A student raises a hand. The teacher has just demonstrated how to solve a problem, and every step looks reasonable. “I understand,” the student says. Five minutes later, faced with a slightly different question, the student is stranded.
Nothing mysterious has happened. Following another person’s reasoning and generating a solution of your own are different intellectual activities. A polished explanation can produce recognition; a new problem demands decisions. Which information matters? What principle applies? Is the first approach wrong, or simply incomplete?
Education is organized to deliver answers. Learning, however, sometimes begins at the moment an answer is not yet available.
That is the tension at the heart of How the Brain Learns by Solving Problems, an English-language Learning Coach Editions title. Its premise invites a difficult question: if a teacher can spare a learner the discomfort of getting stuck, should they?
The answer is neither a romantic defense of failure nor an argument against explanation. It is an argument for designing moments when students must think before they are told what to think—and for making sure those moments lead somewhere.
The answer can end the task before it begins the learning
Imagine two mathematics lessons about proportional relationships.
In the first, the teacher presents the correct method, solves three examples and gives students ten near-identical questions. Students may move quickly because the method has already been selected for them. Their work can be fluent, accurate and useful; rehearsal matters. Yet it may reveal little about what would happen if the question changed its surface features.
In the second, students first encounter a carefully chosen problem: two stores offer different-sized packages at different prices. Which is the better value, and how can you justify your conclusion? Some draw tables. Others compare differences instead of ratios. A few make an intuitive guess. The teacher collects contrasting approaches before explaining the underlying structure.
The second class is not automatically better. It could be confusing, slow or badly managed. But it exposes something the first lesson may conceal: learners' existing models of the situation.
An error made during a thoughtful attempt is information. It reveals what the learner assumed, which quantities were compared and where the reasoning broke. The teacher can now explain not only how a method works, but why it solves a problem that students have already tried to solve.
This changes the meaning of an answer. Rather than arriving as a finished object to be copied, it arrives as a resolution to an intellectual question the learner has begun to own.
What “productive failure” actually means
The idea is closely associated with learning scientist Manu Kapur and the research program known as productive failure. Its counterintuitive move is to put certain problem-solving experiences before formal instruction, not simply after it.
The adjective matters. The point is not to abandon students, celebrate wrong answers or treat distress as evidence of growth. A well-designed initial problem is accessible enough to invite meaningful attempts, rich enough to elicit multiple approaches and followed by instruction that compares those approaches with the target concepts.
The most useful distinction is between failing to solve a task immediately and failing to learn from the task. Those are not the same outcome.
In a 2021 meta-analysis published in Review of Educational Research, Tanmay Sinha and Manu Kapur synthesized 53 studies with 166 comparisons of problem solving before instruction versus instruction before problem solving. Across the comparisons, the average result favored the problem-solving-first sequence by a moderate amount (Hedges' g = 0.36). Effects varied by design, age and learning target. The same review reported patterns favoring instruction-first approaches for some younger learners and domain-general skills.
Read the meta-analysis abstract.
That finding is important precisely because it is not a universal command to start every lesson with a puzzle. It suggests that sequence can matter, and that the quality of the problem, discussion and subsequent instruction determines much of the educational value.
Kapur's 2024 book Productive Failure: Unlocking Deeper Learning through the Science of Failing develops the broader design logic. It is a useful intellectual companion to the research, but the book's persuasive examples should not be confused with the strength or generality of any single experiment.
Explore the book's publisher record.
Difficulty is not a virtue in itself
One of education's most persistent misunderstandings is that a hard activity must produce deep learning. It might. It might also consume attention without building understanding.
Robert Bjork's research on desirable difficulties distinguishes immediate performance from learning that lasts. Spacing practice across time, retrieving information instead of always rereading it, and mixing related problem types can feel effortful even when they support later retention or transfer.
The word desirable does enormous work. A challenge becomes useful when it encourages the processes required for learning, at a level the learner can realistically manage. Confusion that cannot be resolved is not automatically productive; neither is an exercise that merely takes longer.
Explore UCLA's overview of desirable difficulties.
That qualification brings a second tradition into the conversation: cognitive load theory. Working memory has limits, especially when a task contains many unfamiliar elements. Novices may learn more effectively from carefully designed explanations and worked examples than from being required to discover a complex solution with little guidance.
