AI and Learning Transfer | Can Students Use What They Learn in a New Situation?

Imagine a student who produces a detailed comparison of two price plans with AI. The calculations are correct, the table is attractive and the recommendation is clearly written. There are still several unanswered educational questions. Can the student explain what the fixed fee means? Would they notice a new charge? Can they use the same relationship when the quantities are shown in a graph? Could they recognise that the cheapest quoted plan is unavailable for the required order?

These questions identify capabilities that the polished document may not reveal. Assisted performance describes what the learner can accomplish with the permitted system. Independent understanding concerns what the learner can explain or do without a generated solution. Retention concerns what remains after time has passed. Transfer concerns use in a changed problem or context. A task can test more than one outcome, but the teacher should say which evidence supports which conclusion.

The distinction matters in AI research as well as classroom practice. Bastani and colleagues’ 2025 high-school mathematics experiment found that unrestricted GPT-4 assistance improved practice performance while weakening subsequent unaided performance; a teaching-oriented design largely avoided that loss. Its unaided tests were within the study sessions. The finding does not establish long-term transfer, but it shows why successful assisted practice cannot answer every learning question.

The next question is constructive: what task would make the missing understanding visible? A changed fee, an unfamiliar representation or a later explanation can each reveal something useful. The check should help the learner understand their next step. It need not become a larger examination every time AI is used.

Contents · Next chapter

Near transfer usually describes applying learning to a situation similar to the original. A learner who compared two printing plans might compare another pair with different fees. Far transfer refers to a more substantial change, but distance is not a single simple quantity. A task might be mathematically similar while using unfamiliar language, or occur much later while looking almost identical. These differences should be described rather than compressed into one impressive label.

In their 2002 taxonomy of transfer, Susan Barnett and Stephen Ceci examined several dimensions along which a new situation can differ from the learning context. Their framework cautions against treating far transfer as one uniform effect. For this article, the practical implication is to record what changed: the content, setting, representation, time, social arrangement or help available. A taxonomy helps describe the question; it does not guarantee that a teaching method will produce transfer.

Consider three checks after the printing lesson. Asking about 300 flyers instead of 100 changes the quantity. Showing the plans as a graph changes the representation. Comparing two simulated water tanks changes the meaning of the quantities while preserving a relationship between starting amounts and rates. The third is farther from the original story in some respects, but it still shares a linear structure. Calling it proof of general reasoning across all subjects would overstate the evidence.

Learner familiarity also matters to interpretation. A child who knows little about printing may find the first story harder than the tank model. The teacher should notice the reading and background demands that accompany the mathematical change. Transfer is always being tested through a particular task, with particular information and conditions.

Contents · Previous chapter · Next chapter

Before opening the investigation, check the relevant foundations. Students need to distinguish a one-time fee from a charge repeated for every item. They need to work with the given currency units, multiply a rate by a quantity and compare totals. Learners using algebra also need to interpret a variable and an equality. A student can begin with a table before using symbols; the representation should make the relationship easier to understand.

The National Academies’ How People Learn II distinguishes learning goals and discusses comparing cases, examining why ideas apply and developing models as ways to support transfer. It also treats context as consequential. This is a synthesis of learning research, not evidence that the specific printing sequence below has been evaluated. The sequence uses those broad distinctions to create a practical teaching proposal.

A teacher can diagnose the starting point with two short questions. What would a plan charge before the per-item amount is added? What changes when ten more flyers are ordered? If the student multiplies the fixed fee by every flyer, explain the structure directly and practise it with small numbers. More open inquiry will not automatically repair that misunderstanding. The learner needs a reliable relationship to carry into the later variations.

Foundations and exploration can reinforce each other. The comparison gives the arithmetic a purpose, while the arithmetic allows the student to challenge a plausible recommendation. The larger AI and education foundation article develops this relationship between established knowledge and new investigation. Here the focus is on making one part of that relationship observable.

