Why are we so sure about things we’ve never examined?

Picture the last time you zipped a jacket without looking down. Thumb finds the pull, drags it up, two rows of teeth lock together behind it, done in under a second. Ask yourself right now, on a scale of one to seven, how well you understand what happens inside that slider. Most people answer fast, and answer high.

Now say the mechanism out loud. Which piece pushes which. Why the teeth interlock instead of merely touching. Where the two halves separate at the very bottom, and what stops them separating everywhere else. Somewhere in that sentence the explanation runs out of road for most people, and the number they would give themselves now sits lower than the one they gave a minute ago.

Rozenblit and Keil ran exactly this test at Yale in 2002, on sixteen graduate students, with eight ordinary devices on the table, the zipper among them: a speedometer, a piano key, a flush toilet, a cylinder lock, a helicopter, a quartz watch, a sewing machine. Phase one, rate your own understanding, one to seven. Phase two, write the causal chain, step by step, the way you would explain it to somebody standing next to the machine. Then rate yourself again.

Here is the trick your mind plays on you, and Rozenblit and Keil gave it a name: an illusion of explanatory depth. Ask why it happens and their answer is almost simple enough to miss. One factor concerns a confusion between what is represented in the head with what can be recovered from a display in real time. The zipper sits in front of you, its teeth catching the light, available to your eyes on demand, and your mind quietly treats that availability as knowledge already stored, filed in the same drawer as everything you actually learned. Looking at a thing and storing how it works turn out to be two separate acts, and the first gets mistaken for the second every single day.

The size of the drop, measured, holds up. On that same seven-point scale, the average self-rating across their studies fell from 3.89 before anyone attempted an explanation, to 3.10 the moment they finished writing the mechanism down, to 2.49 once a follow-up question forced real mechanical detail out of them. A repeated-measures analysis of variance put the drop beyond one chance in a thousand of being noise. And this was no case of people simply running out of steam over a long afternoon: the same students, given a short expert explanation afterward, rated their understanding back up again, the exact rebound plain fatigue could never produce.

Numbers are cheap when nobody checks them against anything. Put your own on record. Below is the same scale Rozenblit and Keil used, one to seven, this time about the zipper, and this time the slider remembers what you pick. Set your confidence, then ask the figure to show you the mechanism, and watch what the second bar does next to the first.

The gap between the two bars is the whole guide in a single glance. The first bar came from a feeling, the felt ease of watching teeth close a thousand times without a single thought spent on how. The second bar asks a harder question: could you actually draw the slider, the two rows of teeth, and the pin that guides them back together, clearly enough for a stranger to cut metal from your drawing. For nearly everyone who tries it, the second bar lands well below the first, and the size of that drop is the size of the illusion riding along with every confident guess you make about a thing you have only ever watched.

The zipper is the toy example. The same collapse shows up wherever someone is asked to defend a belief instead of just holding it.

Rebecca Lawson handed university students a blank sheet in 2006 and asked for a bicycle, drawn from memory, an object most of them had ridden thousands of times. The result: people were found to make frequent and serious mistakes, such as believing that the chain went around the front wheel as well as the back wheel. Years in the saddle build real skill in a rider’s hands and legs, and almost none in the ability to describe the machine underneath them. Familiarity and a stored mechanism turn out to be two separate accounts, and the errors shrank, without disappearing, for people who rode often, for men more than women, and for anyone tested with a real bicycle sitting in front of them.

The same shape turns up somewhere with real stakes attached. Fernbach, Rogers, Fox and Sloman asked people in 2013 to explain, mechanism by mechanism, how a policy they held strong opinions about would actually work: a flat tax, a cap-and-trade system, a single-payer health plan. Self-rated understanding fell again, 3.82 down to 3.45 on average, across six real policies. A second experiment isolated why: participants asked to build the mechanism showed a far larger drop in self-rated understanding than participants asked merely to list their reasons for a position, and only the mechanism group’s stated position moved toward the middle. The reasons group barely shifted at all. A 2021 attempt to repeat the full study kept the sturdy half of that finding alive and lost the flashier half: the drop in self-rated understanding reproduced cleanly, while the claim that explaining a mechanism reliably drags someone toward the political center failed all three attempts to rebuild it. Keep the half that held.

This is the whole reason one person who took a thing apart with their own hands outweighs a roomful of people who only ever watched it work from a comfortable distance. The room is reporting a feeling. The person with the wrench is reporting a memory of resistance, a part that stuck, the exact spot their own hands found trouble. Only one of those two reports survives being asked to explain itself, which is the whole case for going and finding the people who actually did the thing, and for weighing their word properly once you find them.