AI and Collaborative Learning | How Students Build Shared Understanding

A group can divide a presentation into separate slides and assemble an impressive result while its members understand only their own sections. That may complete a production task. A learning task needs an additional connection: each student should understand how the parts support, qualify or challenge the group’s answer. The teacher must decide which relationships everyone needs to explain.

Suppose the question is whether a club should use reusable containers. One student may calculate costs, another examine waste quantities and another consider the work involved. Shared understanding means that the cost specialist can explain why a waste claim is limited, while the waste specialist can explain which cost the group initially overlooked. Nobody needs to become equally expert in every detail, but the central argument must travel between members.

AI makes the distinction more important because it can produce a coherent-looking final answer very quickly. The class needs opportunities to inspect how that coherence was constructed. Which assumption connects the calculation to the recommendation? Which sentence sounds stronger than the evidence? Where do two reasonable priorities lead to different choices?

This develops the wider argument in AI and the Paradigm Shift in Education. Learners use an expanding set of tools to investigate further, while strengthening the judgement needed to understand and direct the work.

Contents · Next chapter

In Smith and colleagues’ 2009 Science study, students in an undergraduate genetics course answered concept questions, discussed them and then answered a similar question individually. The findings supported improved understanding through peer discussion, including groups in which nobody initially selected the correct answer. The study concerns that instructional setting and immediate conceptual questions. It does not establish that every group activity works, that the effect lasts indefinitely, or that adding generative AI improves collaboration.

Chi and Wylie’s 2014 ICAP framework distinguishes forms of engagement and examines how learners generate and exchange ideas. The authors also acknowledge that a learner can process material deeply while appearing quiet. The framework is useful for designing opportunities to explain and respond. It should not become a personality label or a rule that the most vocal student has learned the most.

These sources suggest questions for a classroom design: are students explaining reasoning, responding to one another and showing understanding on a subsequent task? The additional AI question remains open. Does the tool help the group investigate an important alternative, or does it make everyone accept the same unexamined answer? The lesson should create evidence that helps the teacher distinguish those possibilities.

Contents · Previous chapter · Next chapter

Temporary roles can make a discussion manageable. For a group of three, one learner might trace calculations, another examine the evidence behind claims and another test what changes under a new condition. These are responsibilities for a phase of work. They should not become fixed identities such as the mathematical student, the creative student or the quiet recorder.

Everyone begins with the same short information pack and writes one observation or question. The roles then help the group examine different parts carefully. At the next phase, students exchange responsibilities. The learner who first checked the calculations must now explain a claim boundary, while another learner checks a calculation already discussed. This makes the movement of understanding part of the task.

Operating the AI interface is also a temporary responsibility. The person typing should read the proposed request to the group or make it visible before sending it. Other members can ask for a missing condition to be included. The response belongs to the discussion; it does not give the person holding the device final authority.

A teacher can use pairs, larger groups or individual preparation followed by a partner exchange. The appropriate size depends on the task and available support. What matters is that responsibility stays visible and each student has a workable opportunity to contribute, question and explain.

Contents · Previous chapter · Next chapter

The worked investigation begins with a fictional reading club planning five lunch sessions for 30 participants each. The class must recommend a container arrangement within a budget of $110. Its answer must consider expenditure, discarded container material and the time needed for cleaning. Every quantity is invented for teaching; the example is not an environmental assessment of real products or a recommendation for a particular school.

The teacher supplies all relevant assumptions in one pack. Single-use containers cost $0.40 each and have a mass of 12 grams. Reusable containers cost $3 each and have a mass of 80 grams. The club initially buys 30 reusable containers. In this scenario, two break after the second session and must be replaced before the third. Every serving requires a container, and reusable containers are washed after each use.

The pack also gives a wash cost of $0.08 per use, assumed to include the specified washing consumables and utilities. Washing takes 30 minutes per session; disposing of the single-use containers and cleaning that area takes five minutes. The club has up to 35 minutes available after each session. Other work is held equal in this simplified comparison.

The design follows AI-Powered Assignments: give students enough information to begin reasoning, then use AI to broaden the investigation while preserving a clear evidence boundary.

