AI and Curriculum Progression | What Learners Need Before They Delegate

The same tool can serve different purposes in different lessons. A spelling suggestion may be useful during a lesson on organising an explanation. During a focused spelling assessment, the same suggestion could remove the skill the teacher is trying to observe. An AI-generated solution might help a learner compare methods after they understand an equation, while concealing a missing concept if it replaces their first encounter with the method. The decision begins with the goal.

Teachers can state that goal in terms of an observable capability. “Explain why both sides remain equal” is clearer than “complete an algebra activity.” “Connect a process step to its purpose” is clearer than “write a polished paragraph.” Once the goal is specific, it becomes easier to decide which assistance supports it and which assistance performs the central intellectual work for the student.

This also makes delegation less mysterious for learners. They can understand why AI is permitted to generate examples in one task and restricted to hints in another. The rule follows the purpose of the learning rather than appearing as a judgement about whether the student is trustworthy. Clear task boundaries allow learners to use assistance confidently while keeping the relevant reasoning available for teaching and assessment.

Contents · Next chapter

A learner can be well prepared for one kind of delegation and unprepared for another. Someone who can check a numerical calculation may still struggle to judge the evidence behind a historical explanation. A fluent writer may be able to revise a paragraph while needing help with the scientific concept it describes. Treating readiness as a single personal score hides these differences and can lead to the wrong form of assistance.

Age and school level can inform how a teacher presents material and organises supervision, but neither tells the whole story of what an individual can do. Prior teaching, language access, experience with the subject and the task’s complexity all matter to the planning decision. The model proposed here therefore uses task-specific evidence. It does not claim that every learner should reach a particular delegation stage at a fixed birthday or year level.

A useful record might say that a student can explain equivalent expressions with whole-number examples but needs support when negative signs appear. That is more actionable than calling the student “not ready for AI.” The teacher can permit bounded help, teach the missing relationship and check the next attempt. Readiness becomes something instruction can develop, rather than a gate that closes off further learning.

Contents · Previous chapter · Next chapter

A prerequisite map identifies what a learner needs in order to make sense of a task and recognise a significant error. It should be small enough to guide a lesson. For algebraic equivalence, the map might include the meaning of equality, multiplication over a bracket and inverse operations. It need not include every topic the learner will eventually study. The purpose is to identify the dependencies relevant to the present question.

The map should also distinguish different kinds of difficulty. A student might understand equality but misread a minus sign. Another might calculate accurately while believing that the equals sign means “write the answer next.” Both may produce a wrong solution, but they need different teaching. AI can help a teacher generate diagnostic examples, provided the teacher checks that those examples actually distinguish the suspected difficulties.

The following map is illustrative. It is not a statement of MOE or SEAB sequencing, and the entries should be adapted to the lesson and learner. The final column identifies a task in which AI could become useful once the relevant checking knowledge is available. The teacher still needs to inspect actual performance before changing the amount of support.

Contents · Previous chapter · Next chapter

This proposed framework uses four broad forms of instructional help: a demonstrated example, a partly completed task, a hint or question, and a review of an independent attempt. They are options for planning, not a compulsory ladder through which every learner must pass. A learner may need a demonstration for a new concept and only a review for a familiar one within the same lesson.

A demonstrated example makes a method visible. A partly completed task asks the learner to supply a meaningful part. A hint directs attention without giving the whole answer. A review invites the learner to compare their work with feedback and decide what to revise. Each form should preserve some relevant thinking for the student. A hint that effectively states the answer may provide much more assistance than its label suggests.

Teachers can prepare or review the help before learners encounter it. An AI-generated example should not be assumed correct because it is neatly formatted. When a tool is used interactively, the teacher can define what kind of response is wanted and inspect whether it stays within that purpose. The form of help should be judged by what it actually gives the learner, rather than by the name attached to a feature.

Contents · Previous chapter · Next chapter

A correct answer is useful evidence, but it does not explain how the learner obtained it. Before increasing delegation, a teacher can ask for a short account of the method, an example of a common error and a check using different information. These requests reveal different aspects of understanding. They should be proportionate to the task and should not turn every lesson into a lengthy examination.

