Productive Friction: The Learning Struggle AI Should Not Remove

A professor assigns a case study. A student opens ChatGPT, pastes the prompt, and 15 seconds later has a fluent analysis. The student submits it, earns a B+, and moves on. The answer is correct. But the learning didn't happen.

This is the paradox generative AI forces educators to face. Not how to catch it, but how to preserve the cognitive effort that makes learning durable. The concept that captures this challenge is productive friction: the specific kind of struggle that produces learning, not frustration.

What productive friction means

The Ivey Business School report From Chalkboard to Chatbot defines high-quality learning as involving three distinct forms of cognitive work: struggle with concepts for understanding, repetition and correction for skills, and decisions under ambiguity for judgment (pp. 2-3). Each demands effort, but a different kind, aimed at a different outcome.

Productive friction is the encounter between resistance and engagement. The qualification matters, and the report's Note 9 makes it explicit: resistance without engagement is just difficulty; engagement without resistance is just busywork (p. 9). A confusing textbook with no scaffolding produces resistance but no learning. A well-designed multiple-choice quiz produces engagement but no durable struggle. Productive friction sits in the middle: effort that the learner can handle, directed at the specific cognitive work the objective requires.

This is not the same as "make it harder." Learning to calculate a net present value requires practising the calculation — not reading a more difficult textbook. Learning to negotiate a contract requires making and defending trade-off decisions — not memorising a longer list of negotiation tactics. The friction follows the objective.

How generative AI changes the learning task

Before generative AI, cognitive effort was a default condition of most academic work. You could not write an essay without thinking through the argument, because there was no tool to do it for you. AI changes that structural reality. As the Ivey report puts it, friction moves from a default condition into an opt-in choice (p. 3). The student now decides whether to do the thinking or delegate it.

This choice creates a tension between short-term performance and durable learning. Desirable difficulties research (Bjork & Bjork, 2011) shows that conditions that make learning harder in the moment — retrieval practice, spacing, interleaving — produce stronger long-term retention. AI removes those conditions by giving the learner the answer before retrieval happens. The student performs well on the assignment but hasn't built the mental model they would need in a setting without AI.

What some observers call metacognitive laziness is not a character flaw. It is a rational response to available tools. When a student can get a fluent answer in seconds, the incentive to monitor their own comprehension, identify gaps, and practise weak areas drops. Evidence in the Ivey report points to reduced persistence when AI is available (p. 3; references p. 9). But metacognitive laziness is a description of a behavioural pattern under specific conditions, not a diagnosis of individual students. Change the conditions — change the assessment — and the behaviour changes.

Four kinds of effort worth protecting

Not all cognitive effort is equal. Four kinds, each tied to a distinct learning objective, deserve deliberate protection in an AI-saturated environment.

Retrieving and explaining a concept. Understanding a framework means being able to recall it from memory and explain it in your own words. When AI retrieves and summarises on the learner's behalf, the storage and retrieval process that builds long-term memory never happens. Protect this by requiring unprompted recall: closed-book warm-ups, device-free discussions, or timed explanations before any tool access.

Deciding what information is needed. In professional settings, the hardest step is often figuring out what question to ask and what data to gather. Many assignments short-circuit this by providing all relevant material upfront. Protect it by staging information: give learners partial data, let them identify what is missing, and release additional details only after they request it with a rationale.

Rehearsing a difficult skill. Skills improve through repeated, effortful practice with corrective feedback. AI can skip straight to the polished output. For skill objectives, assessments must capture the process — the draft, the attempt, the revision — not just the final product.

Making and defending a judgment under ambiguity. This is the highest-stakes kind of effort. It means sitting with incomplete information, weighing conflicting perspectives, committing to a decision, and articulating the reasoning. AI can produce a reasonable judgment, but the learner hasn't wrestled with the trade-offs. Protect this through scenarios that require a decision before revealing outcomes, followed by reflection and revision.

Restore friction or introduce new friction?

When an assessment has lost its productive friction, educators have two toolboxes.

