The Questions Change. The Curiosity Doesn't.
Key Takeaways
- Curiosity does not fade with age. The one real longitudinal study, following adults aged 17 to 92, found information-seeking curiosity unchanged. What declined was stimulation seeking, which is a different appetite.
- Children look more curious mostly because of the room. The same student asks 0.11 questions an hour in a classroom and 135 where asking is the only way to learn.
- Curiosity peaks near 50% confidence, so what switches it off is the feeling that the gap is closed, not the availability of answers.
- Adding sources and an explanation to an AI answer cut follow-up questions from 71.4% to 12.7%. Most participants never clicked the sources.
- Answers do not sate curiosity, they relocate it. We seek information partly to learn what to be curious about.
Someone made a picture of Sabeer Bhatia and split it down the middle.
On the left, a man in his late twenties in a dark suit, shaking a hand at what looks like a reception. On the right, the same man three decades later, glasses on, mid-gesture, talking to a room. The captions read He’s still curious. and So is he. Along the bottom: The questions change. The curiosity doesn’t.
It is a good line, and my first instinct with a good line is to check whether it survives contact with the record.
This one mostly does. What is interesting is the part that doesn’t.
What Actually Changed Between Those Two Photographs?
Almost everything except the posture.
The man on the left had a specific question. In the mid-1990s, email was tied to whoever sold you your internet connection. Change providers, lose your address. Read your mail from someone else’s machine, and you generally couldn’t. Bhatia and Jack Smith asked what happens if you put the mailbox in the browser instead, and launched Hotmail on 4 July 1996, a date the founders said they picked for the obvious symbolism.
The scale of the answer is documented in an unusually clean source. When Microsoft announced the acquisition on 31 December 1997, its own press release put Hotmail at “more than 9 million members worldwide.” Eighteen months from launch.
The man on the right is asking about something else entirely. He now works on AI and education: in October 2023 his company pivoted to an AI-driven entrepreneurship course built with Jamia Millia Islamia, carrying university credit, using AI to assess how students actually reason rather than what they can recall.
Two questions with nothing technical in common. Web-based mail is a 1996 problem. Machine assessment of reasoning is a 2026 one. Nobody carries a method from one to the other.
What carries is the shape of the move: notice something everyone has accepted as fixed, and ask why it is fixed.
Here is the thing worth checking, though. A picture of someone at 27 and at 57 can show you that they are still asking questions. It cannot show you that the asking never lapsed. Thirty years compressed into two frames will always look like a straight line.
While We’re Checking: The Number Everyone Repeats
A small detour, because it illustrates the essay’s own point better than any argument I could make.
Ask almost anyone what Microsoft paid for Hotmail and you will hear $400 million. It is in the profiles, the business-school retellings, the LinkedIn posts. It has the texture of a fact.
Microsoft’s press release announcing the deal says, in plain words, that financial terms of the transaction were not disclosed. Neither Microsoft nor the founders have confirmed a figure since. Crunchbase lists the amount as undisclosed. Some sources say $500 million.
So the number is a report. Possibly a good one. But thirty years of confident repetition have quietly converted “reported” into “is,” and almost nobody performs the two minutes of checking that would surface the difference.
That is what an incurious relationship with information looks like, and it is worth noticing that it can happen to a fact you like, about a person you admire, in service of a point you agree with. Those are the ones that get through.
Does Curiosity Actually Survive, or Do We Just Say That?
It survives. This is the part I expected to have to hedge, and the evidence turned out to be better than the slogan.
The only real longitudinal work here is Giambra, Camp and Grodsky (1992) in Psychology and Aging: 1,356 men and 1,080 women aged 17 to 92, with a repeat measure six to eight years later. Their finding, in their own words, is that the scales measuring curiosity as information seeking “did not change across the adult life span,” while the scales measuring stimulation seeking, meaning the need for external input to relieve boredom, significantly decreased.
Two different things, moving in different directions, usually collapsed into one word. The appetite for novelty fades. The appetite for knowing does not.