Contents · Previous chapter · Next chapter

Before deciding, the group checks what each quantity describes. There are five sessions of 30 people, so the plan involves 150 servings. The single-use option needs 150 containers. The reusable option starts with 30, then purchases two replacements. It does not require buying a new set of 30 for every session. Confusing the number of containers with the number of uses would change the comparison substantially.

The reusable option costs $48 more under these assumptions and leaves $2 of the stated budget. Its cleaning takes 125 additional minutes across five sessions. Both options fit the given time allowance per session. These are consequences of the supplied model, not observations from a trial.

Each learner should be able to explain at least one quantity before the group moves on. A calculator or accessible calculation support can be used where appropriate; the learning objective includes knowing what to calculate and what the answer means.

Contents · Previous chapter · Next chapter

Students now prepare a provisional recommendation. One may prefer single-use containers because the plan spends less. Another may prioritise the smaller mass of discarded containers. A third may worry about the cleaning workload or the small budget margin. These are differences in priorities, provided everyone has interpreted the same quantities correctly.

Ask each learner to complete three sentences: our strongest reason is this; our decision depends on this assumption; and we would reconsider if this changed. The sentences make the argument available for discussion. They also reveal whether a student has confused a fact supplied by the pack with a judgement about what matters most.

The group does not need to agree immediately. It can write two defensible options with their trade-offs. If the brief requires one recommendation, the group should state the criterion used to choose. For example, it may prioritise reducing discarded container mass while staying within the budget and time constraints. Another group may justify reserving more of the budget for uncertain expenses. The teacher assesses the reasoning against the task, rather than treating one preference as the only evidence of learning.

This is a useful moment to connect the investigation to AI and the Paradigm Shift in Economics, which asks learners to examine opportunity cost, assumptions and the distribution of effort.

Contents · Previous chapter · Next chapter

The group can now ask AI to compare its provisional arguments. A useful request supplies the fictional data, identifies the group’s criteria and asks for missing assumptions or possible errors. Students might ask for a table separating arithmetic, factual claims within the pack and value judgements. The purpose is to create material the group can inspect.

Avoid letting the response silently redefine the question. If the class is comparing five sessions, a model should not assume the containers will be used for several years unless it labels that as a separate scenario. If the budget is fixed, an imagined sponsor does not make an over-budget plan feasible. The group can investigate those possibilities, but it must recognise that they change the original brief.

A suggested request is: “Use only the quantities in this fictional pack. Identify a calculation we should check, an assumption that could change our decision and a claim the pack cannot support. Explain your reasoning without selecting our priorities for us.” Students can adapt the wording. A prompt is a request for useful assistance, not a guarantee of a reliable answer.

The teacher may instead give every group a common AI response or a prepared candidate analysis. This makes the same comparison available when devices, connectivity or approved accounts differ. The central intellectual task is deciding what to do with the proposed reasoning.

Contents · Previous chapter · Next chapter

The teacher can supply this deliberately flawed candidate response, written for the lesson: “Choose reusable containers. They cost only $12, so the club saves 80%, and they reduce carbon emissions by about 91%.” This is an invented example of faulty reasoning, not a report of what a named AI system produced. Its confident wording creates several useful checking tasks.

The $12 figure covers washing only. It omits $90 for the original containers and $6 for replacements. The stated reusable total is $108, compared with $60 for single-use containers. The difference is $48. Relative to $60, that is 80% more stated expenditure, not an 80% saving. A plausible percentage can reverse the conclusion when its underlying total is wrong.

The carbon claim fails for a different reason. The pack contains no emissions data. Students can calculate that discarded container mass falls from 1,800 grams to 160 grams, a reduction of 1,640 grams or approximately 91.1%. That percentage concerns discarded container mass during the five sessions. It cannot be relabelled as a carbon-emissions reduction.

Nor should the group assume that all purchased reusable material has been discarded. Thirty containers, with a combined mass of 2,400 grams, remain at the end under the scenario. Total reusable container material purchased is 2,560 grams, including replacements. These different quantities describe purchases, remaining stock and discarded material. None alone establishes a complete environmental comparison.

