The feeling of understanding#

How to use this lesson#

Read at your own pace; no presenter is needed. Allow about 40 minutes, and have somewhere to write. This lesson asks you to produce a text and then examine the text you produced.

You do not need to own a zip, hold one, or look one up. A full explanation of how a zip works appears later in this deck, at the point where the activity has finished with it.

The people in this lesson are fictional.

Where Week 1 left us#

Week 1 introduced the examiner: the person deciding what a performance demonstrates. It separated an observation from an inference, insisted that the conditions an answer was produced under belong in the record, and used Clark and Chalmers to argue that consulting a reference does not automatically disqualify a performance.

It ended on a problem. A correct sentence did not tell us what the speaker could produce independently.

The obvious next move, and its problem#

If a repeated sentence is thin evidence, ask the person to explain. An explanation has parts: steps, causes, things that depend on other things. There is more in it to inspect.

This week tests that move. We will find that an explanation is a useful probe — a way of generating evidence — but not a proof. A fluent explanation can still leave out the relations that make the thing work.

Two feelings that are easy to confuse#

Recognition is knowing that you have met something before. You recognise a zip instantly, from across a room, in the dark, by sound.

Familiarity is the broader sense that something is well known to you and easy to follow. Week 1 met it in a Wi-Fi explanation that “made sense” on reading.

Neither feeling reports on what you could produce. This week asks how you would ever find that out about yourself.

The protocol we are going to run#

Rozenblit and Keil (2002) asked people to rate how well they understood an everyday object, then write a detailed explanation of it, then rate themselves again. A question about the mechanism followed, and then a third rating.

We are going to run an adaptation of that on a single object: the zip. The object is theirs. A zipper was one of their eight test items.

Rozenblit & Keil (2002), “The misunderstood limits of folk science: an illusion of explanatory depth,” Cognitive Science 26(5), 521–562. An empirical finding. Full reference and link at the end.

What the attempt does#

Whatever you wrote, it is now a text with edges. Writing forces an explanation out of the state in which it felt complete, and into something that has a beginning, an end, and gaps you can point at.

In Rozenblit and Keil’s studies, self-ratings dropped after this step. Compare your Rating 2 with Rating 1 now. A fall, a rise and no change are all informative, and we will come back to what each one means.

The expert description: the parts#

At this point their protocol supplied a brief expert description. Here is ours, over three slides.

A zip has two tapes: the strips of fabric sewn to the garment or bag. Along the inner edge of each runs a row of teeth, also called elements, either as separate pieces or as one continuous coil. The slider rides on both rows, moved by a pull tab, and a top stop and bottom stop keep it and the tapes from leaving the ends.

Technical background on these three slides: Wikipedia, “Zipper.” Link at the end. It grounds the worked example; it is not evidence for any claim about understanding.

The expert description: closing the zip#

Inside the slider is a Y-shaped channel. Run the slider in the closing direction and the two arms of the Y are its inlets, the single stem its outlet: the rows converge and are pressed together one pair at a time.

Each tooth carries a small projection on one face and a matching recess on the other. The projection is slightly larger than the recess it enters, so the tooth springs apart and closes over its neighbour, like the jaw of a hair pin.

The expert description: holding, and opening#

That spring is an elastic deformation: the tooth returns to its own shape afterwards, so the joint is a snug fit that holds once the slider has moved on. Interlocked projections and recesses resist being pulled apart both across the zip and along it.

Run the slider the other way and the Y is reversed. The stem becomes the inlet, the arms become the outlets, and the wedge between them prises each pair of teeth apart in turn.

Worked response: how one rigid object does both#

The slider needs no moving parts because the work is done by a fixed shape and the direction of travel through it. The same Y is a funnel from one end and a splitter from the other, so the teeth are brought together or forced apart depending on which way they enter it.

The teeth do the rest. They are not held by the slider once it has gone; they are held by their own interlocking geometry.

Why a sideways tug does not open it#

Pull the two tapes of a closed zip apart sideways and you load every interlocked pair at once, in the direction that projection-in-recess geometry is shaped to resist.

The slider does something different. It reaches the teeth one pair at a time, so the whole of the force falls on a single joint instead of being shared along the chain. The same fastening, undone in series rather than resisted in parallel.

One boundary condition worth naming#

A boundary condition is the circumstance in which a mechanism stops working. This one follows directly from the geometry above rather than from any extra fact.

The hold depends on every tooth keeping its own shape and having an intact neighbour to grip. Bend one tooth, or lose one, and at that point nothing is interlocked — so the chain parts there even though the slider has passed.

Worked response: an annotated explanation#

Fictional. Theo wrote: “The slider pulls the two sides of teeth together and they interlock like fingers. Pulling it back down separates them again.”

Every word of that is true. Annotated, it is thin. Missing relation: nothing about what holds one tooth to the next. Missing cause: “interlock” names the outcome rather than the projection entering the recess. Missing dependency: no Y-shaped channel, so nothing in the explanation makes the two directions of travel different.

