You Understand How a Zipper Works. Try Explaining It and Watch That Fall Apart.
Rate, on a scale of one to seven, how well you understand how a zipper works. Most people land somewhere around a five or six — confident, unbothered, moving on. Now write out, step by step, what actually happens between the two rows of teeth when you pull the slider. Where does the interlocking force come from. Why doesn't it come apart sideways. What's the slider actually doing mechanically that your fingers aren't.
If you're like the participants in a study that's been quietly reshaping how psychologists think about knowledge for over twenty years, you'll get a few sentences in, stall, and realize you don't actually know. You knew you could use a zipper. You believed, right up until the moment you tried to explain it, that using it and understanding it were the same thing.
The Study That Named the Gap Between Using and Knowing
In 2002, Yale psychologists Leonidas Rozenblit and Frank Keil published a paper in Cognitive Science that gave this gap a name: the illusion of explanatory depth. Their method was simple and almost cruel in its efficiency. Participants first rated how well they understood a set of everyday mechanical objects — a zipper, a flush toilet, a sewing machine, a speedometer, a piano key, a cylinder lock, a helicopter, a quartz watch. Then they were asked to write a detailed, step-by-step causal explanation of how each one actually worked. Then — and this is the part that does the damage — they rated their understanding again.
Across all eight devices, the second rating came in lower than the first, consistently, for nearly everyone. Not because anyone learned new information in between. Nothing about the toilet changed between the first rating and the second. What changed was that the participants had been forced to produce the understanding they'd claimed to have, and production revealed what mere recognition had been hiding. You can recognize a working zipper, use it a thousand times, feel completely at ease with the concept — and still have functionally nothing when asked to generate the mechanism from scratch. Recognition and generation aren't the same cognitive operation, and the illusion lives entirely in the gap between them.
Why This Matters More in a Boardroom Than a Toolshed
It would be a tidy, low-stakes finding if it stayed confined to zippers and cylinder locks. It doesn't. In 2013, Philip Fernbach, Todd Rogers, Craig Fox, and Steven Sloman extended the same method to something with real consequences, publishing "Political Extremism Is Supported by an Illusion of Understanding" in Psychological Science. Their subjects rated how well they understood specific policies — a flat tax, a cap-and-trade carbon system, sanctions on Iran — with the same initial confidence people bring to zippers. Then they were asked to explain, mechanistically, step by step, how the policy would actually produce its intended effect.
The explanation task did something the original study's zipper-writers didn't get: it moderated people's political positions. Not because anyone changed their values. Because the act of trying to generate the causal chain — this policy leads to this effect, which leads to this outcome — exposed how much of their confident position had been built on the feeling of understanding rather than an actual model of the mechanism. Fernbach and colleagues put the finding plainly: people's mistaken sense that they understand the causal processes underlying policies contributes directly to political polarization. Notably, simply asking people to list their reasons for a position didn't produce the same softening — reasons are recall, and recall doesn't puncture the illusion. Only mechanism does. You can list five reasons you support a policy all day without ever discovering you can't actually trace how it works, because reciting reasons and generating a causal chain draw on entirely different cognitive machinery. The Dunning-Kruger effect gets misquoted constantly as "stupid people think they're smart," but this is the more precise and more universal cousin of that story — it's not about intelligence at all. It's about a specific blind spot that fluent recognition creates in everyone, regardless of how sharp they otherwise are.
AI Didn't Fix This. It Built a Faster Delivery System for It
The newest and most uncomfortable extension of this research arrives from the last two years, and it should worry anyone treating a chatbot as a substitute for actually learning something. A 2024–2025 study running 102 university students through three conditions — one group consulting an AI assistant before explaining a topic, one reading a plain text explanation, one control — found that the AI-assisted group showed the largest gap between their initial confidence and what they could actually produce afterward. The AI's fluent, complete-sounding output didn't transmit understanding into the student's head. It transmitted the feeling of having understanding, which is precisely the illusion Rozenblit and Keil described in 2002, now manufactured on demand by a machine that's very good at sounding like it already did the explaining for you.
This is the mechanism, not a vague complaint about laziness: a good explanation, read passively, satisfies the same recognition system that made you confident about the zipper. You never generated the causal chain yourself. You watched someone else generate it, fluently, and your brain filed the experience under "I understand this" instead of "I watched this be understood." The two feel identical from the inside, right up until someone asks you to reproduce it without the source in front of you.
So Actually — the Cure Was Always the Same One-Step Test
Here's what makes this finding oddly hopeful instead of just damning: the illusion is trivially easy to puncture, once you know where to press. Rozenblit and Keil's method wasn't complicated. It was one instruction — explain it yourself, in writing, from the beginning, without looking anything up — and the confidence collapsed on its own, no argument required. You don't need someone to tell you that you don't understand cap-and-trade, or the tax bill you have strong opinions about, or the technical concept you just had an AI walk you through. You need to try writing the mechanism out loud, alone, and watch where it stalls. The stall isn't failure. It's the most honest piece of information you'll get about what you actually know.
The next time you notice yourself certain about how something works — a policy, a device, a decision someone else made that seems obviously wrong to you — try the one-sentence test before the certainty hardens into a position: could you explain the mechanism, out loud, to someone who'd immediately notice if you faked it?