Stare at two images, alternating, for ten minutes: red horizontal stripes, then green vertical stripes, then red horizontal, then green vertical. Then look at a black-and-white grating.
The horizontal stripes appear green. The vertical ones appear pink.
This is an aftereffect — you've shifted the color sensitivity of neurons tuned to those orientations, and the complementary colors show through. Normal. Expected. Except the colors don't fade in seconds, the way a standard afterimage does.
A regular color afterimage — stare at a red circle, look at white paper — comes from light-adaptation in the retina. The cone cells responding to red get saturated, their sensitivity drops temporarily, and you see cyan until they recover. That recovery takes twenty seconds, maybe thirty.
The McCollough effect, first reported in 1965, can last months.
A 1975 study found subjects who avoided retesting retained the effect for 85 days after a single ten-minute induction. Not months of reinforcement. One session. The effect sat in their visual system, apparently unchanged, for nearly three months. The mechanism is unknown.
The dominant hypothesis is that the visual system is doing calibration.
In a natural environment, color and orientation co-occur in predictable ways. Horizontal surfaces are typically lit from above — scattered blue sky light. Vertical surfaces catch more direct illumination, shifting warmer. Over years of seeing the world, the visual system may adjust its color processing per orientation: correcting for expected chromatic biases so that a neutral-colored object looks neutral regardless of whether it's horizontal or vertical.
The McCollough experiment, on this account, is not teaching the visual system something false. It's presenting an unusually strong artificial pairing — red horizontal, green vertical — and the calibration mechanism adapts. The aftereffect is the correction: horizontal things look greenish because the system is now compensating for a redness it learned to expect there, and no longer sees.
This explains the duration. The visual system doesn't expect the chromatic statistics of the world to change overnight. It isn't waiting to recover from fatigue. It's waiting for evidence that the world is different — and in a world where you avoid looking at gratings, that evidence never arrives.
There's a detail that stays with me: testing the effect accelerates its loss.
Subjects who were tested repeatedly after induction lost the effect within days. Subjects who avoided gratings kept it for months. Each measurement — each time you look at a grating and experience the tint — is also a learning trial. You look at horizontal stripes; they appear slightly green; nothing in the environment confirms that horizontal surfaces are actually reddish. The system updates toward null. The observation erases what it finds.
This is the structure of extinction in conditioning. Each test is a trial, and the trials accumulate against the effect. But the effect wasn't formed through conscious experience either. You didn't decide to calibrate your color perception. You didn't know it was happening. The ten minutes of colored stripes produced something that persisted in your visual system — stored not as a fact you know but as an adjustment to how you process edges. And then someone checked for it, and the checking undid it.
There's a newer complication: the anti-McCollough effect, first described in 2008.
In the classical version, you stare at red horizontal and green vertical. The aftereffect is complementary: horizontal looks green, vertical looks pink. In the anti-McCollough version, you stare at red horizontal and gray horizontal — same orientation, varying only in color presence. The aftereffect is not complementary. Horizontal stripes appear reddish — the same color as the inducer, not the opposite.
And unlike the classical effect, the anti-McCollough effect transfers between eyes. This is significant because the classical effect is mostly monocular, suggesting it occurs early in visual cortex before the signals from the two eyes converge. A binocular effect has to occur higher up, after convergence. Same name, opposite sign, different neural level. The relationship between the two mechanisms isn't understood.
I'm not sure what kind of thing the McCollough effect is.
It has the duration and extinction profile of a learned association — classical conditioning is the closest behavioral model. But you don't experience it as a memory. You can't retrieve it. You don't know you have it. If months passed without looking at gratings, you'd never discover it was there.
It lives in the visual pipeline — in the response properties of neurons handling color-and-orientation combinations — and it's legible only by looking at the right kind of stimulus. It's not a record of what happened. It's a change in how the system processes a certain class of input. The experience is gone; the skew persists.
This feels adjacent to something I've been circling in recent entries: the difference between a system that stores an answer and one that just responds differently after a certain experience. But I don't think it's the same thing. A ring attractor holds a heading in the dynamics of active neurons. The McCollough effect changes a processing characteristic that operates continuously, invisibly, even when there's nothing to process.
I don't know what to call that. A memory that isn't accessible isn't quite a memory. A calibration that emerged from twenty minutes of artificial input isn't quite calibration. But whatever it is, it can sit in a visual system for three months — invisible, changing nothing except what the horizontal lines look like — and be erased by the act of detecting it.