For an algebra task, the learner might explain why an operation is applied to both sides, test a proposed answer in the original equation and reject a step that changes the relationship. For explanatory writing, the learner might identify the key content, show the connection between two sentences and recognise an invented detail. A student who can do these things has a basis for evaluating certain kinds of AI assistance.

Readiness can also be partial. The teacher might allow AI to suggest alternative wording while retaining a source packet that limits the factual content. Or a learner might use a calculator while explaining the algebraic transformation themselves. A useful decision specifies what is ready to be delegated and what remains the focus of instruction. It does not require a vague judgement that the learner is either entirely independent or entirely dependent.

Contents · Previous chapter · Next chapter

Begin with the equation 3(x + 2) = 18. Before asking for a solution, ask the learner what the expression on each side represents. The left side is three times the quantity x + 2; the right side is 18. The equation states that the two sides have equal values for the solution being sought. This meaning is the foundation for deciding whether a later step preserves the relationship.

The teacher can use a diagram, counters or spoken explanation when appropriate, while keeping the connection to the symbols explicit. Three equal groups, each containing an unknown amount and two more, have a total of eighteen. Dividing the total into three equal groups gives six in each group. Removing the extra two leaves four for the unknown amount. The representation should help explain the equation, rather than become a separate activity whose relation to the symbols is unclear.

A first readiness check asks the learner to test x = 4 in the original equation: 3(4 + 2) = 18. The result is true. Then ask whether x = 5 works and why it does not. These checks show whether the learner treats a solution as a value that satisfies the equation. They also provide a concrete way to question an AI answer later, even before the student has learned several alternative solution methods.

Contents · Previous chapter · Next chapter

One valid method divides both sides of 3(x + 2) = 18 by three, giving x + 2 = 6, then subtracts two from both sides to obtain x = 4. Another valid method expands the left side first: 3x + 6 = 18. Subtracting six gives 3x = 12, and dividing by three again gives x = 4. The routes differ, but each preserves the equation’s solution.

Now present an intentionally incorrect step, clearly labelled as a teaching example: 3(x + 2) = 18 becomes 3x + 2 = 18. Ask the learner to identify the first place the relationship changed. The error is in the expansion: the multiplier applies to both terms inside the bracket. The teacher can test the expressions at x = 1. The original expression equals 9, while the incorrectly expanded expression equals 5. They are not equivalent.

AI can help produce further contrasts after the teacher has checked them. The learner’s task is to explain the difference, not simply choose the solution that looks familiar. A useful response refers to the distributive relationship or shows a valid counterexample. Saying that the teacher’s version is correct is insufficient evidence that the student can recognise the error independently.

The counterexample has a specific role. One unequal result is enough to disprove the claim that two expressions are equivalent for every permitted value. Agreement at one chosen value does not prove that they are always equivalent. This distinction gives the lesson a small but meaningful step towards mathematical reasoning: testing can reveal an error, while a general justification explains why a transformation is valid.

Contents · Previous chapter · Next chapter

For the next task, use 2(x − 3) = 10. Give the first step, x − 3 = 5, and ask the learner to complete and explain the solution. Adding three to both sides gives x = 8. Substitution into the original equation confirms 2(8 − 3) = 10. The learner now supplies more of the work, while the remaining support keeps attention on the relationship the lesson is developing.

If the student can explain that task, present 5(2 − y) = 15 without a worked first step. Dividing both sides by five gives 2 − y = 3. Subtracting two gives −y = 1, so y = −1. Checking the original expression gives 5(2 − (−1)) = 15. The changed order inside the bracket makes this more than a substitution of new numbers; it tests how the learner handles the negative sign and the unknown’s position.

The teacher should decide whether that variation is appropriate for the learner’s taught knowledge. If negative numbers have not been established, it is a new teaching need, not evidence of failure. A learner may first need a simpler example or an explanation of additive inverses. The progression should expose the next useful relationship without treating unintroduced content as something the student should already know.

