Ask someone to explain a mechanism and the understanding evaporates

The experiment is a small humiliation

The standard procedure asks people to rate how well they understand how something familiar works — a zip, a flush toilet, a bicycle, a lock — on a numbered scale. They are then asked to write out a full step-by-step explanation of the mechanism. Afterwards they rate their understanding again, and the second rating is substantially lower than the first.

What happened in between is the whole finding. The attempt to produce the explanation revealed that what felt like knowledge was a familiarity with the object and a rough sense that the parts connect somehow. This has been called the illusion of explanatory depth, and the basic effect has been reproduced across a range of materials by different groups, which puts it in better standing than a good many effects from the same era.

Why mechanisms specifically

The illusion does not appear for everything. Asked how well they know a set of facts, a procedure, or a narrative, people rate themselves and then perform roughly in line with the rating. It is causal mechanisms that produce the gap, and the proposed reason is that mechanisms have hidden internal structure while facts do not.

A fact is stored whole, so you either have it or notice that you do not. A mechanism is stored as a small number of visible parts and a strong impression that the connections between them are available if needed. That impression isn’t checked in ordinary life, because nothing in ordinary life requires you to run the explanation end to end. The bicycle works whether or not you can describe how the pedals reach the wheel.

Where the same illusion does real damage

It scales badly upwards. The mechanisms people are most confident about are usually the largest ones: how an economy responds to a policy, how a supply chain fails, how a piece of legislation would work in practice, how an organisation would react to a change. These have the same structure as the zip — a few visible parts and an assumption that the linkages exist — and they’re far less likely to be checked, since nobody is going to ask you to draw one.

It also explains a familiar sequence in projects. A plan looks straightforward until somebody writes the steps down, at which point three of them turn out to have no content. The plan had not got worse. It had been forced through the same procedure the experiment uses, and the outcome was the same.

The extension that’s less well supported

A widely reported follow-up argued that asking people to explain the mechanism of a contested policy reduced how extreme their stated positions were, while asking them to list reasons did not. It is an appealing result, because it suggests a cheap intervention for entrenched disagreement.

That extension has held up less consistently than the core effect. Some replication attempts have produced smaller effects or none, the outcomes depend on the issue and the wording, and the mechanism-versus-reasons distinction may not generalise as cleanly as the original framing implied. The sensible position is that the illusion of explanatory depth is well supported and its application as a depolarising technique is not.

Using it deliberately

The practical value is that the test is cheap, private and unusually informative. Before relying on your understanding of anything, write the mechanism out as a sequence somebody else could follow, without vague connectives. The places where the writing stalls are the places the understanding was never there, and they’re almost never the places you expected.

The same move works on a claim somebody else is making, though it needs to be asked without hostility to get an honest answer. How does that actually happen, step by step, isn’t a rhetorical question. It frequently produces a genuine and useful answer, and when it does not, both parties have learned the same thing at the same moment.

What the illusion is not evidence of

It is not evidence that people are stupid about mechanisms, and reading it that way misses what makes the finding interesting. Carrying a compressed sketch instead of a full mechanism is efficient, since the full mechanism would be expensive to store and is almost never needed. The system works well and it fails in one specific way, which is that it does not signal how compressed it is.

Nor does it mean that everybody must understand everything from first principles, which is impossible in a world with any division of labour. Most of what any person knows is a placeholder pointing at somebody who knows more, and that is a perfectly sound arrangement. The failure is in mistaking the placeholder for the thing, which happens specifically at the moment when you are about to make a decision that depends on the detail.