HSW-0241
Two weeks before an examination, a class discussion changes shape. Students who were quiet begin asking whether a topic will be tested, how two ideas differ, what happens if a condition changes, and which step matters most. It is tempting to say that pressure made them more curious.
That explanation is possible. It is also too quick.
A question is an observable action. Curiosity is a psychological state. The two can be related without being identical. Students ask questions because they want to know, but also because a deadline makes ignorance costly, because a task exposes a gap, because a peer’s answer creates uncertainty, or because the structure of an examination finally tells them which unknowns matter.
A 2026 npj Science of Learning study gives this distinction unusual scale. Farhan Ali, Yehong Yang and Jia-Hou Poh analysed a large naturalistic archive of student discussion on Reddit—about 55,000 users and roughly 570,000 posts and comments from 2019 to 2023. Question-posing rose sharply during recurring educational phases such as examination periods across discussion topics, whereas the diffuse uncertainty associated with COVID-19 showed comparatively little statistical effect on the overall volume of questions. The authors describe question-posing as a manifestation of curious behaviour, not a direct meter of private curiosity.
This article calls the broader studying problem question-posing under uncertainty: how a learner converts not-knowing into a useful question, and why uncertainty becomes productive only when it is structured enough to point somewhere.
The direct answer
Uncertainty does not automatically create useful learning. A useful question usually requires at least three things: the learner must notice a gap, the gap must feel relevant enough to pursue, and the learner must be able to express the gap in a form that can be answered or tested.
Structured educational uncertainty can help because it narrows the search. An approaching examination, a difficult problem, a disagreement between two explanations or a prediction that failed can reveal exactly what the learner does not yet know. Diffuse uncertainty can do the opposite: it can be large, emotionally salient and still provide no clear next question.
The studying lesson is therefore not “make students anxious so they ask more questions.” It is: make the unknown visible, bounded and actionable without manufacturing unnecessary threat.
The information gap is not enough
Suppose a student opens a chapter on electricity and says, “I don’t understand this.” That sentence contains uncertainty, but it is poorly resolved. The learner may not know whether the difficulty is vocabulary, circuit representation, current versus voltage, proportional reasoning, or how to read the question.
Now change the task. Ask the student to predict what happens to current when resistance doubles while voltage stays constant. The student predicts “current doubles,” checks the relation, and discovers the prediction is wrong. The uncertainty has acquired an address.
The better question becomes: “Why does increasing resistance reduce current if voltage is unchanged?” That question can drive explanation, retrieval, worked examples and a fresh test.
Productive uncertainty is not merely a feeling of confusion. It is uncertainty attached to a discriminable object.
What the 2026 naturalistic study found
Ali, Yang and Poh examined discussion on the SGExams Reddit community across several years. Their central comparison was between two forms of uncertainty. One was proximal and task-relevant: recurring educational phases such as examination periods. The other was distal and diffuse: the broader uncertainty associated with the COVID-19 pandemic.
Using time-series analyses, they found that the collective volume of question-posing increased markedly during educational phases across discussion topics. COVID-related uncertainty, by contrast, showed minimal statistical effects on overall question volume. At the shorter conversational timescale, changes in urgency and the trajectory of questioning were more limited and topic-dependent.
The distinction matters. “More uncertainty” was not the pattern. Uncertainty that was organised around a consequential, comprehensible educational problem was associated with more questioning.
The study is observational. It does not show that examination pressure caused each student to become more curious, and it cannot recover every poster’s motive from text. The authors’ question taxonomy identifies linguistic behaviour, not an internal psychological state. Those limits make the paper more useful, not less, because they prevent a common educational overreach: treating every visible question as direct evidence of curiosity.
Questions do at least four different jobs
A student question can seek missing information, test a hypothesis, reduce procedural uncertainty or open a possibility space. These jobs need different answers.
- Information question: “What does osmosis mean?”
- Discrimination question: “How can I tell diffusion from osmosis in a new example?”
- Mechanism question: “Why does water move this way when solute concentration differs?”
- Possibility question: “What would change if the membrane were permeable to the solute as well?”
The first may be answered with a definition. The second needs contrasting cases. The third needs causal explanation. The fourth may require reasoning beyond the taught example.
If every question receives the same response—usually another explanation—the learner misses the chance to learn how questions themselves control the study process.
Illustrative case: “Will this come out?”
The following case is illustrative, not research data.
Leonie is revising history. She asks her tutor, “Will this chapter come out?” The surface request is about examination probability. The tutor could answer yes or no and end the exchange.
Instead, the tutor asks: “What would you need to know if it did?” Leonie names three events but cannot explain why one caused the next. The original question was driven by uncertainty about relevance; the follow-up exposes uncertainty about causal structure.
They convert it into three study questions: What changed? Why did it change? What evidence would distinguish this explanation from an alternative?
The examination did not magically create curiosity. It created a boundary around the unknown. Good teaching used that boundary to reveal a learnable problem.
Question volume is not question quality
A learner can ask many questions and still outsource the difficult work. “Is this right?” after every line may function less like inquiry and more like continuous reassurance. “What is the answer?” can terminate search rather than deepen it.
That does not make simple questions bad. Novices often need factual and procedural questions. The diagnostic issue is whether the questioning gradually improves the learner’s ability to locate uncertainty independently.
A useful progression is:
- What do I not know?
- Which part of the task exposed that gap?
- What answer would change my next move?
- Can I test the question before asking someone?
- What new question appears after the answer?
Question quality improves when questions begin to discriminate between explanations rather than merely request information.