Paul Kirschner, John Sweller and Richard Clark made this case forcefully in their influential paper, “Why Minimal Guidance During Instruction Does Not Work.” Their target was not thinking, curiosity or active engagement. It was the assumption that beginners can efficiently build complex knowledge without sufficient instructional support.
Read the paper's bibliographic record.
The two traditions are often turned into opposing educational camps: let them struggle versus show them how. But the practical question is better framed as: What should learners attempt independently, what should an expert model, and when should the balance change?
A 2026 study in Instructional Science examined task complexity and interleaved practice with 132 participants. Its results illuminate the interaction between challenging practice, cognitive engagement and performance, but a single study does not settle the best sequence for every subject or age group.
The credible educational position is not maximal difficulty. It is well-timed difficulty with a path to understanding.
A mistake becomes useful only after something happens to it
Consider a student who incorrectly assumes that doubling the sides of a square doubles its area.
The teacher can say, “No, the area increases fourfold,” and move on. That correction is accurate. Yet the student's underlying assumption—that length and area change at the same rate—may remain untouched.
Alternatively, the teacher can ask the student to draw a one-by-one square and then a two-by-two square, count the unit squares and explain the discrepancy. The drawing makes the error visible. The learner now has a reason to replace one mental relationship with another.
What made the experience productive was not the wrong answer. It was the sequence:
Attempt → comparison → explanation → revision → new attempt.
Omit comparison, and a misconception may survive. Omit explanation, and the learner may remember an isolated trick. Omit a new attempt, and it remains uncertain whether the corrected idea can be used independently.
This is also why “learning from mistakes” should not become a moral slogan. Feedback must carry information. A student needs to know which inference failed, how a better one works and what would count as improvement.
Research on productive failure itself points to design limitations. Studies examining cases where the method underperformed have highlighted the importance of fidelity to its core features. Simply giving a problem before a lecture is not the same as carefully designing a productive-failure lesson.
Read Sinha and Kapur's analysis of when productive failure fails.
Three kinds of struggle that should not be confused
A meaningful debate about problem-solving begins by separating three experiences that classrooms too often label alike.
1. Generative struggle
The learner has enough initial knowledge to attempt the task, but not a ready-made solution. Several plausible approaches can be compared. The teacher is prepared to consolidate the underlying concept.
This is where problem-solving-first designs may shine.
2. Practice struggle
The learner has encountered the concept and is now retrieving, selecting and applying it without all the original prompts. Errors reveal what still requires correction.
This is where varied practice, spacing and useful feedback matter. A student who cannot remember a procedure may need a refresher, not an entirely new discovery exercise.
3. Unproductive overload
The learner lacks essential vocabulary, background knowledge, an accessible task, time or the support needed even to begin. Repeated attempts produce neither a better hypothesis nor clearer understanding.
Here, increasing the challenge can make the situation worse. Model a step, teach a prerequisite, adjust the representation or reduce irrelevant complexity. For some students, accessibility accommodations and specialized instruction are necessary parts of sound teaching, not concessions that undermine rigor.
Distinguishing these experiences protects the idea of productive struggle from becoming a justification for neglect.
The social cost of saying “try harder”
Who gets to struggle productively?
A student who has a patient teacher, enough time, accessible materials and permission to ask questions may experience a difficult problem as an invitation. A classmate under acute time pressure, with missing prerequisites or a history of being publicly shamed for errors, may experience the same task very differently.
This is not an argument for lowering intellectual expectations. It is an argument against confusing equal exposure to difficulty with equal opportunity to learn.
Even the phrase “learn from failure” carries emotional and institutional baggage. When schools use mistakes as public rankings, children may learn to conceal uncertainty. When adults praise persistence without offering feedback, responsibility for inadequate teaching can quietly be transferred to the learner.
A humane classroom makes first attempts low-stakes, evaluates reasoning without humiliating the person, and treats revision as part of the task. It also respects the student's right to receive a clear explanation.
The educator's job is not to manufacture frustration. It is to ensure that cognitive effort has educational direction.