The limitation this week exposes#

A fluent explanation can conceal missing relations, missing causes, boundary conditions and dependencies. Fluency is a property of the telling, not of the model behind it.

Notice what that does not say. It does not say explanations are worthless, or that everybody fails, or that a confident speaker is probably bluffing. It says the telling and the model can come apart, so the telling on its own cannot settle the question.

If your explanation held up#

Some readers will have written most of the description already. Take that seriously rather than explaining it away: you produced, unaided and on demand, a causal account that survived comparison with a technical one.

Then say precisely what it establishes. It establishes that you could produce this explanation, of this object, in writing, at this moment, with nothing in front of you.

It does not establish that you would survive a probe you had not anticipated, or that the same holds for any other mechanism.

Naming it: the illusion of explanatory depth#

Rozenblit and Keil call this the illusion of explanatory depth: people believe they understand how everyday things work in more mechanistic detail than they can actually produce, and that belief shrinks as soon as they try to explain.

In their Study 1, sixteen graduate students at one university rated four items each on the 1 to 7 scale, five times in all; we have used the first three ratings.

Rozenblit & Keil (2002), Study 1. Empirical finding. A repeated-measures ANOVA found the effect of time significant, F(4, 56) = 16.195, p < .001. The rating after explaining, the rating after the diagnostic question, and the re-rating of prior knowledge after reading an expert description were each lower than the first rating, p < .002.

What that study does and does not support#

It is an empirical finding, shown across twelve studies in the one paper. It is not a theory of understanding and not a settled consensus. Study 1 alone is sixteen graduate students at one institution; the paper’s broader claims rest on the other studies too.

Now notice what was measured. The thing that fell was a self-rating. The study shows people revising their estimate of their own understanding downwards.

Two different things your session produced#

Only one of these is independent of your own account of yourself.

Our critique: a rating is not a performance#

This point is ours, not Rozenblit and Keil’s. If Rating 3 fell below Rating 1, you have evidence that your confidence changed. You do not yet have evidence about your performance.

The performance evidence is the other artefact: the annotated text. A drop in the number is a report about a feeling, and Week 1 already warned us that feelings of familiarity are not observations of what someone can do.

And you have not replicated anything#

Running this on yourself, once, with no control condition and nobody to compare against, is not a replication of Rozenblit and Keil. You also knew the design before you started, which is exactly what their participants did not.

Treat your three ratings as the record of one person’s session, useful for examining yourself. Their studies are the evidence for the general claim. Your session is not.

The paper-swap: the paired version#

Fictional. Nadia names a topic she genuinely knows: the thing she does at work, an instrument she plays, a process she has run a hundred times. Theo researches it freely for ten minutes, using whatever he likes, then explains it back to Nadia with his sources closed.

Nadia then probes. She is not marking Theo and she is not trying to catch him out. She is finding where his explanation stops being load-bearing.

What Nadia’s probe can do that yours cannot#

Nadia has something no self-check has. She knows what a complete answer contains, and she did not learn it from the page Theo read five minutes ago.

Her probes come from the real structure of the topic: the exception, the case where the usual account breaks, the step everyone skips. If Theo’s explanation survives those, then somebody who knows the material was satisfied by it. That is an observation, and a stronger one than Theo’s own sense of how it went.

What the solo route loses#

State the loss exactly. With no second person, there is no independent check on whether your explanation actually satisfied someone who knows the material.

You grade your text against a description you have only just read, so you cannot tell whether you would have produced the missing step unprompted. And every probe you meet is one you thought of yourself, so you are never surprised. A later week in this course returns to what an unforeseen question reveals.

Keep a short evidence record#

Write four lines and keep them; later weeks will ask you to compare against this one.

1. Ratings: 1, 2 and 3, with the conditions — written, alone, no sources.
2. Observation: what your explanation actually contained.
3. Interpretation: the specific thing it shows you can produce.
4. Limit: the specific thing it does not show.

“I don’t really understand zips” is not a limit. “My text never said what holds two teeth together” is.

Explaining as a probe, not a proof#

An explanation is better evidence than a repeated sentence, because it has parts that can be missing and you can put your finger on them.

It is still only a probe. It tells you what this person produced, in writing, on this occasion, about this object. A confident telling and a complete model are different things, and the telling is the only one of the two you observed.

What this leaves for Week 3#

Explaining exposed gaps, but the conditions were generous. The reader had unlimited time, no interruption, and a lifetime of handling zips to draw on.

So Week 3 changes the conditions. Study something, lose access to it, then rebuild it. Its question is what becomes visible when the answer is no longer in front of us, and the claim to test is that reconstruction is harder to fake than explanation.

Read further#

Sources and your next step#

Technical background for the zip: Wikipedia, “Zipper”. An encyclopaedia article is adequate background for a mechanism of this kind; it is not evidence for a claim about learning or understanding.

Nadia, Theo and their exchange are fictional teaching examples, not reported data.

For the full written activity and a record to keep, open the Week 2 investigation. The lecture notes give a continuous reading version.