Why diffuse uncertainty can be cognitively expensive
Consider the difference between “I don’t know whether I can solve simultaneous equations” and “I don’t know what is going to happen to school, examinations, my family schedule and the wider world.” Both are uncertainties. Only the first immediately specifies a learning test.
Diffuse uncertainty can produce monitoring without progress. The learner repeatedly scans for information but lacks a criterion for when enough has been learned. In practical terms, the search space is too large.
That is one reason the 2026 study’s contrast is instructive. The natural experiment did not show a simple proportional relation between societal uncertainty and questioning. Educational phases were much more reliably associated with question volume than the pandemic period.
A study system should therefore shrink uncertainty into units that can be acted upon.
The uncertainty-to-question protocol
1. Name the object
Replace “I don’t get it” with the smallest identifiable object: a term, step, relation, assumption, evidence claim or decision.
2. State what you currently think
A question is more diagnostic when it contains a provisional model. “I think the denominator increases because the sample space increased. Is that the right reason?” exposes more than “How do I do this?”
3. Ask what evidence could change your mind
This converts information-seeking into hypothesis testing. It is especially useful in comprehension, science and humanities, where several interpretations can sound plausible.
4. Choose the answer source deliberately
Some questions should be answered by the textbook, some by a worked example, some by an experiment, some by a teacher, and some by attempting a fresh problem. The fastest source is not always the best learning source.
5. Close with a test
After the answer, remove it. Ask a fresh version. If the learner can now predict, explain or solve independently, the question has done educational work. If the learner can only repeat the answer, the uncertainty may have been covered rather than resolved.
Delayed and independent performance check
A good questioning routine should eventually survive without a tutor standing nearby.
After a topic has been taught, wait until the next session and give a fresh problem before reviewing notes. Ask the learner to write three things: what is known, what is uncertain, and the single question whose answer would most improve the next decision.
Then require an attempt before external help. The target is not whether the learner produces an elegant philosophical question. The target is whether the question locates the real bottleneck and changes what the learner does next.
Repeat with a different topic. Transfer matters because a student who can ask excellent questions only in one familiar chapter may have learned a local script rather than a general regulation skill.
For parents: do not answer the first version too quickly
When a child asks, “Is this correct?”, the fastest response is yes or no. A more diagnostic response is sometimes, “What are you uncertain about?”
This is not a trick for withholding help. The child should receive help when needed. The purpose is to expose whether the uncertainty concerns the answer, the method, a definition or confidence.
Parents can also separate emotional uncertainty from study uncertainty. “I’m scared about the exam” deserves emotional support. “I cannot tell when to use completing the square” deserves a discriminating set of examples. Conflating the two can produce either cold tutoring or endless reassurance.
For tutors and teachers: design question opportunities, not anxiety
The 2026 findings should not be turned into an intervention claim that pressure improves learning. Examination periods are bundled social events: deadlines, task relevance, peer activity, information needs and stakes all change together. The observational data cannot isolate a beneficial dose of stress.
Teachers can obtain the useful part without deliberately increasing threat. Use prediction before explanation. Present two plausible answers and ask what would distinguish them. Give an incomplete worked solution and ask what information is missing. Ask students to identify the first point at which their confidence drops. End a lesson with a question that cannot be answered by copying a sentence from the slide.
The adjacent eduKateSG guide How Epistemic Emotions Work owns the broader job of explaining curiosity, surprise, confusion and boredom. HSW-0241 has a narrower job: turning a learner’s uncertainty into an observable, diagnostic question without pretending that the question reveals the whole internal state.
Misconceptions
“More questions mean more curiosity.” Not necessarily. Questions are behaviour with multiple possible motives.
“Exams make students curious.” The 2026 study found a temporal association between educational phases and collective question-posing. It did not randomly assign examination pressure or establish a universal causal effect on curiosity.
“Confusion is productive.” Sometimes. Unresolved or diffuse confusion can waste attention. Productive uncertainty needs a path toward discrimination and testing.
“Good students ask sophisticated questions.” Expertise changes what can be asked, but simple factual questions can be exactly right for a novice. Judge a question by whether it locates and advances the learning problem.
“Teachers should never answer directly.” False. Direct explanation is often efficient. The issue is whether answering removes the learner’s chance to learn how to locate the uncertainty themselves.
Evidence and limits
The main source is Ali, Yang and Poh, “Curious behaviors in times of uncertainty: dynamics of student question-posing in online educational discussion”, published in npj Science of Learning in 2026. The researchers analysed large-scale student discussion from 2019–2023 and compared question-posing across recurring educational phases and the COVID-19 period at multiple timescales.
The strongest educational inference is descriptive: question-posing can vary systematically with the structure and relevance of uncertainty in real student communities. The study is not a classroom intervention, does not prove that question-posing improves examination scores, and cannot infer every student’s internal curiosity from text. Platform users are self-selecting, posting behaviour is shaped by community norms, and historical events introduce many simultaneous changes.
Those limits are exactly why the practical applications in this article are framed as teaching proposals to be tested through learner performance, not as effects established by the Reddit analysis.
The return
A learner cannot study every unknown at once.
The useful skill is to turn a field of uncertainty into one question with an address: this term, this relation, this prediction, this missing reason.
Pressure can make gaps visible, but pressure is not the mechanism we should worship. Structure is. A clear task makes uncertainty local. A good question makes it inspectable. An answer becomes learning only when the student can later act without it.
That is the movement worth training: not from no questions to many questions, but from vague not-knowing to a question that changes the next move.
Continue through the How Studying Works Numbered Series Reading Index, or return to the How X Works Hub.