Curse of Knowledge

Once you know, you can’t imagine not knowing.

Named by economists Camerer, Loewenstein, and Weber in 1989: once you know something, you can no longer simulate the mind of someone who doesn’t. It’s why experts write confusing docs and teams ship jargon.

Origin

Colin Camerer, George Loewenstein, and Martin Weber coined the term in a 1989 economics paper: better-informed traders couldn’t ignore their own knowledge when predicting the behavior of less-informed ones.

Elizabeth Newton’s 1990 tapping study made it famous — tappers drumming well-known songs predicted listeners would name half of them; listeners got 2.5%. Chip and Dan Heath’s Made to Stick carried the idea into communication and design.

The Curse

Knowledge is one-way glass: once acquired, it silently rewrites your model of what other people can see. Experts systematically overestimate how much novices know — not from arrogance, but because un-knowing is cognitively impossible.

Coined by
Camerer, Loewenstein, Weber (1989)
Source
The Curse of Knowledge in Economic Settings
In practice
Test on people who don’t know
The study 01 / 08

Tappers and Listeners

Newton’s tappers heard the full song in their heads and predicted a 50% success rate. Listeners heard knuckles on a table and named 2.5%. Every expert demoing their product is a tapper.

Tappers predicted
50%
Listeners scored
2.5%
Every demo
Is a tapping
Language 02 / 08

Jargon Is Invisible to Its Speakers

Inside the team, “sync,” “workspace,” and “instance” are plain words. To a new user they’re a locked door. The curse hides exactly which of your words are jargon — only outsiders can see the list.

To the team
Plain words
To new users
Locked doors
Only outsiders
See which is which
Onboarding 03 / 08

Written by People Who Know

Onboarding fails when it’s designed by people who can’t remember not understanding the product. The empty state that “explains everything” explains it in the vocabulary of someone on the other side of the curse.

Written by
People who know
Read by
People who don’t
Gap
The curse itself
RelatedPower Law of Practice
Interfaces 04 / 08

The Obvious Icon Isn’t

To its designer, the icon obviously means export. To users it means whatever their history says it means — often nothing. Pair icons with labels until recognition is proven, not assumed.

Obvious to
Its designer
Fix
Icon plus label
Assume
Nothing is obvious
Antidote 05 / 08

Watch Someone Who Doesn’t Know

The only reliable cure is borrowed ignorance: five first-time users narrating their confusion will find what no expert review can, because the experts literally cannot see it. Usability testing is curse removal.

Cure
Borrowed ignorance
Five users
Beat expert review
Testing
Is curse removal
Writing 06 / 08

Explain Like They’re Smart, Not Informed

The fix for cursed writing isn’t dumbing down — it’s supplying context. Assume intelligence, assume zero shared vocabulary, define terms at first use, and lead with what the reader can already picture.

Assume
Smart, uninformed
Define
At first use
Lead with
The familiar
✦

The Curse of Knowledge in the Age of AI

Prompts assume context the model doesn’t have — and models explain with knowledge the reader doesn’t have. The curse now runs in both directions.

In the Age of AI 07 / 08

Prompts Assume Context

A prompt that fails often fails as a tapper: the writer heard the full song — the codebase, the constraints, the goal — and typed three words of it. The model, like the listener, gets knuckles on a table.

Failed prompt
A tapped song
Fix
Say the context out loud
In the Age of AI 08 / 08

The Model Is Cursed Too

A model that knows everything explains accordingly — the response assumes vocabulary the reader may not have. Asking for the audience (“explain for a new hire”) is de-cursing the machine.

Cursed output
Assumes your vocabulary
De-curse
Name the audience
Further Reading

Wikipedia

Books