Curiosity: the information-gap engine.

Curiosity has always had a good reputation and a bad budget. Companies praise it in values statements and probe for it in interviews. But almost no course is engineered to produce it. It is filed as a trait some learners bring, rather than a state good design can open.

The research says otherwise, and it says so with unusual mechanistic detail. There is a formal theory of when curiosity switches on. There are controlled experiments on what it does to memory while it runs. There is a developmental model of how a moment of interest becomes a durable one. And there is longitudinal evidence — data tracking people over years — that the trait itself rivals effort as a predictor of academic performance.

This article assembles that evidence in reading order. First the theory: Berlyne’s founding taxonomy and Loewenstein’s information-gap account, which together specify when curiosity appears. Then the memory experiments — the trivia paradigms showing that curious states improve retention even for bystander material, and the brain-imaging work behind the (hedged) dopamine explanation. Then the four-phase model of interest development, which separates what designers can trigger from what they can only cultivate. Then the hungry-mind findings on trait curiosity. And finally the most practical lever in the set: questions asked before answers are available — where the pretesting research and the curiosity research turn out to be the same research wearing different labels.

From drive to gap

The scientific study of curiosity begins in earnest with Daniel Berlyne, who in 1954 did the unglamorous groundwork of making the concept usable. He distinguished perceptual curiosity, aroused by novel sights and sounds, from epistemic curiosity, the appetite for knowledge itself. He also split specific curiosity, aimed at one missing piece, from diversive curiosity, the general hunger for stimulation (Berlyne, 1954). Berlyne treated curiosity as a drive state aroused by what he called collative properties: novelty, surprise, incongruity, complexity, uncertainty. That inventory still describes most of what course designers reach for when they try to make material interesting. What his framework lacked was a precise trigger. When, exactly, does an exposed mind tip into wanting to know?

Forty years later, George Loewenstein reviewed the accumulated literature and supplied the answer most designers now use without knowing its source. The information-gap theory holds that curiosity arises when attention focuses on a gap between what one knows and what one wants to know (Loewenstein, 1994). The result is a state of felt deprivation, closer to an itch than to an appetite. It motivates the specific behaviour of closing the specific gap.

Three consequences fall out of this framing, and each one is a design instruction. First, curiosity requires knowledge: a person who knows nothing about a domain perceives no gaps in it. That is why novices are not curious about the questions experts find electric — and why the least knowledgeable learners are the hardest, not the easiest, to make curious. Second, the itch should grow as a gap narrows: the missing last piece pulls harder than the missing first one. Third, the state can be induced by the situation. Questions, puzzles, partial exposure and violated expectations all work by steering attention onto a gap — which means curiosity is a property of the encounter, not just of the person (Loewenstein, 1994).

One refinement matters for design. Litman distinguished curiosity as wanting from curiosity as liking. Deprivation-type curiosity is the uncomfortable need to close a gap; interest-type curiosity is the anticipated pleasure of discovery (Litman, 2005). The two feel different and arise under different conditions. They reward different design moves too: the cliffhanger serves the first, the invitation to explore serves the second. A curriculum that only ever needles produces anxiety; one that only ever invites produces browsing. The strongest designs alternate.

What a curious state does to memory

The theory would matter less if curiosity were merely pleasant. The reason it belongs in learning architecture is that curious states change what memory keeps. Kang and colleagues ran the paradigm that opened this line. People read trivia questions, rated how curious they were about each answer, and were scanned while they waited for answers to arrive. High-curiosity answers were remembered better — immediately, and one to two weeks later. Pupils dilated as the answers approached, and fMRI showed activity rising in reward-related regions, notably the caudate, before the answer appeared (Kang et al., 2009).

The same study produced the finding with the most practical value per word: curiosity as a function of confidence followed an inverted U. People were most curious when they half-knew — when their confidence in a guess was intermediate. They were least curious when the answer felt either obvious or hopeless (Kang et al., 2009).

