate attention across the unit. It is not a validated scale, a school grading policy or a basis for high-stakes decisions.
For example, a learner might correctly calculate both rates but claim the sign caused the entire change. The feedback should preserve credit for the mathematics while teaching the causal distinction. Another learner might identify confounding but omit the resource constraint; that calls for a different next task. A single total can conceal these different patterns. The curriculum reviewer should therefore inspect the disciplinary explanation alongside any score, keeping the focus on what should be learned next rather than treating a number as a complete educational judgement.
Back to contents · Next: 21. Independent transfer: the complete Juniper Library packet
21. Independent transfer: the complete Juniper Library packet
Juniper Library is a fictional workshop with a different proposal. It wants to reduce discarded printed handouts at open learning events. The publicity draft says: “Our new digital sign cut waste by 40 per cent. It proves that every event should replace paper, and the change will pay for itself immediately.” The class must interpret the claim, calculate fair comparisons, assess the evidence and choose a feasible response. No real library, attendee record or individual behaviour is represented. All counts are invented aggregate teaching data.
Before the sign, one week had 600 event attendances, 500 handouts printed and all 500 distributed, and 200 discarded handouts counted at the end of events. After the sign, the next week had 400 attendances, 300 handouts printed and all 300 distributed, and 120 discarded handouts counted. These are the complete printed counts for the two weeks; there were no undistributed copies. “Attendance” means one event visit, not a count of unique people. The second week contained different event topics and shorter sessions. The packet does not identify which attendees used the sign, whether they read the handout, or what happened to material taken away.
The library can spend at most 90 dollars once and 3 dollars per week on incremental running costs, with at most thirty staff minutes weekly in total. Status quo costs nothing to establish and needs fifteen staff minutes weekly. Digital costs 80 dollars to establish, 4 dollars weekly and forty staff minutes weekly. Mixed costs 25 dollars to establish, 1 dollar weekly and twenty-five staff minutes weekly. Mixed combines a reusable notice with optional printed handouts; no measured waste-reduction result is supplied for it. The fictional printing cost is 0.04 dollars per handout.
The teaching brief is four forty-five-minute lessons for twenty-four learners in six groups of four, using six existing shared devices. The materials cap is 100 dollars, preparation cap sixty minutes and marking cap eighty minutes. Personal accounts and homework device access cannot be required. Proposal Flash allocates fifty minutes to video, fifty to dashboard design, fifty to presentations and fifty to generated reflections. It budgets 115 dollars, seventy minutes of preparation and ninety minutes of marking. Treat these as complete stated estimates for the exercise.
Back to contents · Next: 22. Transfer questions to attempt before reading the answers
22. Transfer questions to attempt before reading the answers
First, repair the publicity claim in two or three sentences. Identify which reduction the forty-per-cent figure actually describes and which causal or value claims it fails to establish. Second, calculate the raw change in discarded handouts, discarded handouts per one hundred attendances in each week, and the percentage reduction in that attendance-adjusted rate. Third, calculate the discarded share of distributed handouts in each week. Explain why these quantities answer different questions even though they use some of the same counts.
Fourth, calculate the change in printing expenditure between the two weeks. Decide whether that amount establishes a recurring saving caused by the digital sign or immediate payback. Fifth, test Digital and Mixed against all three operating limits. State one feasible recommendation, including what remains unknown. Sixth, identify two features of the evidence that weaken a causal conclusion, and sketch one more informative comparison. Keep the proposed comparison within a paper-design exercise; do not recruit real attendees or collect personal data.
Seventh, audit Flash against the learning-unit constraints. Calculate each stated overrun and name the disciplinary content that its activities fail to protect. Eighth, propose a four-lesson allocation that fits 180 minutes and preserves language, mathematical, scientific and judgement work. Use the following available cost inputs for a repaired plan: printing 20 dollars, accessible-format preparation 24 dollars and reusable materials 16 dollars. Preparation components are eighteen, twelve, twelve and twelve minutes. Marking can use two minutes for each individual response, four minutes for each group sheet and six minutes for feedback planning.
Finally, consider two changes independently. In Change A, Digital’s running cost is reduced to 2 dollars weekly and total staff time to twenty-eight minutes; its setup cost remains 80 dollars, and no new outcome evidence is supplied. In Change B, the original operating costs remain, but a careful comparison provides convincing evidence of reduced discarded handouts for the tested events. For each change, say which conclusion changes and which barrier remains. Do not combine the changes unless the question explicitly supplies both together.