Contents · Previous chapter · Next chapter

A fictional school club needs single-page, one-sided flyers for an event. Plan A charges an $8 setup fee plus $0.12 per flyer. Plan B charges a $20 setup fee plus $0.06 per flyer. For the first version, both plans can supply the required quantity, quality is equivalent and no other charge applies. These are invented classroom prices, not quotations from real businesses. State the conditions so the learner knows what the model includes.

Ask the student to predict which plan will be cheaper for a small order and which may become cheaper for a larger one. A useful first explanation might be: A begins with a smaller fixed charge, while B adds less for every extra flyer. The learner does not need the exact crossover quantity yet. That initial explanation gives the teacher a way to distinguish understanding of the relationship from a guess based on one attractive number.

The task is to make a recommendation that depends on the order quantity, rather than name a plan that is always best. The student should produce checked examples, locate where the plans cost the same and explain which side favours each plan. A calculator can be permitted if calculation speed is not the capability being assessed. State that decision before the activity begins.

Contents · Previous chapter · Next chapter

For 100 flyers, Plan A costs 8 + 0.12 × 100 = $20. Plan B costs 20 + 0.06 × 100 = $26. A is cheaper by $6. For 300 flyers, A costs 8 + 36 = $44, while B costs 20 + 18 = $38. B is cheaper by $6. These examples show that the initial recommendation cannot be carried unchanged to every quantity.

To find the equal-cost point, set the expressions equal: 8 + 0.12n = 20 + 0.06n. Subtracting 0.06n and then 8 gives 0.06n = 12, so n = 200. At 200 flyers, both totals are $32. The $12 extra setup fee for B is exactly offset by saving $0.06 on each of 200 flyers. This interpretation connects the algebra to the original situation.

For positive whole-number quantities below 200, A is cheaper. Above 200, B is cheaper; this means B first becomes strictly cheaper at 201 flyers. At the equal-cost point, neither has a price advantage. If some other relevant difference exists, such as delivery time, that information would need to be supplied before choosing on that basis. Students should not invent a benefit to force a unique answer.

The result applies to the conditions stated. It assumes that the per-flyer charges remain constant and that both plans can supply the order. It does not prove that a lower unit price always gives a lower total, or that every real pricing system is linear. The learner should be able to name those boundaries. Knowing when a result applies is part of the understanding that later tasks will examine.

Contents · Previous chapter · Next chapter

After the first attempt, AI can provide another representation or a challenge. The student might ask it to explain the equal-cost point using a table, identify a quantity that tests the recommendation or suggest an assumption worth checking. The learner then compares the response with the stated plans. A request for an explanation is useful only if the student examines what the explanation says.

One possible prompt is: “Use these two fictional plans. Ask me a question that will test whether I understand the fixed fee and the rate. If you suggest a calculation, explain what it checks. Do not add fees or conditions that are not in the task.” This is a proposed interaction pattern, not a guarantee of correct behaviour. The teacher should inspect the output and be ready to provide a checked example.

Require a short record of one consequential decision: what AI suggested, how it was checked and what the student accepted, adapted or rejected. If the model is correct, the student can show why. If it forgets the setup fee, the student can identify the omission. There is no need to engineer a false answer or reward disagreement for its own sake. Judgement includes justified agreement.

The companion guide to AI-powered assignments describes the broader inquiry process. This task makes one additional demand: after the supported work, the student must use the relationship in a changed case. The product and the later check therefore provide different evidence.

Contents · Previous chapter · Next chapter

Begin with a variation that changes one feature. Return to the original plans, then add a $6 delivery fee to Plan B only. Its new expression is B(n) = 26 + 0.06n. Plan A stays at 8 + 0.12n. Ask the learner to predict the direction of the change before calculating. B now needs a larger order before its lower rate compensates for the higher initial charge.

The revised equality is 8 + 0.12n = 26 + 0.06n, giving 0.06n = 18 and n = 300. At 300 flyers, both plans cost $44. At 250 flyers, A costs $38 and B costs $41, so A is cheaper. A learner who repeats the earlier threshold of 200 has not adjusted the model to the changed condition, even if the earlier solution was correct.

