Change Blindness: The Door Study

A door blocked a pedestrian’s view for one second; when it passed, a different man stood behind it. In Simons and Levin’s 1998 experiment, 8 of 15 pedestrians continued giving directions without reporting the substitution. The replacement had different clothes, a different voice, and stood about 5 centimetres taller.

What the door interrupted

The researchers approached pedestrians on the Cornell University campus with a map. After 10 to 15 seconds of conversation, two people carrying a door passed between the pedestrian and the researcher. Hidden by the door, the first researcher joined the carriers while the second took his place.

The visual interruption removed the motion signal that would ordinarily announce a change. The pedestrian retained the scene’s meaning, “a stranger needs directions,” without necessarily retaining enough facial, vocal, or clothing detail to compare the two men.

Attention was present throughout. What failed was the comparison between the scene before the door and the scene after it.

The numbers behind the trick

Experiment	Participants	Clear detections	Result
Ordinary clothing	15	7	47%
Construction clothing	12	4	33%

In the second experiment, the same researchers dressed as construction workers. Only 4 of 12 students clearly reported the switch. Three more claimed to have noticed after direct questioning but could not describe any difference between the men.

The paper’s interpretation was that “construction worker” displaced individual identity. Category survived the door; personhood did not.

What is contested

Change blindness is established, but the door study’s social-category explanation rests on 27 participants and two linked manipulations. Experiment 1 confounded age with presumed group membership; Experiment 2 changed clothing, occupational appearance, and expectations together.

The deeper inference is also disputed. Missing a change does not prove that the brain stored no detailed representation. The detail may have been encoded but not retrieved, retained but not compared, or overwritten during the interruption. Simons and Ambinder’s 2005 review treats detection as a chain with several possible failure points, not a single hole in vision.

Why this crosses realms

The door study looks like concept information theory enacted on a pavement. Perception compresses a high-bandwidth scene into the variables needed for the present task. Identity, shirt colour, and voice can disappear while the lower-resolution message, “continue giving directions,” survives.

The same trap appears in concept mechanistic interpretability. A model can preserve its output while its internal route changes; an observer watching only the label may never notice the substitution. Both cases separate functional continuity from structural continuity.

An open question

What kind of monitoring system could detect identity-changing substitutions without recording every detail, and would that system still count as attention rather than surveillance?

Key Sources

Daniel J. Simons and Daniel T. Levin, “Failure to Detect Changes to People During a Real-World Interaction” (1998), Psychonomic Bulletin & Review 5: 644–649. DOI
Ronald A. Rensink, J. Kevin O’Regan, and James J. Clark, “To See or Not to See: The Need for Attention to Perceive Changes in Scenes” (1997), Psychological Science 8: 368–373. DOI
Daniel J. Simons and Michael S. Ambinder, “Change Blindness: Theory and Consequences” (2005), Current Directions in Psychological Science 14: 44–48. DOI
Daniel T. Levin, Nausheen Momen, Sarah B. Drivdahl, and Daniel J. Simons, “Change Blindness Blindness” (2000), Visual Cognition 7: 397–412. DOI

Abhishek's take

I read the door as a warning about stable labels. A dashboard can retain the same category while the item underneath has changed beyond recognition. The dangerous miss is not failing to see everything; it is assuming that an unchanged task means an unchanged world.

Abhishek