Back to contents · Next: 23. Transfer answers: quantities and interpretation
23. Transfer answers: quantities and interpretation
The raw discarded count falls from 200 to 120, a reduction of 80 handouts. Relative to 200, that is 40 per cent. Attendances also fall, from 600 to 400, so the raw count alone does not describe waste per visit. Before, 200 ÷ 600 × 100 is approximately 33.33 discarded handouts per one hundred attendances. After, 120 ÷ 400 × 100 is 30 per one hundred attendances. These rates allow a comparison that accounts for the different number of event visits.
The rate difference is about 3.33 discarded handouts per one hundred attendances. Using exact fractions, the relative reduction is one tenth, or 10 per cent: the earlier rate is one third per attendance, the later rate is three tenths, and their difference of one thirtieth is one tenth of one third. Rounding the two rates first can produce a slightly different final decimal. Retaining the fractions or sufficient precision avoids that avoidable discrepancy. Ten per cent is an adjusted description, not a proven causal effect.
The discarded share of distributed handouts is 200 ÷ 500 = 40 per cent before and 120 ÷ 300 = 40 per cent after. Thus the packet shows no change in that particular share. It simultaneously shows a forty-per-cent reduction in raw discarded count and a ten-per-cent reduction in discards per attendance. There is no contradiction: the denominators differ. A responsible report names the denominator alongside the percentage, rather than selecting whichever number sounds most impressive for the preferred proposal.
Printing expenditure is 500 × 0.04 = 20 dollars before and 300 × 0.04 = 12 dollars after, an eight-dollar difference. Attendance, event topics and session length changed, so this does not establish a recurring eight-dollar benefit caused by the sign. Immediate payback is also not shown: Digital has an eighty-dollar setup cost and recurring costs. Even a conditional calculation would need a justified future printing comparison. A complete model repair is: “Counted discarded handouts fell from 200 to 120, a forty-per-cent raw-count reduction between the two weeks. Because attendance, event topics and session length also changed, the records do not isolate the sign’s effect or establish that all events should abandon paper. They also do not establish immediate payback.”
Back to contents · Next: 24. Transfer answers: science, options and curriculum repair
24. Transfer answers: science, options and curriculum repair
Digital fits the ninety-dollar setup cap with ten dollars to spare, but exceeds the weekly cost cap by one dollar and staff-time cap by ten minutes. Mixed fits all three, leaving sixty-five dollars of setup headroom, two dollars per week and five staff minutes. A feasible recommendation is to retain the existing arrangement or propose Mixed for a bounded, separately authorised investigation. Neither conclusion claims that Mixed has demonstrated a reduction. Replacing all paper also raises an access question that the waste counts cannot answer: attendees may need different ways to obtain the material.
The comparison is weakened by unequal attendance and changed event topics or duration, as well as its simple before–after timing. A more informative paper design would compare similar events using a planned assignment of the sign condition, a consistent counting method and both attendance and distribution totals. It would state how departures are recorded and preserve accessible information in every condition. The exact design would depend on practical constraints. The useful student answer names a relevant improvement and its remaining limits, rather than promising that a larger sample alone eliminates every bias.
Flash totals 200 minutes, exceeding 180 by twenty. Its materials overrun is fifteen dollars, preparation overrun ten minutes and marking overrun ten minutes. Its activities fail to protect precise claim repair, selection of denominators and cost calculations, diagnosis of the causal comparison, and a feasible decision supported by reasons. A repair can allocate forty-five minutes each to claim reading, rates and costs, evidence and options, and an integrated recommendation with a short individual changed-case response. Within each lesson, protect time for explanation and response rather than assigning the whole slot to production. This plan preserves four disciplinary contributions while reducing the demand for polished media.
The supplied repaired materials total 20 + 24 + 16 = 60 dollars, leaving forty. Preparation totals 18 + 12 + 12 + 12 = 54 minutes, leaving six. Marking totals 24 × 2 + 6 × 4 + 6 = 78 minutes, leaving two. In Change A, Digital becomes feasible under all operating caps: eighty dollars, two weekly dollars and twenty-eight minutes all fit. Its causal evidence remains weak. In Change B, the evidence improves for the tested events, but the original weekly-cost and staffing failures remain. Stronger evidence and greater feasibility solve different problems.