Ask what stayed the same. The per-flyer saving remains $0.06, but the difference in fixed charges increased from $12 to $18. This gives the student a way to explain the result without starting from a memorised equation. The goal is to relate the new information to the part of the model it changes.

Record whether the learner noticed the new fee independently or only after a prompt. Both responses can be useful during teaching, but they are different observations. A hint such as “check the fixed charges” can help the lesson continue. It should not later be described as completely unprompted transfer.

Contents · Previous chapter · Next chapter

Now use a separate variation: restore the original prices with no delivery fee, but state that Plan B accepts orders of at most 250 flyers. The club needs 300 flyers in one order. The old arithmetic still produces $38 for B, but that quote is not an available option under the new condition. A can supply the order and costs $44. The capacity limit changes the feasible choices.

This variation tests a different capability from recalculating a fee. The student must inspect whether the model’s recommendation can actually be used. A plan can be cheaper on paper and unavailable in the situation. Ask the learner to highlight the sentence that changes the decision and explain why calculating more decimal places would not resolve the issue.

Another separate variation leaves both plans available but says the final quantity may be anywhere from 180 to 240 flyers. At 180, A costs $29.60 and B costs $30.80. At 240, A costs $36.80 and B costs $34.40. No plan is cheapest at every quantity in that interval. The learner should ask whether the quantity can be confirmed or what decision criterion the club wants to use.

A defensible answer can therefore include a request for information. The task has changed from calculating one known quantity to choosing under uncertainty. AI may generate a confident recommendation, but the student should notice what would justify it. This is a useful form of transfer: recognising that a previously sufficient procedure no longer answers the whole question.

Contents · Previous chapter · Next chapter

A further task uses a fictional simulation of two water tanks. Tank X begins with 18 litres and gains 0.5 litres per minute. Tank Y begins with 30 litres and gains 0.2 litres per minute. For the period being considered, there is no loss or overflow and the rates stay constant. Ask when their volumes will be equal. This is a mathematical model, not an instruction to operate equipment.

The learner can write X(t) = 18 + 0.5t and Y(t) = 30 + 0.2t, where t is time in minutes. Equality gives 0.3t = 12, so t = 40 minutes. Both tanks then contain 38 litres. X starts with less but increases faster. The relationship resembles the printing problem: an initial difference is overcome by a difference in rates.

Ask students to explain the connection before showing them the equations. Can they identify starting amount and rate in the new setting? Can they explain why the answer is time rather than a number of flyers? Can they recognise that “cheaper” no longer describes the comparison? These questions examine whether the mathematical relationship survives a change in meaning and units.

Success here is evidence for transfer to this particular model. The algebra is still linear and the relevant quantities are explicitly supplied. It would be too strong to claim that the learner can now solve any science problem or apply the principle without cues in everyday life. A useful report describes the achievement precisely and identifies a sensible next challenge.

Contents · Previous chapter · Next chapter

An immediate explanation can show what a student understands at that moment. A later check asks whether the idea remains available after the original lesson and conversation are no longer fresh. These are different questions. A delayed task can also introduce a new context, but then two things have changed: time and situation. Record both so the result can be interpreted carefully.

Andrew Butler’s 2010 research compared repeated testing with repeated studying across four experiments using prose materials. A final test one week later included repeated or new inferential questions, and the testing conditions produced stronger retention and transfer in those experiments. The result supports examining active retrieval and delayed use. It does not establish a universal schedule or validate this article’s AI-supported pricing lesson.

A teacher might check a new price pair in the next lesson and revisit a rate comparison one or two weeks later. Those timings are examples for planning, not an experimentally established prescription for every learner. The later task should be manageable and should reveal the central idea. If it becomes much harder in several unrelated ways, an incorrect answer will be difficult to interpret.

After the learner attempts the check, provide feedback and further teaching. Do not preserve an error merely to keep the assessment conditions untouched. The educational purpose is to find out what needs attention and help it improve. A later check can become the starting point for the next learning cycle.

Contents · Previous chapter · Next chapter

“Without help” is an incomplete description of an assessment condition. Does it mean no generated answer, no calculator, no reference sheet, no read-aloud support or no conversation? Different restrictions test different capabilities. A task intended to assess model choice may permit a calculator. A task intended to assess arithmetic fluency may specify other conditions. The teacher should define the intended capability before choosing the restrictions.

