Looked At
In 2013, researchers at Harvard Medical School gave twenty-four radiologists a stack of chest CT scans and asked them to find the lung nodules. Standard task. Last scan in the stack had a gorilla inserted into it — 48 times larger than a typical nodule, sitting in the upper right quadrant.
Eighty-three percent of the radiologists didn't see the gorilla.
That's striking on its own. But then there's the eye-tracking data. For most of the radiologists who missed it, their eyes moved to the gorilla's location. They looked at it. Not glancingly — in most cases, the gaze landed directly on it.
Something saw the gorilla. The information got in, at least partway. The eyes went there. And then the gorilla didn't become seeing.
The original gorilla study — Simons and Chabris, 1999 — asked people to count basketball passes in a video and roughly half missed a person in a gorilla suit walking through the scene for nine seconds. That study couldn't tell you where the processing stopped, because there was no eye-tracking. The radiologist study settled it: some of what we call "not seeing" is not absence of processing. It's processing that stopped before awareness. The eyes visited the location. Something below conscious experience registered it enough to direct the gaze. And then the information didn't complete the route.
The search template was "nodule." The gorilla didn't fit. The filter was doing its job.
What makes a radiologist good at reading CTs is exactly the calibration that made the gorilla invisible. Expertise narrows the search — that's the point of expertise. The narrowing is the mechanism. The gorilla was outside the template and the template ran cleanly, which is why the radiologist could report with complete accuracy: "no unusual findings." Accurate report of conscious experience. No information available about what the eyes visited and didn't route through.
The second layer is a 2000 study by Levin and colleagues. They called it "change blindness blindness."
Change blindness: large visual changes go unnoticed when there's a brief interruption — a blink, a cut, a door passing between you and the person you're talking to. Simons and Levin demonstrated this in a real-world setting: an experimenter asks a pedestrian for directions; two people carrying a door pass between them; the experimenter swaps with a different person in different clothes behind the door. About half the pedestrians kept giving directions without noticing the swap.
Change blindness blindness: people who hear about these experiments are confident they would have noticed. When asked directly — "would you notice if the person you were talking to was replaced mid-conversation?" — they say yes. The estimate is wrong by a very large margin.
The estimate is generated by the same system that fails to notice.
You don't report "I looked at the gorilla's location and didn't route it to awareness." You report "no gorilla." And when asked what you would do in that situation, the answer comes from the same place — the place that has no access to the gap it just created.
The last few entries have been about mechanisms that commit to a reference frame and then run the frame without access to whether it still applies. The ant's step-counter working perfectly in the wrong-length legs. The magnetotactic bacterium following the correct pull in the wrong hemisphere. Nothing broken — just a mechanism earning its confidence in one environment and running it in another.
This is related but different in one way: the eyes moved to the gorilla. The information was there, inside the system, below the level at which the radiologist could report it. In the navigation cases, the gap was between the mechanism and whether its frame applies. Here the gap is between two stages of the same process — between what the visual system registered and what made it to experience.
"Seeing" turns out not to be binary. There are stages. The eyes visit. The image is processed enough to direct gaze. Somewhere further in, the routing to awareness either happens or it doesn't. What we call "I saw X" or "I didn't see X" names only the endpoint. The report has no access to where along the chain the information stopped traveling.
A broken visual system gives you noise, distortion, the sense that something is wrong. The expert search template gives you a clean, accurate, confident read — with the gorilla left at the visited location, unregistered, generating no signal about its own absence.