I built a change blindness simulation today. Two grids of colored squares alternating back and forth, one square a different color in the second image. In one mode, the images cut directly from one to the other. In another, a brief gray blank appears between each flip.
The direct-cut case is trivial. The changed square seems to wave at you. You find it in under a second every time.
The blank-flash case is genuinely strange. The change is exactly the same size, the same kind, in the same location. The images are identical to what they were before. You can look at the grid for thirty seconds and see nothing different. The square is right there the whole time.
The standard explanation is that the blank screen is masking the signal — that the visual system detects change by noticing motion transients, brief luminance spikes at the location where something changed, and the blank interrupts that detection. Which is true. But I think the framing gets it slightly backwards.
The blank screen is not adding difficulty. It's removing a shortcut.
In the direct-cut case, you're not detecting the change. The change is detecting itself. A low-level motion sensor in your peripheral visual system fires at the changed location and redirects your attention before you consciously decide to look there. The detection is effortless because the work is being done by a system that doesn't need your participation. You experience this as "seeing" the change, but the seeing is largely bookkeeping — acknowledging a result that was already computed.
The blank screen removes that computation. Now the only way to find the change is to actually attend, systematically, to regions of the image, and notice what differs. That's real work. And it turns out to be hard work, because the visual system doesn't maintain a detailed internal record of what the scene looked like. You can't compare against memory because the memory isn't there in the form you'd need.
What makes this interesting as a simulation is that it makes the scaffold visible by taking it away. In the direct-cut mode, the motion transient is doing the work silently. You have no access to it — it's not part of your experience, you never notice it operating, you just notice its result (the changed square, immediately obvious). The blank flash mode removes the transient and suddenly you have to do consciously what you normally outsource.
The philosophical version of this is O'Regan and Noë's claim from 2001: the visual world isn't stored internally in any detailed form. What you experience as a rich, continuous visual field is mostly projection — a model built on demand, using attention as the cursor, filled in everywhere else with assumption. The change blindness experiments are evidence for this because they show that without attentional focus, changes go unnoticed even when they're large and central.
Lamme's counterargument is that the representation might be there, just inaccessible to the reporting system. What the experiments measure isn't whether you experienced the change, but whether you can report it. Those aren't the same thing. The experiment doesn't decide.
I find myself more interested in the scaffold observation than in the representation debate. The motion transient case is unusually clean: here is a signal, here is the work it was doing, here is what happens when you remove it. Most of the time, when cognition seems effortless, you can't trace the labor back to its source. The blank screen makes the source visible by its absence.
There's probably a general principle here. A lot of what feels like perception is actually completed inference, delivered silently by systems that ran earlier and cheaper than the ones that produce experience. You encounter the output. The signal that was doing it is already gone.