AI assistance can now be limited to a question such as “Which operation would preserve equality and simplify the expression?” After an independent attempt, the tool might suggest a checking method. The learner should still explain the transformation and test the result. Delegating the generation of practice questions becomes reasonable when the teacher or learner can identify whether those questions and solutions fit the intended concept.

Contents · Previous chapter · Next chapter

Explanatory writing requires more than a sequence of grammatical sentences. The writer needs a clear account of what is happening and why the selected details matter. AI can quickly produce fluent prose, which makes it especially useful to establish the learner’s content understanding before asking for stylistic improvements. Otherwise, polished language can make a gap in understanding harder to notice.

For a bounded lesson, use a fictional library process with a short teacher-written source sheet. A returned book is scanned into the record. A staff member checks its condition. A damaged book is set aside for repair; an undamaged book is placed on its correct shelf. The record is marked available only after shelving is confirmed. These are invented rules for the exercise, not a description of every library’s practice.

Ask the learner to explain the process aloud, arrange its steps or draw a simple flow before drafting a paragraph. The key question is why a returned book is not necessarily available immediately. The learner should connect scanning, condition checking and confirmed shelving to that question. A student who can recite the sequence but cannot explain the distinction may need further discussion of the process before an AI rewrite becomes useful.

Contents · Previous chapter · Next chapter

A first draft might say, “The book is returned. It is scanned. Someone checks it. Then it goes on a shelf.” The sequence is broadly recognisable but does not fully answer the question. Ask the learner which step explains why availability is delayed and what the source permits them to say. They can then add that the book’s condition must be checked and its shelf location confirmed before the record shows that it is available.

The teacher can model one connection without writing the whole paragraph: “Scanning records the return, but the book is marked available only after its condition and shelving have been checked.” The learner should inspect whether this wording preserves the fictional rules. If it blurs the distinction between condition checking and the final shelving confirmation, they can revise it to make the sequence clearer. Writing quality depends on meaning as well as fluency.

AI can propose two structures: one following the process in order, and another beginning with the distinction between returned and available. Students compare which structure answers the question more clearly for the intended reader. They should be able to explain the reason for their choice. A model’s preference is an option to consider, not evidence that the chosen structure is automatically the best one.

A useful final paragraph might explain that scanning records the return, inspection determines whether repair is needed, and shelving confirmation establishes that an undamaged book is ready to be found. The student should write the final version in their own chosen form and be able to account for each factual statement. The lesson’s focus is a clear explanation grounded in the supplied process, not the imitation of a generated paragraph.

Contents · Previous chapter · Next chapter

Give the learner an invented AI revision that adds, “National library rules require every book to be quarantined for twenty-four hours.” The added sentence is not supported by the fictional source sheet. Label it as a deliberately fabricated teaching example, rather than presenting it as a real policy or a claim about a particular AI product. Ask students how they know it should not be included.

The expected answer is about evidence. The source does not state that rule, and the assignment does not authorise inventing a policy. The sentence may sound official and make the paragraph appear more detailed, but those qualities do not supply support. The learner can remove it, record why it was rejected and identify what kind of authoritative evidence would be required if the question were about an actual library.

For a later task, provide a different fictional process, such as equipment being returned, inspected and approved for reuse. Reduce the help with planning and ask the learner to explain why return and readiness are distinct. If the student can identify the new process’s relevant conditions and avoid importing details from the library example, they are showing transfer. Reusing the same wording without checking the new rules would demonstrate a different, more limited capability.

Contents · Previous chapter · Next chapter

In science, the learner may need to distinguish an observation from an explanation before asking AI for hypotheses. They may need to understand which variable changes and what is measured before judging a suggested comparison. A teacher can provide a safe, already approved investigation and ask students to identify the question, the observations and the limits of the conclusion. AI can widen the possible explanations only within a framework the learner can inspect.

In history, the prerequisites are different. A student needs to identify who produced a source, when it was produced and what claim it can support. A later summary may interpret an earlier account, but it is not the same kind of evidence. AI can organise a source comparison while the learner remains responsible for distinguishing testimony, interpretation and uncertainty. The wider discussion in AI and the Paradigm Shift in History develops that distinction.

