The Curse of Knowledge: Why Experts Explain Things Badly

Try this. Pick a song everyone knows and tap the rhythm on a table for someone. Before they answer, guess how likely they are to name it. Most people guess something around a coin flip. It feels obvious, because in your head the song is playing: melody, words, arrangement, the lot. Your listener has knuckles on wood.

That is Elizabeth Newton's tappers-and-listeners study, run for her Stanford dissertation in 1990. Tappers predicted their partners would identify the tune roughly fifty percent of the time. The actual figure was about two and a half percent, around three successes in a hundred and twenty attempts. The gap is not a failure of listening. It is the tapper being unable to subtract what they know.

Where the term comes from

The phrase entered the literature through economics. In 1989 Colin Camerer, George Loewenstein and Martin Weber published work on what they called the curse of knowledge, showing that better-informed agents could not fully discount their private information when predicting how less-informed people would behave, even when doing so would have paid them. It was a rational-actor problem before it was a teaching problem.

That origin matters, because it rules out the easy explanation. The bias is not laziness or arrogance. These were people with a financial incentive to model ignorance accurately, and they could not do it. Knowledge does not sit in a separate compartment you can set aside. It reorganises how the material looks to you, permanently.

What it does to an explanation

The curse shows up in specific, predictable ways, and once you can name them you start seeing them in your own drafts.

• Compressed steps. An expert experiences three actions as one move, so the write-up has three steps where the novice needs seven.
• Unmarked jargon. Words that stopped feeling technical years ago pass through unexplained, because they no longer register as words at all.
• Missing purpose. The expert knows why the step exists, so the instruction says what to do without saying what it is for, which makes recovery from any error impossible.
• Overestimated pace. Material that takes the expert ninety seconds is budgeted at ninety seconds, and the room falls behind in minute two and stays there.
• Assumed structure. The expert sees the shape of the whole subject, so they start in the middle, where the interesting part is.

The last one is the most damaging, because the novice has no framework to hang anything on. Chunking research is relevant here: expertise partly consists of having assembled larger units of meaning, and an explanation pitched at those units is unreadable to someone who has not built them yet.

Who it hurts

Teaching

The best mathematician in a department is not automatically the best person to teach the first-year course, and the reason is structural rather than personal. Their fluency has removed exactly the friction the beginners are stuck in. Newer teachers, closer to their own confusion, often explain the hard bits better and the advanced bits worse.

Medicine and law

A consultation compresses years of pattern recognition into a few minutes of speech. Studies of patient recall are sobering: a large share of what is said is not retained or is misunderstood, and the patient rarely signals it, because not understanding your doctor is socially costly. The same holds for legal advice and financial paperwork.

Documentation and onboarding

Internal docs are written by the person who built the thing, at the moment they know it best, which is the worst possible moment for the task. This is why the step that everyone gets stuck on is so often the one the author did not think to write down.

Fixes that actually work

Every reliable remedy has the same shape: stop consulting your intuition about what is obvious, and get external evidence. Introspection is precisely the faculty the bias has compromised.

1. 1Teach-back. Ask the other person to explain it back in their own words, not to confirm they understood. Their version shows you exactly which joint you skipped, and it is standard practice in patient communication for this reason.
2. 2Test on a real novice. One person from outside the domain, working through your instructions while narrating where they stall, will find more problems than three rounds of expert review.
3. 3Write the explanation before you finish learning. The window where you understand something and still remember not understanding it is short and valuable. Take notes in it.
4. 4Use concrete analogies, then say where they break. An analogy imports a structure the reader already owns. Naming its limits stops it hardening into a misconception.
5. 5Ban the words obviously, simply, just and of course. They are markers of the bias, and they also tell a struggling reader that their struggle is illegitimate.

The most systematic version of this is the Feynman technique: explain the thing in plain language, notice where you resort to jargon or hand-waving, go back to the source, repeat. It is usually sold as a comprehension check, and it works as one, but its real power is that it forces you to simulate a listener who knows nothing.

Preparing to explain also changes how you learn in the first place. That is the protege effect: learners who expect to teach material organise it differently and retain more, because they are hunting for the joints where a listener would fall out.

A caution about the fix

Simplifying has its own failure mode. Explanations pitched so low that they remove all difficulty produce the comfortable sensation of understanding without the substance, which is the trap in the illusion of knowing. The target is not easy. It is honest about difficulty while supplying the missing structure.

This is one reason our own lessons are built as short narratives with a question at the end rather than as summaries. Disclosure: MindSnap is our app, so treat that as an example of the principle rather than an endorsement. A story forces you to state the causal steps in order, and a question at the end gives you evidence about whether the reader followed them, which is teach-back at scale.

The practical upshot is small and slightly humbling. You cannot introspect your way to a good explanation, because the thing you would be consulting is the thing that changed. You can only find out by watching someone try.

The tapper hears the song. The listener hears knocking. Both are certain the task is simple.