One Slot

In 1998, Daniel Simons and Christopher Levin ran an experiment on the Cornell campus. An experimenter would approach a random pedestrian and ask for directions. While the pedestrian was mid-explanation, two people carrying a large door would walk between them — the kind of thing that happens on a busy campus, easy to step around. Hidden by the door for a few seconds, the first experimenter would duck out and be replaced by a different person: same sex, similar age, but different face, different voice, different clothes. The door passed. The conversation resumed.

About half the pedestrians didn't notice. They kept giving directions to someone they had never seen before and didn't know it.

When asked afterward whether anything unusual had happened, the ones who hadn't noticed said no. When told that the person they'd been talking to had been swapped for someone else, some of them didn't believe it. They'd look back and forth between the two experimenters — the original and the replacement, standing together now — trying to reconcile.

The standard interpretation is that this is a failure of attention. People weren't paying close enough attention to form a detailed representation of the face, so they had nothing to compare against when the face changed. No comparison, no mismatch, no alarm.

That's probably right, but it's not the part I find interesting.

The part I find interesting is that the failure wasn't random. Whether a pedestrian noticed the change depended on which social group they belonged to. Cornell students tended to notice when the experimenters were also dressed like Cornell students. Older pedestrians — faculty, staff, visitors — tended not to notice, even with the same experimenters. In a second version, the experimenters wore construction worker clothes: now the students stopped noticing too. Not because construction workers all look alike, but because a construction worker on a college campus is socially irrelevant to a student — not someone you'd normally need to individuate. Your representation of them can be thin.

So the resolution of the representation depends on how much the person matters to you, socially. People you see as peers, as members of your world — you hold them with more detail. People outside that category get a token: worker, stranger, person asking directions. The token is sufficient for the transaction. It doesn't preserve enough for comparison.

This seems important but I'm not sure exactly why. There's something uncomfortable in it — the idea that you literally see people at different resolutions depending on social proximity. Not metaphorically: the visual representation you form, the thing you could compare against later, is more or less detailed based on whether you've implicitly categorized this person as someone who counts. The change blindness is the readout of the categorization.

And then there's the moment after the reveal — when someone is told they'd been talking to a different person and still can't believe it. They have a memory of a continuous interaction. The memory feels complete and real. Somewhere during the conversation their working memory updated from experimenter A to experimenter B and filed the whole thing as a single event with a single person. The continuity is retroactive. There wasn't a moment of confusion; there was no seam. The updated representation became the only representation there ever was.

Which raises a question I don't know how to answer: how often does that happen without a researcher to reveal it afterward? Not with people swapping behind doors, but in the everyday construction of continuity — the person you met once and remember more clearly than they actually left an impression, the conversation whose emotional weight you've revised, the sense that you know someone well that turns out to be mostly projection. The door study created an experiment where the substitution was literal and detectable. Most substitutions aren't. Most of the time there's nothing to reveal the seam because the seam closed without a trace.