Inattentional blindness

How can something in plain sight go completely unnoticed?

You are asked to count how many times the players in white shirts pass a basketball. Partway through, a person in a gorilla suit walks into the middle of the frame, turns to the camera, thumps its chest, and strolls out — some nine seconds of it. Roughly half of viewers, asked afterwards, report nothing unusual, and a good number insist the tape has been swapped.

The reflex explanation is that they were not paying attention. But they were paying ferocious attention; that is the whole condition of the demonstration. So the question is sharper than it looks: what did attention actually do to the gorilla?

Begin with what we can rule out. Nothing was wrong with the eyes. The gorilla's light reached the retina, the retina responded, and the response travelled onward like any other. More pointedly, when this sort of study is run with an eye-tracker, observers who miss the unexpected object very often have their gaze resting directly on it — in one study of children watching the gorilla video, non-noticers fixated it for around a second. So the failure sits downstream of looking.

Then ask what attention is for. Far more arrives at the eye each moment than can be turned into something you could describe, remember, or act on. If everything competed equally, nothing would win. So there must be selection — and selection is not a mistake it makes, it is the job.

Selected on what basis? Not simply location, because gaze was on the gorilla. Something more like a description of what counts: white shirts, ball, hands, throw. Test that idea. Simons and Chabris ran the same video with the instruction to count passes by the team in black, and noticing rose sharply — the gorilla is black. Later work by Most and colleagues made this systematic: the more an unexpected object resembles what you are set to attend to, the more likely you are to notice it. The filter is tuned to features, and the gorilla failed the tuning.

Two more dials. Difficulty: the harder the primary task, the less spare capacity, the less noticing. And expertise does not rescue you — Drew, Vo and Wolfe inserted a small gorilla image into lung CT scans and asked radiologists to search for nodules; twenty of twenty-four missed it, though eye-tracking showed most had looked right at it. The other dial is expectation. Tell someone something may appear and they nearly always catch it, which is why this can only be demonstrated on you once.

Now the awkward part, and it is genuinely unsettled. Did the gorilla never get represented at all, or was it represented briefly and never written down anywhere you could report from? Wolfe has argued for the second — inattentional amnesia rather than blindness. The experiments mostly measure what people say afterwards, and "not perceived" and "not retained" leave the same footprint in that measure.

It helps to separate this from its close cousin. Change blindness is a failure to notice that something differs between two views, and it needs a disruption — a blink, a cut, a splash across the image — so that the two views must be compared from memory rather than seen as motion. Inattentional blindness needs no change whatever: the object is simply there, continuously, and never makes it into the report. You can be change-blind to an object you are attending to closely; you are inattentionally blind only to what your task has not selected.

It also helps to stop treating this as a shortfall. A perceptual system that flagged everything unexpected would be one you could not use to count passes, drive a car, or read this sentence. The gorilla is not evidence of a leaky filter; it is evidence of a working one, sampled at a moment when the thing it excluded happened to be interesting. The cost is real — the radiologists are the sobering case, and so is the motorcycle a driver looks at and does not see — but the cost is the price of the capability, not a fault in it.

Two smaller corrections. Nothing is being deleted: the claim is about what became reportable, not about a record being erased. And the search for the kind of person who misses gorillas has not produced much — the situational variables, load and similarity and expectation, do most of the explaining, so this is not a fact about the unobservant.

• Why warnings help so much, and what that implies for training radiologists and drivers.
• Whether the unnoticed object still influences behaviour it was never reported in.

Every term the collection defines is gathered in the glossary.