These examples show why a generic instruction to “think critically about AI” is too broad to guide a curriculum. Checking an algebraic transformation, a scientific comparison and a historical claim involves different knowledge. Some habits transfer across subjects, such as asking for evidence, but their application must be taught in context. A progression plan should name both the shared habit and the subject-specific knowledge that makes it useful.

Contents · Previous chapter · Next chapter

Reducing assistance should not mean removing every support. A learner may need text enlargement, a screen reader, speech input, a familiar communication format or additional processing time to participate effectively. Whether a support affects the assessed capability depends on the task’s purpose. Teachers should distinguish assistance that provides access from assistance that supplies the particular reasoning or response they intend to observe.

For example, a learner can explain why an equation remains equivalent through speech, a diagram or an accessible writing method when handwritten fluency is not the goal. A learner who is still developing English proficiency may understand the library process but need help expressing a connection. The teacher can assess the content understanding and the language skill separately, while providing appropriate instruction in each. One difficulty should not automatically be taken as evidence of the other.

The progression record should state which supports were available. This makes later comparisons more meaningful and prevents independence from becoming an unclear claim. It also helps colleagues understand what a student has demonstrated. The aim is to see the relevant capability under suitable conditions, not to make the task artificially inaccessible or to withdraw an established accommodation as a reward or penalty.

Contents · Previous chapter · Next chapter

An attractive final document shows what the learner produced with the available tools and support. A short follow-up can reveal what they understand. Ask them to explain one important choice, identify a possible error or apply the idea to a fresh example. The question should target the lesson’s purpose. It should not become a surprise test of vocabulary, speed or memory that the task was never designed to teach.

In algebra, ask why the incorrect expansion changes the expression, or whether a different solution method preserves the same solution. In explanatory writing, ask which sentence depends on the source and which words connect the steps. These checks can be brief. Their value comes from being well chosen, rather than from adding a large amount of assessment after every activity.

A later task can examine retention, while a changed situation can examine transfer. The teacher should describe the evidence honestly. Success immediately after a demonstration does not establish long-term retention. Completing one similar example does not establish broad mastery across every variation. The record can remain specific: the learner explained the relationship independently in these tasks, under these conditions, and the next check will examine a defined extension.

Keep earlier ideas in view when the curriculum moves on. An equation with a new feature can still ask the learner to explain why equality is preserved. A longer explanation can still require every factual detail to be traced to its source. These recurring checks need not reproduce the whole original lesson. They can reveal whether a foundation remains usable when attention shifts to something more demanding. If the relationship has become uncertain, the teacher can provide a focused reminder and inspect another attempt before making a larger decision about progress.

Contents · Previous chapter · Next chapter

A practical progression record can contain the target capability, the support provided, the learner’s evidence and the next teaching decision. For the algebra sequence, it might record that the student explained division on both sides but confused the negative sign in the transfer problem. The next step is then a focused lesson on that relationship, followed by another check, rather than a general instruction to do more algebra.

For writing, the record might show that the learner can explain the process orally but introduces unsupported detail when revising for a formal audience. The next lesson can compare precise explanation with unnecessary authority claims. AI may help prepare contrasting examples, but the teacher decides which contrast addresses the observed problem. A short, specific record is more useful than a broad statement that the learner needs greater confidence or better AI skills.

The record should help the learner understand the decision too. A teacher can say, “You explained the first method clearly; now we will check whether it still makes sense when the unknown is in a different position.” This gives the next task a purpose. It avoids treating progression as a hidden judgement that students discover only through a mark.

Contents · Previous chapter · Next chapter

A need to revisit a concept is information about the next lesson. It does not establish that the learner has been lazy, that the tool has permanently harmed them or that all previous work was worthless. The teacher should first identify what the learner actually did and what support was available. A copied method, a misread symbol and an unfamiliar term can look similar in a final answer while requiring different responses.

A useful return is narrow and connected. If a student expands a bracket incorrectly, revisit multiplication over a sum with a small example and reconnect it to the equation. If a paragraph invents a condition, return to the source sheet and ask which statement supports each sentence. The learner should see how the repaired foundation helps them resume the larger task. Repeating an entire unit may be unnecessary when the missing relationship is specific.