The correction should be explained by someone who did not make the original calculation. This lets the teacher see whether the reasoning travelled across the group. The same distinction between a measurement and a broader conclusion appears in AI and the Paradigm Shift in Science.

Contents · Previous chapter · Next chapter

Some disagreements concern errors that the group can resolve directly. If someone says 32 containers at $3 cost $93, the multiplication can be checked. Other disagreements concern missing information. The group cannot resolve an emissions comparison by taking a vote when the relevant data are absent. A third kind concerns priorities, such as how much additional cleaning time is acceptable.

Naming the disagreement helps students choose a useful next action. For a calculation, reproduce the method. For a disputed source claim, inspect the source. For a missing quantity, state what evidence would be needed. For a value judgement, explain the criterion and its consequences. AI can suggest questions, but it cannot turn one category into another by producing an elegant summary.

A simple conversational move is to restate another person’s argument before responding. “You are saying that the reusable option fits the budget, but leaves only $2 for uncertainty. Is that the concern?” The other learner can correct the restatement. This makes the discussion more precise without requiring agreement.

Students can also record a remaining difference: “We agree on the totals. We differ on whether the waste reduction justifies the extra cleaning time.” Such a statement is often more informative than forced consensus. The goal is to understand the choice well enough to explain it, including the reason a thoughtful person might choose differently.

Contents · Previous chapter · Next chapter

After groups present their first recommendation, the teacher changes the fictional scenario. Instead of two containers breaking, five break and must all be replaced. The original budget and all other stated assumptions remain the same. Students first work out the consequences individually, then compare their reasoning.

The reusable option now requires 35 purchases at $3, costing $105. Washing still costs $12 because there are still 150 servings. The total is $117, exceeding the $110 budget by $7. Under the revised brief, a recommendation to buy that arrangement cannot satisfy the stated budget. The group needs to revise its choice or identify an authorised change to the conditions.

A student may propose borrowing containers or finding a cheaper supplier. Those are reasonable possibilities to investigate. They are not facts in the pack. The group should mark the proposal as conditional and state what information or permission it would require. AI should not be allowed to make the budget balance by inventing a source of money.

This is a close transfer check: the context remains familiar but an assumption changes. A later task should use a different context, such as shared equipment with an initial purchase cost and maintenance per use. The teacher then checks whether students recognise the same structure without relying on the container example’s numbers or wording.

Contents · Previous chapter · Next chapter

Role rotation works best when it changes the reasoning a learner practises. Simply moving the keyboard does not ensure that responsibility has changed. A student who originally calculated expenditure might next explain the difference between discarded mass and emissions. Another who focused on claims might check the revised budget. The third might explain why a proposed solution changes the brief.

Use a short handover. The outgoing role-holder explains the main finding and the most important unresolved question. The next learner restates it, asks for clarification and then applies it to a new part of the task. This makes an exchange of understanding visible without requiring every student to repeat the whole investigation.

Students can produce a shared decision record with four entries: the original proposal, the evidence that challenged it, the revised conclusion and the condition that could change it again. A few well-explained decisions are sufficient. The record should help the group remember its reasoning; it need not become a transcript of every interaction.

Teachers should avoid rewarding a learner simply for being the first person to find an error. Credit can also go to the student who asks a clarifying question, develops a counterexample, explains the correction to a peer or notices that an earlier conclusion no longer applies. These contributions support the group’s ability to reason together.

Contents · Previous chapter · Next chapter

A quiet learner may be following a difficult argument, preparing a written response or waiting for enough time to formulate a question. A fluent speaker may be making a valuable contribution, repeating an answer or thinking aloud towards an unfinished idea. The amount of speech does not settle what either student understands.

Teacher observation should therefore focus on specific evidence. Can the learner identify the assumption behind a calculation? Can they explain a correction in their own terms? Can they respond to a changed condition? A written note, annotated table, short private explanation or accessible alternative can provide evidence that public discussion alone misses.