A 2025 preregistered study of 1,218 adults aged 20 to 84 adds a wrinkle worth having. Trait curiosity declined slightly with age and state curiosity rose slightly, both by tiny amounts, and the curve was not a slide at all: middle-aged adults sat lowest, older adults highest. Whatever is happening, it is not a long fade from a childhood peak.
Three things I nearly wrote, and had to throw out. That children ask 300 questions a day and adults ask six. That NASA found 98% of five-year-olds are creative geniuses. That curious people measurably live longer. The first traces to a retailer’s marketing poll, the second has no published method or sample, and the third rests on a single 1996 study whose effect size is not in the abstract and whose full text is behind a paywall with no open copy anywhere. All three are repeated constantly, including by people making exactly the argument this essay is making.
So Why Does It Look Like Children Ask More?
Because they are usually somewhere that permits it, and adults usually are not.
The cleanest number in this literature comes from Graesser and Person (1994). A student in a classroom asks 0.11 questions per hour. The same kind of student in one-to-one tutoring asks 26.5. Put them in an environment where asking is the only way to get anything, and it goes to 135.
That is a two-hundred-fold swing with the person held constant. In an ordinary classroom, 96% of the questions asked are the teacher’s.
The same pattern shows up in the small British study people usually cite as evidence that school kills curiosity. Tizard and Hughes recorded four-year-olds asking roughly 27 questions an hour at home and two or three at nursery. The same children, in the same week. It is quoted as a developmental finding and it is nothing of the kind. It is a finding about rooms.
Curiosity is not a quantity that drains. It is a behaviour that environments either permit or suppress, and most adult environments suppress it, because asking marks you as the person who did not already know. It is also the quiet argument inside what a teacher was ever for: rooms that reward asking are rare, and they were always built on purpose.
What Happens to Curiosity When Answers Get Cheap?
Here the picture genuinely darkens, though not in the way the usual complaint has it.
Start with what curiosity actually responds to. Loewenstein’s information-gap account predicts, and Kang and colleagues confirmed, that curiosity is an inverted U in how much you already know. It peaks when you put your confidence around 50%. Know nothing and there is no gap you can perceive. Know everything and there is no gap at all.
Which means the thing that switches curiosity off is not an abundance of answers. It is the feeling that the gap is already closed. And a model cannot warn you that it is still open, because inference begins only after the question is fixed.
And that feeling is precisely what a fluent answer manufactures, whether or not it is earned. In a 2021 PNAS study of 1,917 people, participants who had used a search engine predicted they would score substantially higher on later questions than those who had not, 5.95 against 4.58. Then they took the test. They scored 3.73 against 3.17, a difference that did not reach significance. The confidence moved. The knowledge did not.
The sharpest result is newer. A preregistered CHI 2025 experiment with 308 participants measured how often people asked a follow-up question after an AI answer:
| What the answer included | Asked a follow-up |
|---|---|
| The answer alone | 71.4% |
| Answer plus sources | 34.4% |
| Answer plus explanation | 28.2% |
| Answer plus both | 12.7% |
Adding the apparatus of rigour cut follow-up questions by a factor of nearly six. And 189 of the 308 participants never clicked a single source.
Sit with the shape of that. The sources did not make people check. They made people stop asking, which is the opposite of what a citation is for.
The threat was never cheap answers. It is answers that feel complete. A bare answer leaves a visible seam and you keep pulling at it. A well-sourced, well-explained answer closes the seam whether or not the underlying question was settled, and closing the seam is exactly what turns curiosity off.
Does Getting an Answer End the Curiosity?
No, and this is the finding that earns the picture’s caption.
The intuitive model is that curiosity is an itch and answers scratch it. A 2026 PNAS study of 5,831 people tested that directly against the opposite model, where satisfying curiosity reinforces it. The reinforcement account won. Getting a satisfying answer increased subsequent curiosity, but only for semantically related material. The authors’ conclusion is the sentence I would put on the wall: people seek information partly in order to “learn what to be curious about.”
So curiosity does not deplete. It relocates, and it relocates along the lines your existing knowledge makes visible.
Which is exactly what thirty years of one person’s questions look like from the outside. Not a fixed quantity of wonder being spent down, but a moving front, each answer opening the adjacent question, the map growing in the direction you already walked.
Which Part of Curiosity Is Worth Protecting?