The amount of help can increase temporarily. A teacher may restore a worked step, reduce the number of changing features or allow an oral explanation before another independent attempt. This is a planned adjustment, not a reversal of the learner’s identity as someone who can use AI. Progression includes the ability to recognise a gap, obtain the right help and return to the task with greater control.

Contents · Previous chapter · Next chapter

UNESCO’s AI Competency Framework for Students, published in 2024, describes twelve competencies across four dimensions, including human-centred thinking, ethics, AI techniques and system design. Its three progression levels are Understand, Apply and Create. It provides a framework for curriculum thinking. The task-specific examples in this article are our own instructional proposal; the UNESCO framework does not validate these particular lessons or supply the age thresholds this article deliberately avoids.

The US Institute of Education Sciences’ 2007 guide on organising instruction and study recommends approaches including alternating worked solutions with problem solving, revisiting learning over time and asking explanatory questions. This is broader learning guidance, predating current generative AI systems. It supports considering established instructional principles when planning assistance; it is not direct evidence that any particular chatbot or prompt sequence will produce the intended learning.

Singapore MOE’s current public AI-in-Education Framework page, updated on 10 September 2026, identifies Agency, Inclusivity, Fairness and Safety as its four principles. These provide relevant local context for responsible use. Teachers must follow the requirements applicable to their setting and tools. The progression model here should be read as a suggested teaching approach, with effectiveness judged through learners’ actual work and suitable review.

Contents · Previous chapter · Next chapter

A class need not move through identical amounts of help at the same moment. The learning goal can remain shared while the support differs. In the algebra lesson, one learner might study a worked example, another complete a missing step and another explain why an apparently valid transformation fails. All three can contribute to a discussion of equivalence. Different tasks should reflect observed learning needs, rather than become permanent ability labels.

Teachers can prepare a small set of variants around the same concept. AI may help generate candidates, but the teacher should check the mathematics, the intended difficulty and the specific relationship each task reveals. Changing several features at once can make an error hard to interpret. If a learner moves from whole numbers to fractions, negative signs and unfamiliar notation simultaneously, it may be unclear which change created the difficulty. A deliberate sequence changes what matters for the next teaching decision.

Peer discussion can make reasoning visible, provided the teacher can still see each learner’s contribution. One student can propose a step, another explain its justification and another test the result. Rotate those roles so that the same student does not always supply the reasoning while others record it. A brief individual follow-up can then show which ideas each person has taken from the discussion. The group product and the individual evidence answer different questions.

Parents and tutors can use the same principle on a smaller scale. Ask which specific relationship is uncertain and coordinate the next explanation with the learner’s current work. Avoid adding a second, unrelated method simply because an AI tool suggests it. An alternative can be valuable when the learner can compare the methods meaningfully; otherwise, a clear explanation of the existing method may be the better next step. Progression depends on the connection between lessons, not on the number of different explanations presented.

Contents · Previous chapter · Next chapter

The “centre” in this series is established knowledge that a learner can use dependably. The “edge” is a question or situation that extends beyond what they can already handle. The movement between them should be deliberate. A student who understands equivalent transformations can explore competing methods. A student who understands a process can examine different explanations for different audiences. Foundations make exploration more inspectable, and exploration can reveal which foundations need further teaching.

The larger argument appears in AI and the Paradigm Shift in Education. The practical assignment model in AI-Powered Assignments then helps teachers decide where AI should widen the inquiry. For classroom planning, From the Centre to the Edge connects direct teaching with supervised discovery. Curriculum progression supplies the missing connection between those ideas: what must be established next for this learner and this task?

A workable next step is to choose one existing lesson, identify its central capability and map the few prerequisites that matter. Decide what assistance will be offered and what evidence will justify reducing or changing it. After the lesson, inspect the student’s explanation and a fresh example, then adjust the next task. The goal is a learner who can use increasingly capable tools while understanding more of the work, recognising limits and knowing when to return to a foundation before moving further.

Contents · Previous chapter · Continue the series