In 1990, a Stanford graduate student named Elizabeth Newton ran an experiment that has become one of psychology's most quoted demonstrations. Participants were divided into two roles. Tappers were given a list of well-known songs and asked to tap out the rhythm of one on a table. Listeners had to identify the song.
Before each attempt, tappers were asked to predict how often listeners would guess correctly. Their average prediction was 50%.
The actual success rate was around 2.5%.
The gap comes from something the tappers could not switch off. As they tapped, they heard the song — the melody, the words, the full arrangement playing in their heads. The listener heard a series of disconnected knocks. Knowing the tune made it feel obvious, and the tappers could not simulate a mind that did not have it.
This is the curse of knowledge: once you know something, you lose the ability to accurately imagine what it is like not to know it.
Why it is so persistent
The curse is not a failure of effort or empathy. People genuinely trying to explain clearly still fall into it, because the mechanism operates below conscious control.
Knowledge changes how information is processed, not just how much is stored. An experienced programmer does not see individual characters in a line of code; they see a pattern with a meaning. A doctor reading an X-ray does not scan pixel by pixel; the abnormality effectively announces itself. A fluent reader cannot look at a word in their own language and perceive it as shapes.
This is **chunking** — expertise reorganises information into larger, meaningful units. It is what makes expertise powerful. It is also what makes expert perception unrecoverable. You cannot un-chunk on demand and see the raw pieces again.
So when the expert says "it's obvious", they are reporting an accurate description of their own experience. The word does not mean the thing is simple. It means their brain has stopped showing them the steps.
Where it shows up
**Technical documentation.** Instructions that skip a step because the author does not register it as a step. "Simply configure the environment variables and run the build" contains, for a beginner, perhaps a dozen unstated decisions.
**Teaching.** The relationship between subject mastery and teaching quality is much weaker than intuition suggests. A brilliant researcher may be a poor first-year lecturer precisely because the material is too deeply internalised to be decomposed.
**Product design.** Interfaces designed by people who built the system, tested on people who understand the system, shipped to people who do not. The designers cannot see the confusion because they cannot unlearn the model in their heads.
**Writing.** Jargon feels like precision to the writer and like a locked door to the reader. Technical writers describe fighting this constantly — the vocabulary that makes communication efficient among specialists is exactly what excludes everyone else.
**Management and handovers.** Instructions that assume shared context. "Handle it the usual way" is complete information to the person who has done it fifty times and nearly useless to the person who has not.
**Personal relationships.** Assuming others know what you mean, what you want, or why you are upset, because your reasoning feels self-evident from inside your own head.
The intermediate-explainer effect
One consequence is counterintuitive but widely observed: the best explanation often comes from someone only slightly ahead of the learner.
Someone who learned the concept last month still remembers the confusion. They know which part was the sticking point, which analogy finally worked, and which piece of standard terminology was actively misleading. That memory is a resource, and it decays.
This is why peer tutoring frequently outperforms expert instruction for beginners, why the most useful forum answers often come from ordinary users rather than developers, and why a colleague who recently onboarded is usually the best person to write the onboarding guide.
It is also an argument for documenting things while you are still confused, not after you have mastered them. The window of useful memory is short.
Why it is hard to fix by trying harder
Simply being told about the curse of knowledge does not remove it. Studies of related biases repeatedly find that awareness produces only modest correction. The knowledge stays active, and the estimate stays anchored to it.
Worse, effort can misfire. Experts told to simplify sometimes strip out detail while keeping the underlying structural assumptions, producing an explanation that is shorter but no clearer. The missing piece was never the detail — it was the scaffolding.
What works better is replacing introspection with external feedback.
What actually helps
**Watch someone use it.** Not ask them — watch. People routinely report that instructions were clear while visibly struggling. Usability testing exists because self-report is unreliable here. Sit behind someone, stay silent, and note every hesitation.
**Ask them to explain it back.** Comprehension failures surface immediately when the listener has to reproduce the idea in their own words.
**Name the audience precisely.** "Explain to a new graduate hire on their first day" produces measurably different output from "explain simply", because it forces you to model a specific person with specific gaps.
**Write down the prerequisites explicitly.** Before explaining, list what the reader must already know. The act of listing surfaces assumptions you were not aware of making.
**Use concrete examples before abstractions.** Abstract statements are compressed summaries of experience the listener does not have. Examples supply the experience first, so the abstraction has something to attach to.
**Keep a record of your own confusion.** When learning something, write down what confused you and why. Later, that document is more valuable than anything you could reconstruct from mastery.
**Let a non-expert edit.** Someone outside the field will flag the sentences that do not land. Their confusion is data, not a deficiency.
The flip side
The curse also runs in the other direction, and this half is less discussed. Because experts cannot easily model ignorance, they also underestimate how much someone else knows, and over-explain in ways that feel condescending.
Both errors come from the same root: the difficulty of accurately modelling another mind's contents. The fix in both directions is the same — stop guessing and check.
Summary
Knowledge is irreversible. Once you understand something, your brain reorganises it into patterns you cannot decompose on demand, and you lose access to what confusion felt like. This explains bad documentation, confusing teaching, and the strange effectiveness of people who learned the topic recently. Awareness alone does not fix it. What fixes it is external feedback — watching real users, asking for explanations back, naming a specific audience, and treating other people's confusion as information about your explanation rather than about them.
Be the first to share a thought.