Restore friction by removing the option to outsource. Closed-book moments, device-free phases, and withheld answers force the learner to retrieve and think without AI. These are clean, low-tech interventions that work when the objective is retrieval or rehearsal.

Introduce friction by adding layers AI cannot easily flatten. Counterarguments that the student must address, staged information released only after an initial decision, consequences that follow from choices, and revision cycles that require the student to improve their own work. These create a structural environment where effort is necessary, not optional.

The two approaches coexist. A single course might include closed-book concept checks alongside a multi-stage simulation where AI is welcome — the tool does not replace the judgment call. The choice depends on the learning objective, not the instructor's philosophy about AI.

A friction audit for any assignment

Before designing or revising an assignment, run through four questions. They form a reusable audit that applies across disciplines and formats.

What should become easier? Administrative friction — formatting, citation management, basic summarisation, data lookup — is not productive. AI can reduce it, freeing cognitive capacity for harder work. Name the elements that genuinely do not need to be effortful.

What must remain effortful? Identify the specific cognitive work the learning objective demands. If the objective is to analyse a balance sheet, the calculation matters. If it is to recommend a course of action based on that analysis, the reasoning and trade-off judgment matter. Preserve effort on those.

What evidence shows the learner did the work? Process evidence — drafts, annotations, decision logs, recorded think-alouds, revision history — tells you more than a polished final product. Design the assignment to generate that evidence as a natural by-product of the work.

How can feedback improve the next attempt? Productive friction is not a single encounter. It is a cycle: effort, feedback, revision. The best assignments build iteration into their structure so that the learner's second attempt is informed by the first.

How structured simulations make effort purposeful

Structured simulations illustrate what productive friction looks like in practice. A well-designed simulation presents learners with incomplete information, introduces stakeholder perspectives with conflicting goals, creates trade-off decision points, attaches consequences to choices, and builds in debrief and feedback loops.

In the LiveCase platform, learners navigate unfolding scenarios through a chat interface, making decisions under partial information and seeing their consequences. The friction is deliberate: the learner must decide what question to ask next, which stakeholder to trust, which trade-off to accept. The effort maps to the learning objective — judgment under ambiguity — not to arbitrary difficulty.

But not every simulation automatically creates productive friction. A simulation that linearises choices into obvious paths or rewards surface-level answers adds busywork, not struggle. The quality depends on the learning objectives and the design choices that implement them. The same principle applies: resistance without engagement is just difficulty.

For a deeper look at how AI changes cognitive ownership of assignments, read Rebuilding Cognitive Ownership. And for the broader shift from passive content to active learning, What Does Experiential Learning Actually Mean? offers a grounding framework.

Download the one-page Productive Friction Audit. Apply it to one current assignment and see where the friction needs redesign. Then explore how the free LiveCase AI Authoring Studio lets you build structured simulations that preserve the right kind of cognitive effort — without writing code.

FAQ

What is productive friction in learning?

Productive friction is the specific kind of cognitive effort that produces learning: struggling with concepts, practising skills with correction, and making decisions under ambiguity. It requires both resistance and engagement, and it varies by learning objective.

Is productive friction the same as desirable difficulty?

Productive friction is a broader concept that includes desirable difficulties, which research shows improve long-term retention. Desirable difficulty is one mechanism within productive friction. All desirable difficulties create productive friction, but not all productive friction takes the form of a desirable difficulty.

What is metacognitive laziness?

Metacognitive laziness describes reduced self-monitoring and persistence when AI provides fluent answers quickly. It is a behavioural pattern under specific conditions, not a diagnosis of individual students. Change the assessment conditions and the behaviour changes.

How can AI support learning without doing the thinking?

AI is most supportive when it explains rather than solves, structures information rather than retrieves it, or generates options the learner must evaluate. The key design principle: AI should reduce administrative friction but never replace the retrieval, rehearsal, or judgment the learning objective requires.

Can simulations create productive struggle?

Yes, when designed for the right learning objectives. A simulation that requires learners to weigh incomplete information, make trade-offs, and face consequences can create productive friction. A simulation that linearises choices into obvious paths or rewards surface answers creates busywork, not struggle.