Accessibility support should remain appropriate to the learner and purpose. Reading the question aloud, enlarging text, using an agreed communication method or allowing a normal access arrangement does not automatically supply the mathematical decision. Removing those supports may change the question from “Can the student reason with rates?” to “Can the student access this particular presentation?” The record should distinguish access support from a hint about the solution.

The same care applies to explanation. A student can explain through writing, a diagram, a calculation with annotations or a conversation, depending on what the task is designed to assess. A hesitant spoken answer is not proof of weak understanding or improper authorship. If evidence is unclear, use another suitable opportunity to examine the reasoning.

Students should know the conditions in advance. During the learning phase, help can be generous and explicit. During a diagnostic check, the teacher can first observe the attempt, then add a recorded cue if needed. Both stages can inform instruction as long as the account of performance is honest.

Contents · Previous chapter · Next chapter

A useful record can be short. It should name the capability, the change in task, the assistance available, the observed reasoning and the next step. It need not preserve an entire chatbot history. The example below illustrates one possible learner’s responses; it is invented and does not describe a real student. Its purpose is to show how different observations can lead to different teaching decisions.

Notice that the record does not reduce the learner to “can transfer” or “cannot transfer.” They handled a capacity limit but needed help to revise a fixed charge. They represented a later problem correctly but made an arithmetic error. These distinctions matter because they call for different responses. Repeating the whole lesson may be unnecessary; a focused comparison may be more useful.

Teachers can add a later entry after feedback. What changed? Did the learner independently recognise a new fixed fee in another example? The record then tracks a capability developing over time. It remains a local teaching record, not a validated scale for comparing every student or school.

Contents · Previous chapter · Next chapter

When a student struggles with the changed problem, first inspect the type of difficulty. They may have forgotten a relevant fact, misunderstood a word, selected the wrong relationship, misread a condition or made an arithmetic error after choosing a sound method. The same incorrect final number can result from each of these. Ask for enough reasoning to distinguish them, then choose a teaching response that addresses the observed gap.

Compare the demands with the original task. If the new example adds unfamiliar vocabulary, complex units and several extra constraints at once, it may be testing much more than the intended transfer. That can be appropriate in an advanced investigation, but it is less useful as the first diagnostic. A carefully chosen small change helps the teacher locate the difficulty.

A hint can reveal something too. If the student succeeds after being reminded to identify the fixed charge, the concept may be available but not selected spontaneously in this situation. That differs from being unable to explain the charge even after it is identified. The next practice might focus on recognising the relevant structure across examples, or on rebuilding the underlying concept. Avoid converting this observation into a permanent label about the learner.

One successful transfer also needs restraint. It may depend on a helpful cue or familiarity with the story. Repeat observation across a few suitable tasks can produce a more dependable picture. The goal is enough evidence to teach well, not an endless series of checks that displaces the learning itself.

Contents · Previous chapter · Next chapter

A teacher can spread the investigation across ordinary lessons. First, teach or revisit fixed charges and rates, and collect a short prediction about the original plans. Next, solve and explain the comparison using a table, calculation or graph. Permit a bounded AI interaction after the initial attempt. Finish with one changed condition that each learner explains individually. This gives the teacher a small amount of evidence without turning the task into a lengthy project.

In a later lesson, introduce either the delivery fee or the capacity restriction. Make clear that these are separate variations rather than a growing list of conditions carried forward silently. Students compare what changed and what stayed the same. A small group can discuss the alternatives, but each learner should produce one short explanation if individual understanding is the purpose of the check.

A subsequent lesson can use the tank model or another carefully matched rate problem. A further revisit can examine retention. The exact order and spacing should fit the class, topic and available time. This is a sample teaching sequence, not a tested intervention. A teacher can shorten it if the central distinction is already secure or add direct instruction when the baseline shows a gap.