What a well-designed problem-first lesson could look like
There is no universal five-step recipe, but the following sequence is a practical interpretation of the research rather than a validated method in its own right.
Start with a question worth answering. Present a problem with a genuine choice to make: compare two claims, predict an outcome, explain an observation or decide which strategy fits. A decorative riddle unrelated to the concept is unlikely to help.
Invite a bounded first attempt. Give individuals or pairs a few minutes to propose a method. Let them write, sketch or test. Establish that an initial attempt is not a graded verdict on ability.
Make the thinking visible. Collect two or three contrasting approaches. Ask what each assumes. The purpose is not to publicly display who failed; it is to surface ideas that the explanation will address.
Teach the missing structure. This is the indispensable consolidation phase. Name the principle, compare it with the initial methods, show a correct worked example when needed, and explain where the tempting errors went wrong.
Test transfer. Change the surface of the problem. Can the learner select the same principle without copying the original solution? Follow up later to check what remains.
A lesson like this can fit mathematics, data literacy, scientific reasoning or source evaluation, though the appropriate amount of guidance will vary. Teachers should adapt it to curricular goals rather than use difficulty as a performance of modern pedagogy.
And not every concept warrants this sequence. Some factual and procedural foundations are most efficiently introduced directly. The art of teaching lies partly in knowing which.
The AI question: who is actually solving the problem?
Generative AI introduces a new version of an old temptation. When a system can produce a correct answer instantly, the learner may bypass the uncertainty that would otherwise reveal a misunderstanding.
But an assistant is not necessarily an enemy of productive struggle. It can ask for a prediction before supplying an explanation; offer a hint instead of a complete solution; compare two approaches; or generate a new problem after a learner studies a worked answer.
What matters is the allocation of cognitive responsibility.
One productive question for an AI-supported study session is: What part of this solution can I explain and reproduce without the tool? If the student can only admire the output, the problem may have been solved while the intended learning remains untested.
Equally, withholding all assistance in the name of “independent thinking” can be a poor choice. When a learner is hopelessly stuck, a timely explanation may accomplish more than another ten minutes of guesses.
AI makes that judgment more urgent, not less.
The promise—and the limit—of a problem-centered education
How the Brain Learns by Solving Problems, from Learning Coach Editions, belongs to a broader conversation about learning as active model-building rather than the passive accumulation of answers. The book's central direction is valuable: meaningful problems can create reasons to predict, explain, revise and transfer knowledge.
Yet no single metaphor for the brain, no matter how compelling, should be mistaken for a universal law of classroom instruction. Intelligence is not simply manufactured by exposure to obstacles. Prior knowledge, expert teaching, practice, feedback, emotion, context and access all shape what a learner can gain from a problem.
The strongest defense of productive struggle therefore refuses two easy stories. It rejects the claim that answers alone are education; it also rejects the claim that answers should be withheld to make learning authentic.
A good explanation can open a door. A carefully chosen problem can give a learner a reason to walk through it.
Perhaps the better measure of a lesson is not how quickly students reach the right answer, nor how long they endure difficulty, but whether they emerge with a more powerful way of thinking.
The answer finishes one problem. The newly understood principle gives the learner a chance to solve the next.
Further reading and source notes
Sinha, T., & Kapur, M. (2021). When Problem Solving Followed by Instruction Works: Evidence for Productive Failure. Review of Educational Research, 91(5), 761–798. Meta-analysis; effects vary with context and implementation.
Kapur, M. (2024). Productive Failure: Unlocking Deeper Learning through the Science of Failing. Wiley. Book-length account of the theory and its applications.
Kirschner, P. A., Sweller, J., & Clark, R. E. (2006). Why Minimal Guidance During Instruction Does Not Work. Educational Psychologist, 41(2), 75–86. A central counterargument emphasizing novices' needs.
Bjork Learning and Forgetting Lab — Research. Background on desirable difficulties and the distinction between performance and retention.
Zeitlhofer, I., & Zumbach, J. (2026). Germane Processes Moderate the Effectiveness of Complex Desirable Difficulty Tasks. Instructional Science. Context-specific experimental evidence, not a universal prescription.
Sinha, T., & Kapur, M. When Productive Failure Fails. Research discussion of design fidelity and weaker results.