If a student contributes little visible work, begin with an invitation that makes the next step clear. Ask them to choose a claim to check, annotate where a quantity came from or explain one part to a partner. The teacher can then see whether the difficulty concerns the material, the instructions, the available time or the way the group is organised. Observation should guide support rather than attach a label to the student.

Likewise, a student taking most of the speaking time may need a different role for the next phase. Ask them to summarise someone else’s argument and invite a correction, or to record questions while another member explains. The aim is a more balanced learning opportunity. Public blame or a competition over who talked most would distract from the problem the class is trying to understand.

Contents · Previous chapter · Next chapter

The group recommendation can show whether the combined work is accurate, clear and appropriately qualified. It cannot establish that every member understands it. An individual explanation can reveal a learner’s reasoning, but it does not by itself describe how effectively the group exchanged ideas. A useful assessment keeps these questions visible.

Tell students what will be assessed before they begin. They should know that they may disagree with an AI suggestion, revise an earlier answer and acknowledge missing evidence. A careful correction should count as intellectual progress when the learner explains why it was necessary.

The teacher can use a short individual question, rather than a lengthy second assignment. “Which part of the $108 changes when another container breaks, and why?” checks a meaningful relationship. Appropriate accessibility support should remain available. The aim is to understand learning, not to conduct authorship detection or reconstruct every keystroke. From the Centre to the Edge develops the broader teaching approach.

In that individual check, one additional replacement adds $3, making the total $111. Washing expenditure remains $12 because the number of uses has not changed. A learner who identifies both relationships has shown more than recall of the original total. If a student changes washing expenditure as well, ask which quantity they believe increased. Their explanation helps the teacher identify the next teaching step. The group can then compare the different methods after individual responses have been recorded, allowing a useful correction without concealing the original misunderstanding.

Contents · Previous chapter · Next chapter

A group should not depend on one student owning the best device or paying for a particular service. The teacher can provide a shared approved tool, conduct a common demonstration or distribute the same candidate responses for analysis. The learning objective remains the ability to compare, check and explain, whichever access route is appropriate.

Language support can help learners enter the discussion. A vocabulary list might distinguish expenditure, replacement, discarded material, assumption and criterion. A student can prepare an idea in a familiar language and then work towards a shared explanation with suitable support. Any translated technical statement should be checked against the original quantities and meaning.

UNESCO’s 2024 AI Competency Framework for Students emphasises human agency, critical judgement and inclusive participation across its learning dimensions. It supplies educational aims, not experimental proof that the group design in this article improves attainment. Those aims are useful when deciding who can participate and what students should retain control over.

Use the school’s approved arrangements and the relevant tool’s conditions. The fictional pack can be entered without students’ personal details or private accounts of classmates. The group does not need to upload names, recordings of peers or private school information to compare container options. Keeping the task bounded also makes it easier for everyone to see what evidence the response could reasonably use.

Contents · Previous chapter · Next chapter

A group can become stuck because members lack a needed concept, disagree about an assumption or do not know how to move from separate ideas to a decision. The teacher’s response should match the difficulty. A short explanation of fixed and per-use costs may be more useful than another round of open discussion. A shared table may help when learners are talking about different versions of the data.

Ask the group to state its current obstacle in one sentence. “We agree on the cost, but do not know whether to prioritise cleaning time or discarded mass” requires a decision criterion. “We have two different totals” requires checking the method. “We do not know what carbon emissions mean here” calls for acknowledging a missing measurement and, if appropriate, further teaching.

AI can be asked to propose questions that clarify the obstacle. Students should choose one useful question and explain why it helps. Generating a longer list of possibilities is not always progress; sometimes the group needs to settle a small definition before expanding its investigation.

When a disagreement becomes personal, pause the decision and return to the claim. Students can point to the number, source or assumption under discussion. A teacher may use a brief individual writing period before resuming the exchange. The purpose is to restore a workable conversation and preserve each learner’s opportunity to understand, contribute and change their mind.

