The Curse of Knowledge: Why Experts Explain Things Badly

The tapping study is the memorable version. The underlying problem is one of the most consequential in this hub, because it quietly degrades almost all teaching, documentation and instruction.

Tappers and listeners

Elizabeth Newton ran the experiment as part of her Stanford doctoral work in 1990. Participants were assigned to one of two roles.

Tappers were given a list of well-known songs and asked to tap the rhythm of one on a table. Listeners had to name the tune.

Before the guess, tappers estimated how often listeners would get it right. Their average prediction was around 50 percent.

Across 120 tapped songs, listeners identified three.

The reason the estimate is so far off is that the tapper cannot hear tapping. They hear the song — melody, words, arrangement, the whole thing playing in their head while their finger keeps time. The listener hears a sequence of knocks.

And the tapper cannot switch that off. Asked to imagine the knocks without the tune, they fail, because the knowledge is not optional once it is there.

Why it is called a curse

The term comes from economics. Colin Camerer, George Loewenstein and Martin Weber introduced it in 1989 to describe better-informed agents in a market being unable to ignore their own information when predicting the behaviour of less-informed ones.

The economists' interest was that this violates a standard modelling assumption: that people can accurately reason about others who know less. The finding was that they systematically cannot.

It is the same failure of perspective-taking behind assuming your nervousness is visible, pointed in the opposite direction. There you over-attribute access to your inner state; here you over-attribute access to your knowledge.

Where it does damage

• Documentation that skips the step everyone gets stuck on, because the author stopped finding it difficult years ago.
• Teaching that moves at the pace of someone who already understands the material.
• Interfaces designed by people who know what the button does.
• Jargon used without definition, because it no longer registers as jargon to the person using it.
• Experts judging a beginner's confusion as inattention rather than as the ordinary condition of not yet knowing.

The pattern in all of these is not arrogance. It is an honest inability to simulate a mind without a piece of information that your own mind now contains — and the failure is invisible from inside, because the explanation feels perfectly clear to the person giving it.

There is a practical asymmetry here that makes the curse particularly hard to catch. When an explanation fails, the expert receives a confused expression, which is ambiguous — it could mean anything. When it succeeds, they receive understanding. Neither outcome tells them which parts of the explanation did the work.

Without that feedback, the estimate never gets corrected. The tapper never finds out how the tapping sounded.

The hindsight connection

The curse of knowledge is closely related to hindsight bias, the tendency to see an outcome as having been predictable once you know what happened. Both stem from the same root: new information contaminates your model of a state that did not contain it.

Hindsight bias is the version pointed at your past self. The curse of knowledge is the version pointed at other people. In both cases you are trying to reconstruct a mind with less information than you now have, and you cannot do it cleanly.

The two have a shared practical consequence. Asked why a project failed, the person who knows the outcome will describe it as obvious. Asked to explain something they understand, the same person will describe it as simple. Both descriptions are sincere and both are wrong.

Children and the false-belief version

The curse has a developmental counterpart that makes the mechanism unusually visible.

In classic false-belief tasks, young children shown where an object really is will predict that someone who did not see it moved will look in the new location. They cannot model a mind without the information they have just acquired.

Adults pass those tasks easily, which was long taken to mean the limitation is outgrown. Work by Susan Birch and Paul Bloom argued instead that it is reduced rather than eliminated — adults show the same bias in weaker form, and it strengthens as the information becomes more familiar and harder to set aside.

On that reading, an expert explaining their subject badly is running a milder version of a three-year-old's mistake, on material they have known for twenty years.

What actually helps

Not willpower. You cannot decide to forget something, and trying to imagine ignorance more vividly does not work — the estimate stays anchored on what you know.

The things that do help are all structural.

Testing on someone who genuinely does not know is the only reliable method. Watching a real novice get stuck is information you cannot generate by introspection, and it is usually uncomfortable in a way that guessing is not.

Writing down what a reader must already know before the first sentence forces the assumptions into view, where they can be checked rather than assumed. And where a step feels too obvious to include, that feeling is itself evidence you have crossed the gap and can no longer see it.

One reframe is worth holding onto. Knowing something and being able to convey it are genuinely separate capacities — the same separation that makes a word available but not retrievable. Having knowledge is not the same as having access to what its absence was like.