The expensive part, which is also the part nothing on the market is trying to sell you.
| Answer-getting | Question-forming | |
|---|---|---|
| What it costs now | Nearly nothing | Same as it ever was |
| What it feels like | Fast, resolving, productive | Slow, uncomfortable, unproductive |
| Who is working on it | Every AI lab on earth | Almost nobody |
| Fails how | Loudly. You get a wrong answer | Silently. You get a good answer to the wrong question |
| Trained by | Using tools | Working where you might be wrong |
The bottom-left cell is why this is not a story about machines making us stupid. Answer-getting genuinely got better, and that is genuinely good.
The bottom-right cell is the whole problem. A wrong answer corrects itself eventually, because reality pushes back. A confident, well-sourced, elegantly reasoned answer to a question that should never have been asked does not correct itself. There is no error signal. You just proceed.
What This Looks Like in Practice
Three things, none of them glamorous.
Write the question down before you ask it. In your own words, including why you want to know. It takes under a minute, and about half the time you discover the question you were about to ask is not the one you actually wanted answered. That discovery is the entire mechanism, and no system was ever positioned to hand it to you. It is the same effortful step that never announces its own absence.
Stay near an edge. Curiosity needs a felt gap to point at, and comfortable expertise stops producing them. This is the underrated reason people who keep asking good questions late in life tend to keep changing fields. Not restlessness. Gap manufacture.
Check the fact you like. The $400 million was in this essay’s own first draft as plain fact, because I had absorbed it the same way everyone else did. Curiosity that only points outward, at things you doubt, is not curiosity. It is argument with extra steps.
The picture is right, then, though slightly more interesting than its caption.
The questions changed completely. Web-based email in 1996 and machine-assessed reasoning in 2026 share no technique, no vocabulary, and no tools. What persisted was not a subject or a skill. It was the willingness to keep standing somewhere the answer was not already known.
That was always the rare part. It was rare when answers were expensive, and it is rarer now that they are free, because the thing that used to force you into it has been optimised away. Which is the narrow, practical shape of the larger question about what is left for us.
The questions change. The curiosity doesn’t, if you keep paying for it.
Frequently Asked Questions
Does curiosity decline with age?
The longitudinal evidence says no. Giambra, Camp and Grodsky (1992) followed adults aged 17 to 92 and found that scales measuring curiosity as information seeking did not change across the adult lifespan. What did decline was stimulation seeking, the need for external input to relieve boredom. Those are two different appetites usually collapsed into one word: the taste for novelty fades, the taste for knowing does not.
Why do children seem to ask so many more questions than adults?
Mostly because of the room, not the age. Graesser and Person (1994) measured 0.11 questions per hour from a student in a classroom, 26.5 in one-to-one tutoring, and 135 in an environment where asking was the only way to learn anything. In an ordinary classroom 96 percent of the questions are the teacher's. The often-quoted finding that four-year-olds ask 27 questions an hour at home and two or three at nursery involved the same children in the same week, so it is a fact about environments rather than about development.
Does AI make people less curious?
Not by making answers cheap. Curiosity is an inverted U in what you already know, peaking near 50 percent confidence, so what switches it off is the feeling that the gap is closed. A preregistered 2025 experiment found that follow-up questions after an AI answer fell from 71.4 percent to 12.7 percent once the answer carried both sources and an explanation, while 189 of 308 participants never clicked a source. The apparatus of rigour stopped people asking rather than making them check.
Does getting an answer satisfy curiosity or increase it?
Increase it, along related lines. A 2026 study of 5,831 people tested the idea that answers sate curiosity against the idea that they reinforce it, and reinforcement won: satisfying answers raised subsequent curiosity, but only for semantically related content. The authors conclude that people seek information partly in order to learn what to be curious about. Curiosity relocates rather than depletes.
How do you stay curious as you get older?
Choose rooms that permit asking, since the environment does more work than the disposition. Notice when an answer feels finished, because that feeling is manufactured by fluency and is the thing that actually stops inquiry. And put a delay between the question and the lookup: in a 2021 study, a 25-second wait before search was enough to remove the gap between how much people thought they knew and how much they did.