Contents · Previous chapter · Next chapter

Parents can help by asking a child to describe the group’s reasoning in ordinary language. “What did your group decide?” is a useful start. Follow it with a question about the evidence: “Which number mattered most?” or “What changed after someone challenged the first answer?” These questions invite explanation without requiring the parent to know the entire project.

It is also useful to ask what another member contributed. The child might describe a calculation, a missing condition or a question that changed the discussion. This keeps attention on learning from peers, rather than only on whether the final product looked impressive or who completed the largest share of it.

If the child cannot explain part of the answer, identify the unclear step calmly. They may need to reread the pack, ask a teammate or bring a specific question to the teacher. Producing a better answer for the child would remove an opportunity to resolve that gap. A parent can help the child formulate the question and decide what evidence would make the explanation clearer.

Concerns about group participation should also remain concrete. A child may report that the device was unavailable to them or that the group decided before they could contribute. The useful next step is to describe what happened and ask for an adjustment to the process. A fair learning arrangement gives students meaningful opportunities to practise, while recognising that contributions can take different forms.

Contents · Previous chapter · Next chapter

The container investigation combines arithmetic, interpretation and decision-making. The same structure can be adapted to other subjects, but the evidence and standards must change with the discipline. A mathematical claim may need a derivation or counterexample. An explanation of a text needs attention to its actual wording. A scientific proposal needs a question that the available observations can address.

In English, students might compare two arguments about a familiar school activity. Each learner identifies a claim and its support before the group asks AI to offer a counterargument. Students must decide whether that counterargument addresses the original reasoning or introduces an unsupported premise. The individual follow-up uses a different argument and asks the learner to explain the relationship between claim and evidence.

In mathematics, a group can compare two methods for the same problem, use AI to suggest a third, and determine where each method is valid. Rotating roles can include explaining a step, checking a condition and finding a case where an apparent shortcut fails. A correct final answer is useful, while the method and its conditions remain part of the learning.

Younger learners may need fewer quantities, visual representations and more teacher-led comparison. Older learners can handle competing assumptions and incomplete evidence. The task should remain demanding in the intended subject without making unfamiliar vocabulary, interface operation or public performance the main obstacle to showing understanding.

Contents · Previous chapter · Next chapter

Teachers can begin with a bounded sequence rather than changing every group assignment. One lesson might establish the question, check the common data and compare initial arguments. A second could introduce the candidate AI response, rotate roles, revise the recommendation and include a short individual check. The pacing should match the class and the amount of prerequisite teaching required.

Record a small amount of useful evidence. Which misconception appeared? Which peer explanation helped another learner revise it? Could students handle the changed breakage assumption? Did access arrangements allow everyone to take part? How much teacher preparation and checking did the activity require? These questions connect the design to both learning and practical classroom conditions.

Keep claims about effectiveness proportionate. A class may improve after the activity because of instruction, peer discussion, repeated practice, AI assistance or their combination. A single successful lesson cannot isolate each contribution. A more formal evaluation would need suitable comparisons, consistent outcomes and attention to differences between tasks and learners.

The teacher can still make a useful local decision. Keep a prompt that helped students expose assumptions, simplify a role that caused confusion, replace a misleading example or provide more time for individual reasoning. A design becomes stronger when the teacher can identify what changed and why. The aim is a repeatable opportunity for better thinking, supported by evidence appropriate to the scale of the claim.

Contents · Previous chapter · Next chapter

AI can add possibilities to a discussion: another explanation, a challenging example, a reorganised table or a question the group had not considered. The educational value depends on how students use those possibilities together. They need to inspect claims, listen to alternative reasoning and decide what the available evidence can support.

The container case illustrates a modest but important form of progress. Students may begin with a confident recommendation, discover an omitted cost, correct a claim about emissions and revise their choice when the budget changes. The final recommendation matters. So does the ability of each member to explain how the group reached it and where its limits remain.

A classroom can make that ability visible through common information, temporary responsibilities, patient explanation and individual follow-up. Students then practise a form of collaboration that extends beyond sharing the workload. They learn how to build an answer with other people, use AI to broaden the investigation and retain the judgement needed to improve the answer when new evidence arrives.

Contents · Previous chapter · Continue the series