Back to contents · Next: 25. Adapt depth without pretending one project is the whole curriculum
25. Adapt depth without pretending one project is the whole curriculum
For younger learners or those new to the mathematics, simplify the numbers while preserving the reasoning structure. Compare whole counts, use a labelled bar model and ask whether the two groups had the same number of visits or planters. The teacher can supply the percentage calculation and focus the student’s work on interpreting it. Simplification should remove unnecessary computational load, not erase the difference between a count, a rate and a cause. The same text can be read aloud or discussed through an appropriate access format.
For learners ready for more mathematics, introduce symbolic relationships and sensitivity checks. Let the projected weekly water saving be S litres, the fictional price be p dollars per litre and weekly maintenance be m dollars. The limited recurring financial benefit is Sp − m. A positive value is necessary for a positive simple payback on a positive setup cost within that restricted model. Students can explain which variables would need to change and why that still leaves the evidence for S unresolved. This extension develops the model rather than merely adding harder arithmetic.
For deeper language work, compare a publicity paragraph, an observation note and a committee recommendation. Ask how audience and purpose change what information must be foregrounded without permitting false claims. For deeper science work, compare alternative designs and discuss what each can identify. For deeper judgement work, introduce a clearly specified competing priority, such as preserving paper access within a fixed cost cap, and require a recommendation that addresses it. Extensions should add a genuine reasoning demand, not repeat the same answer in a more elaborate format.
Keep the wider curriculum visible. This unit does not replace sustained reading, mathematical progression, substantive science or thoughtful engagement with literature, history and the arts. It demonstrates a connection among disciplines around one bounded decision. Schools and families should compare any proposed AI-related programme with what it displaces. If a new activity consumes practice time for an essential concept, its benefit must justify that trade-off. A useful programme description says what it adds, what it assumes learners already know and what continues elsewhere.
Back to contents · Next: 26. Judge future curriculum proposals by what they preserve
26. Judge future curriculum proposals by what they preserve
A proposal for education around advanced AI should be readable as an argument. It needs a purpose, a body of content, a sequence of work, a realistic account of resources and a way to examine whether the intended learning is visible. If one of these is missing, ask for it directly. A vendor’s demonstration or a learner’s attractive output can be interesting evidence about a tool’s operation, but it cannot substitute for all five parts of the curriculum argument.
The stronger future possibility deserves attention too. Suppose a future AI system provides correct calculations and reliable causal analysis for the supplied records. The curriculum could use that support differently, perhaps spending less time correcting machine errors and more time interpreting models, comparing explanations and examining priorities. It would still need to specify what students are learning through those activities. Reliable assistance changes the design possibilities; it does not itself decide the distribution of lesson time among language, mathematics, science and judgement.
The practical conclusion is a balanced one. Preserve disciplinary knowledge and give students opportunities to connect it. Use assistance where it serves a named learning purpose. Keep the limits of evidence, resources and authority visible. Require curriculum proposals to explain the capabilities behind their products and to account honestly for what they consume. Avoid promises that a particular subject is permanently job-proof, that a fashionable tool replaces all foundations or that one short project demonstrates every form of understanding.
For students, the immediate learning target is concrete: read the exact claim, reproduce the relevant quantities, ask what the comparison can establish and defend a feasible next step. For adults choosing a programme, the target is equally concrete: check whether students actually practise those contributions and whether the proposed time and support make that possible. The garden and library packets provide inspectable examples of that decision. Their value is in the reasoning they expose, not in predicting the future of schooling from a fictional spreadsheet.
Back to contents · Next: Sources and further reading
Sources and further reading
Source scope checked on 1 October 2026. UNESCO’s 2024 student AI framework is used as a curriculum framework, the National Research Council’s science-practices chapter as a conceptual reference, the IES 2007 guide as bounded instructional guidance, and Roediger and Karpicke’s 2006 primary study for a limited point about delayed retention. These sources do not validate this original unit or establish learning outcomes under hypothetical ASI. The linked sources appear beside the specific propositions they support; all case quantities, options, budgets and example answers are invented for this guide.
For practical AI-user and technical learning priorities, continue with What You Actually Need to Learn About Super Intelligence. For the wider institutional question, read Future of Schooling. The related discussion of Human Augmentation addresses assistance and human capability. These are neighbouring questions; this guide’s particular job is choosing and evaluating the disciplinary content of a bounded curriculum around advanced AI.
Back to contents · Return to the Super Intelligence master guide