Gruber, Gelman and Ranganath then extended the paradigm in the direction that matters most for designers. In their studies, states of high curiosity improved memory not only for the answers people were curious about. They also improved memory for incidental material — unrelated faces — shown while people waited. Anticipatory activity in the dopaminergic midbrain and nucleus accumbens rose during high-curiosity states. And the coupling between midbrain and hippocampus tracked the incidental-memory benefit (Gruber, Gelman & Ranganath, 2014).

The reading on offer — one account, consistent with the imaging rather than proven by it — is that curiosity places the brain’s reward-and-learning circuitry into an expectant state. In that state, the hippocampus encodes whatever arrives more strongly, for as long as the state stays open. The behavioural fact, though, stands on its own: a curious state is a promiscuous encoding window. Open the loop with a question a learner cares about, and the material delivered inside that window — including material that was not the question — is retained better.

Interest is a process, not a trait

A curious state lasts minutes. The commercially interesting question is how minutes become a lasting appetite — how a learner who was made curious once becomes a person who returns unprompted. Hidi and Renninger’s four-phase model is the standard map (Hidi & Renninger, 2006). Interest begins as triggered situational interest. A spark in the setting — surprise, a mismatch, personal relevance, a well-aimed question — captures attention, with no commitment from the learner. Sustained support turns it into maintained situational interest: involvement continues because tasks are meaningful and the environment keeps holding the thread.

From there, some learners cross into emerging individual interest. They begin to reengage on their own, building stored knowledge and stored value. A few arrive at well-developed individual interest. There, reengagement is self-sustaining, questions are self-generated, and persistence through difficulty comes cheap.

The model’s two design consequences point in opposite directions, and both are routinely ignored. The first is optimistic: situational interest is designable. Triggers are cheap, repeatable, and work on people who bring no interest of their own — which is exactly the population a mandatory training program serves.

The second is a boundary: individual interest cannot be installed. It develops — or fails to — across repeated, supported reengagements. And the later phases need different fuel than the earlier ones: autonomy, choice, depth and identity, where the early phases needed novelty and structure (Hidi & Renninger, 2006). Most corporate content is engineered, at best, for phase one — a strong open, then nothing. The model says the open was the easy part. The program earns its keep in the hand-off, where a triggered state either meets a scheduled return or evaporates.

The hungry mind

Alongside the state literature runs a trait literature. Its strongest single claim is von Stumm, Hell and Chamorro-Premuzic’s: intellectual curiosity is the third pillar of academic performance. They pooled the predictors of university-level achievement meta-analytically. A stable appetite for engagement — typical intellectual engagement, the tendency to seek out, enjoy and persist in effortful thinking — predicted performance alongside intelligence and conscientious effort (von Stumm, Hell & Chamorro-Premuzic, 2011). Curiosity mattered about as much as effort, and effort and curiosity together rivalled intelligence. Crucially for the construct’s independence, the curiosity contribution was incremental: not a proxy for being clever, but a separate input.

The mechanism they proposed is the compounding one the state research supports from below. Curious people place themselves in front of more information, more often, and encode more of what they meet. Small per-encounter advantages accumulate across years into measurable performance differences. For organizations, the hungry-mind finding reframes what curiosity is worth: it behaves less like a charming accessory and more like a rate parameter on cumulative learning. What the finding does not license — because it is correlational — is treated honestly in the limits section below.

Questions before answers

The most deployable lever in this whole field was found mostly by researchers who were not studying curiosity at all. Richland, Kornell and Kao asked whether attempting to answer questions before studying material would help or hurt later memory. The attempts were guaranteed to fail, since the material had not been read yet. Pretesting helped. Readers who first attempted related questions and then studied the passage beat readers given the same time for extended study (Richland, Kornell & Kao, 2009). The benefit held for the pretested content.

The surface explanation is attentional — a prequestion tells the reader what matters. The deeper reading is Loewenstein’s: a question converts an unknown-unknown into a known-unknown. It makes a gap visible, and a visible gap recruits the machinery this article has been describing (Loewenstein, 1994).

Put the strands together and a design grammar appears. Open loops before content. A unit that begins with a question the learner cannot yet answer starts inside a curious state; a unit that begins with an agenda slide does not. Calibrate the gap to the learner’s edge. The inverted-U says curiosity peaks at middling confidence (Kang et al., 2009). So the productive question is the one a learner half-knows — too easy opens no gap, too hard reads as someone else’s subject.

