The Half-Known Thing

You are flat on your back with foam blocks against your ears, and the room smells faintly of hospital. Above your eyes is a small angled mirror, and in the mirror is a screen, and on the screen — nothing yet.

Then a sentence arrives.

Which instrument was invented to imitate the human singing voice?

Something happens in you before you decide anything. A small lean. You don't know the answer, but you feel that you almost do — it's near, it's a thing with strings probably, and there is a distinct, faintly uncomfortable pull toward finding out. That pull is the entire subject of this morning's experiment, and the machine humming around your skull is measuring it.

Nineteen Caltech students took a turn in that scanner. Forty trivia questions each. Before every answer, two ratings: how curious are you, 1 to 7, and how confident are you that you know it, 0 to 100%. Then the answer, then the next question, forty times, while the magnet took pictures of their brains.

The paper that came out of it, in Psychological Science in August 2009, is called The Wick in the Candle of Learning. Its authors were Min Jeong Kang, Ming Hsu, Ian Krajbich, George Loewenstein, Samuel McClure, Joseph Tao-yi Wang and Colin Camerer. And the thing they found is stranger and more useful than "curiosity is good."

Fifteen years before the scanner, an economist at Carnegie Mellon had written down what he thought curiosity actually was. George Loewenstein went looking through a century of psychology on the question and concluded that most of it was describing the wrong thing. Curiosity, he wrote, arises

“when attention becomes focused on a gap in one's knowledge. Such information gaps produce the feeling of deprivation labeled curiosity.”George Loewenstein, “The Psychology of Curiosity”, Psychological Bulletin 116(1), 1994

Read that twice, because it contains a trap that most people never notice. If curiosity is a feeling of deprivation — a hole — then you cannot be curious about something you know nothing whatsoever about. A hole needs edges. You have to know enough to know that something is missing.

Which predicts something you can test. Curiosity shouldn't rise steadily as you know less. It should peak in the middle — at the point where you half-know. Know nothing and there's no gap you can feel; know everything and there's no gap at all.

That is exactly what the nineteen students in the magnet produced. Plotted against confidence, their curiosity made an upside-down U, and it peaked almost precisely at fifty per cent. Three quarters of them peaked between 0.45 and 0.55. The correlation between curiosity and P × (1 − P) — the mathematical shape of maximum uncertainty — was r = .44.

Now stop reading about it. The instrument below is the one they used, minus the magnet. Sixteen questions. Rate, guess if you like, and choose whether to open the answer.

A rating is cheap. Anyone can say they are curious. So the researchers ran a third experiment on thirty more students, and this time curiosity had a price.

Ten of them got twenty-five tokens for fifty questions — spend one to see an answer now, or keep it and never find out. Twenty of them got something crueller: after guessing, a wait of somewhere between five and twenty-five seconds, randomly chosen, with nothing to do. Sit it out and the answer appears. Quit and it's gone forever.

Both groups paid. For every one standard unit of extra curiosity, people sat there for an additional 3.7 seconds of empty time. In twenty-eight of the thirty individuals, taken one at a time, the correlation held.

Which is a small, exact, slightly embarrassing fact about human beings: we will spend real seconds of a finite life to close a gap about the colour of a bear's skin.

Sixteen students did the forty questions and went home. Eleven to sixteen days later, twelve of them came back — and were handed the same forty questions again, with no answers, and twenty-five cents for every one they could remember.

The answers they had been most curious about stuck. Roughly seven in ten of the high-curiosity answers came back, against something closer to four or five in ten of the low-curiosity ones, two weeks after a single reading.

But here is the sentence in the paper that does the real work. That memory advantage only appeared for the questions they had guessed wrong. Where their first guess had been right, curiosity predicted nothing at all.

A gap you filled correctly closes quietly. A gap you filled wrongly, and then had corrected, leaves a mark.

Which turns the wrong guess — the thing school taught most of us to be faintly ashamed of — into the active ingredient.

Five years later, Matthias Gruber, Bernard Gelman and Charan Ranganath at UC Davis pushed it further. They made people wait fourteen seconds for each answer, and during the wait they showed a photograph of a completely unrelated stranger's face. Curiosity about the trivia made people better at remembering the faces — information they had no reason to care about at all. Gruber described it as a vortex: “curiosity may put the brain in a state that allows it to learn and retain any kind of information… and also everything around it.”

We'll come back to how well that particular finding has held up. First — and without warning you in advance, which is the whole point — the instrument below.

Everything so far has been about somebody else's questions. Now the harder version.

Loewenstein's gap explains why a trivia question grips you for eleven seconds. It does not explain why you specifically will lose an afternoon to how a lock works, or to why that friend said that thing, or to the weather over a mountain, while the person next to you would rather chew gravel.

Curiosity has a shape as well as a size. And most people have never looked directly at the shape of theirs, because the things that pull them feel unrelated — a video game, a grandmother's accent, the noise a train makes. They look like a pile. They usually aren't.

So: put the pile on the table. Then say, for each thing, what the pull actually was. The tool will draw the threads.

Curiosity is a field with a lot of folklore stapled to it, so here is the honest ledger before the mirror.

How the idea arrived

• 1954 — D. E. Berlyne publishes “A Theory of Human Curiosity” and splits the thing in two: perceptual curiosity, which wants to look, and epistemic curiosity, which wants to know.
• 1960 — Berlyne's Conflict, Arousal, and Curiosity argues that novelty, surprise and incongruity are what raise the arousal that curiosity then works to lower.
• 1994 — George Loewenstein reviews the whole century and reframes curiosity as an information gap: a felt deprivation, which means it needs edges.
• 2007 — Michelle Chouinard's monograph on children's questions, built from transcripts, diaries, zoo visits and 67 four-year-olds, argues that question-asking is a mechanism of development, not a nuisance.
• 2009 — Kang, Hsu, Krajbich, Loewenstein, McClure, Wang & Camerer publish The Wick in the Candle of Learning: the inverted U, the caudate, the tokens, the 3.7 seconds, and the two-week memory advantage that only appears after a wrong guess.
• 2014 — Gruber, Gelman & Ranganath report that curiosity primes the hippocampus, and that unrelated faces seen during the wait are remembered better.
• 2018 — Golman & Loewenstein formalise information gaps in economics — including their dark twin, deliberate information avoidance, which turns out to be just as real a behaviour.
• 2021 — A larger study across younger and older adults replicates the core effect and finds the celebrated spillover to unrelated material missing.
• 2026 — A review of children's question-asking concludes that older children ask fewer but sharper questions — so the famous “curiosity dies as we grow up” curve may partly be children getting better at aiming.

Take your page with you

Nothing you wrote has left this browser. Copy it out, or print it, and it's yours on paper.

Try it at home: the Sealed Envelope

Write six questions on six slips of paper — three you're sure you know the answer to, three you're sure you don't. Ask someone to rate each one, out of seven, for how badly they want the answer, before anyone says anything. Then seal the answers in an envelope and put it on the mantelpiece. Whoever opens it first, and how long it takes, is the whole 2009 experiment in your kitchen — and the interesting slips are always the middle ones nobody could place.

Boss level: a week later, with no warning, ask them which answers they can still remember. Then ask which ones they had guessed wrong the first time. Watch whether those two lists are the same list.

You now know that you are most curious about the things you almost know. So: what have you been standing at the edge of for years without ever stepping in — and what would it cost you to find out?