Plan the review before the activity. Decide what would count as successful model choice, which supports are available and which misconception would prompt a return to foundations. The guide on teaching discovery with AI gives a broader approach to piloting lessons and reviewing workload. Here the important decision is which limited change will reveal the capability the teacher wants to see.

Contents · Previous chapter · Next chapter

A parent can help by asking a child to explain one important decision rather than requesting a performance on the whole assignment. Why did the plan change when the order became larger? Which sentence made the old answer unsuitable? What would you need to know before choosing? These questions keep attention on the child’s reasoning. They do not require the parent to become an expert in every tool or reproduce the teacher’s assessment.

Choose a calm moment and allow the child to use a representation that helps. They might draw the fixed fee as a starting amount and show repeated additions for each flyer. If the explanation is unclear, ask which part they would like to revisit. The conversation should make a learning need visible without implying that needing help is a failure. Follow the school’s instructions where the task has specified rules for assistance.

Parents can also help keep claims proportionate. Completing one comparison does not mean a child is now ready to make every purchasing decision independently. Conversely, struggling with one new context does not erase what they learned. Look for a specific next step: interpreting a rate, spotting an additional condition or explaining how a conclusion changed. A small clear improvement is more useful than a broad judgement about intelligence.

Contents · Previous chapter · Next chapter

Before using AI, write what you think the problem is asking and name the quantities that matter. After receiving an explanation, try to point to the part that changed your understanding. If the system says one plan is best, ask under which conditions. Check a small example and a larger one. A result that works at only one quantity may not support the general claim you were about to make.

Then close or set aside the generated solution and try a fresh version under the conditions your teacher permits. Explain why your method fits. If you need a hint, record it and use it to identify what to practise. This is a learning routine, not a contest to prove you never need support. The useful question is whether you can gradually take responsibility for more of the reasoning.

You can also create a test case for someone else. Change one condition and write the answer with its explanation before sharing it. Check that the question has enough information and that your solution matches the revised conditions. Creating an ambiguous question by accident is an opportunity to improve the task. AI can propose variations, but you remain responsible for checking whether they test the idea you intended.

Contents · Previous chapter · Next chapter

A classroom sequence provides evidence about the learners and tasks examined. It can show that a student revised a model, recognised a constraint or retained a principle at a later check. It cannot, by itself, establish that AI caused the improvement. Teaching, feedback, practice, peer discussion and prior knowledge may all contribute. If the aim is to evaluate AI’s added effect, a more careful comparison is needed.

Even a comparison between classes needs attention to topic, starting knowledge, teaching time and assessment conditions. A small local pilot can guide a department without being presented as a general research result. Report what happened, what support was provided and what remains uncertain. This makes the evidence useful to the next teacher who adapts the lesson.

Review practicality as well. Did the checks expose useful misconceptions? Could all students access the task? Did the AI interaction generate more checking than the lesson could support? Could a simpler shared output packet have served the same purpose? These questions help keep the design focused on learning. The number of tools introduced is not the outcome being sought.

For later educational and professional stages, the same distinctions remain useful. The article on building expertise alongside AI applies them to supervised practice and workplace decisions. The settings differ, but both require an honest account of what a person can understand, check and adapt.

Contents · Previous chapter · Next chapter

The aim of this branch is to help people use AI to extend their capabilities while developing the knowledge needed to direct that assistance. Transfer gives that ambition a practical test. Can the learner recognise the idea in a new representation? Can they change the model when a fee changes? Can they reject a recommendation that violates a constraint? Can they return later and explain the relationship again?

Each answer should be read at the scale the evidence supports. Understanding a printing comparison is a modest achievement, but it contains important habits: identify quantities, state assumptions, test a result and revise a conclusion. When these habits are taught across appropriate tasks and connected to strong subject knowledge, educators have more concrete evidence to examine than a polished output alone.

The education and civilisation article asks how changing human capability might reshape institutions. That larger proposition begins with specific learning that people can carry beyond its original setting. AI can widen the examples available, challenge a first explanation and support investigation. The educational work is to help the learner make those possibilities usable, explainable and open to correction.

Contents · Previous chapter · Continue the series