And sequence what matters into the window. If curious anticipation enhances encoding of what arrives during it — the pattern the incidental-memory work is consistent with (Gruber, Gelman & Ranganath, 2014) — then the moments after a well-aimed question are premium placement. Burying the key idea three screens after the reveal wastes the state the question paid for.

What the evidence doesn’t show

The research here is genuinely encouraging — which is exactly when a research library owes its readers the boundaries.

• The dopamine story is an account, not a verdict. The neural findings are anticipatory activations and connectivity patterns consistent with a reward-circuit mechanism (Gruber, Gelman & Ranganath, 2014); they are not a demonstrated causal chain in humans. Design decisions should lean on the behavioural effects, which stand independently.
• The paradigms are narrow. The memory effects were established largely with trivia questions and adult volunteers (Kang et al., 2009). Field evidence from classrooms and workplaces is far thinner, and deployed effect sizes should be assumed smaller than laboratory ones.
• Nobody knows how to train trait curiosity. The hungry-mind findings describe a stable disposition (von Stumm, Hell & Chamorro-Premuzic, 2011); no validated short intervention has been shown to raise it durably. Courses promising to make people curious are selling past the evidence.
• The peak is not a computable constant. The inverted-U appears across operationalizations — confidence, feeling-of-knowing, gap size — but its exact placement varies by person, topic and moment (Loewenstein, 1994). Calibration is an ongoing measurement problem, not a formula.
• The trait findings are correlational. Curious students may choose richer environments, and richer environments may breed curiosity; selection and reverse causation are not excluded (von Stumm, Hell & Chamorro-Premuzic, 2011). On their own, these findings do not justify treating curiosity scores as causal levers in selection decisions.
• Triggers are not learning. Interest sparked by material irrelevant to the learning goal decorates attention rather than directing it, and triggered interest fades without support — a caution the interest-development literature itself raises (Hidi & Renninger, 2006).

Where the evidence stops

1. 1The dopamine story is an account, not a verdict
2. 2The paradigms are narrow
3. 3Nobody knows how to train trait curiosity
4. 4The peak is not a computable constant
5. 5The trait findings are correlational
6. 6Triggers are not learning

© 2026 FUTURE PROOF™

Designing for the gap

Translated into practice, the evidence reads as a short grammar for anyone building courses, assessments or onboarding — six moves, each anchored to a finding.

Open a loop before every unit. Lead with a question, a prediction prompt, or a pretest the learner is expected to fail productively (Richland, Kornell & Kao, 2009). The point is not assessment; it is converting the invisible gap into a felt one before the content arrives (Loewenstein, 1994).

Calibrate questions to the knowledge edge. Curiosity peaks where confidence is middling (Kang et al., 2009). That makes diagnostic knowledge of each learner a curiosity instrument, not just a placement tool: the right question is the one a learner half-knows, and finding it requires knowing what they know.

Deliver the important material inside the window. Sequence the key idea immediately after the question that opened the state, with a beat of anticipation before the reveal. That placement is consistent with the anticipatory-encoding account, and it is cheap to build even while the mechanism is still being settled (Gruber, Gelman & Ranganath, 2014).

Trigger situationally, then hand off deliberately. Use novelty, mismatch and relevance to open interest in learners who brought none; then support the later phases with what they actually need — choice, autonomy, depth, and scheduled reengagement (Hidi & Renninger, 2006). A trigger with no return path is an expense, not an investment.

Build the knowledge that makes gaps visible. Because curiosity requires knowing enough to notice what is missing (Loewenstein, 1994), the basics are not the boring prelude to engagement — they are the substrate that makes engagement possible. Courses that chase interest while skipping fundamentals sabotage the very state they are chasing.

Serve both kinds of wanting. Deprivation-type curiosity responds to withheld resolutions and near-miss feedback; interest-type curiosity responds to open exploration and choice (Litman, 2005). Learners and moments differ; a design that only needles or only invites leaves half the motivation on the table.