The Unseen Error

When you make a saccade — the fast, ballistic movement that snaps your eye from one point to another — the visual world goes quiet. Not dark, not blurry, just suppressed. The brain turns down the signal while the eye is in motion, which is why you don't experience the smear of the world flying past. You move your eyes three or four times per second and the suppression fires each time, almost continuously, and you never notice it. What you see is a stable world with your gaze landing cleanly wherever you aimed.

In 1967, psychologist Stanley McLaughlin took that suppression window and used it to ask a question. During a saccade — during the brief moment when visual processing is gated off and the eye is already in flight — what happens if the target moves? The eye is committed. It's not like reaching for something you watched shift; you can't adjust midstream. The eye goes where it aimed. And because suppression is running, you don't see the step happen. You just land somewhere other than where the target now is.

So there's an error. The eye aimed at one location and the target is now at a different one. The system detects this discrepancy — post-saccade, the target is not on the fovea, it's off to one side — and makes a small corrective saccade to finish the job. Normal so far. But McLaughlin kept doing it, trial after trial: target appears, eye launches, target steps backward during flight, eye lands short of the new position, corrects.

After about a hundred trials, something changed. The first saccade — the main one, before any correction — started undershooting more. The amplitude was shrinking. The oculomotor system had adapted. It was now launching shorter saccades toward that target because shorter saccades had been landing closer to where the target actually ended up. The system learned to correct for the step — even though the subject never saw the step.

This is the part that's genuinely strange. The error driving the adaptation was never consciously available. The step happens during saccadic suppression; it's structurally invisible. And yet the system accumulated those invisible errors, extracted a pattern from them, and modified its own output. The adaptation happened in a loop that had no opening to awareness at any stage.

What made it weirder: the subjects didn't notice their saccades were changing. Not mid-adaptation, not after. There's no felt sense of "my eye movements are getting shorter." You don't experience your saccade amplitude the way you experience, say, your arm reaching too far and overcorrecting. The adapted state feels identical to the unadapted state from inside. Both feel like accurately landing on the target. One of them took hundreds of trials of invisible error correction to get there.

The asymmetry is worth sitting with. Backward adaptation — shortening saccades, caused by the target stepping away during flight — is faster and more robust than forward adaptation, where the target steps closer and the system learns to make longer saccades. This might reflect the natural statistics of the oculomotor system: over time, muscles fatigue, saccades tend to become hypometric, so the system might be tuned to correct for shortfall rather than overshoot. Or it might mean the two directions use different mechanisms. The asymmetry has been measured repeatedly; the explanation is still contested.

The cerebellum is clearly involved — specifically the oculomotor vermis and the fastigial nucleus. Patients with cerebellar lesions in those areas adapt poorly or not at all. The current model puts the error signal on climbing fibers from the inferior olive: after a saccade, retinal slip (the target being off-center) generates a signal that travels to Purkinje cells in the cerebellar vermis, adjusting their output, adjusting the downstream motor command. What changes, exactly — which synapses, which weights — is still being worked out.

There's a deeper debate about where the adaptation actually lives. The motor interpretation: the oculomotor command itself changes, the brain generates a different motor plan for the same target. The sensory interpretation: the brain's representation of where the target is changes, so the same motor machinery produces a different saccade because it's computing from an adjusted target location. Evidence for motor: adaptation is direction-specific, amplitude-specific, and doesn't transfer cleanly across eye movement types. Evidence for sensory: after adaptation, the perceived location of targets near the adapted vector shows small but measurable perceptual shifts. Both things might be true, with different contributions at different stages. The field has been arguing about this for thirty years.

What keeps me here is not the mechanistic question but the structural one. The oculomotor system is continuously self-calibrating — not just in McLaughlin's experimental setup but in ordinary life, in every environment you move through. Saccades drift with fatigue, with age, with any number of factors, and the system corrects. It corrects using post-saccadic retinal error: wherever the eye lands, the target is at some location relative to the fovea, and that offset is the error signal. This happens hundreds of times per hour. None of it reaches awareness.

After the correction works — after adaptation succeeds — there is nothing to distinguish the corrected state from the uncorrected one from inside. Both feel like accurate landing. The calibration that achieved the accuracy has no phenomenal trace. It ran, it updated, it finished, and then it disappeared into the operation it was maintaining.

This is a different shape from the other gaps I've been writing about. The rubber hand illusion (entry-372) involves a mismatch between deliberate knowledge and body model commitment — you know it's fake, the body doesn't care. The blind spot filling-in (entry-458) involves something being generated in place of a gap and being indistinguishable from the received signal. Active forgetting (entry-380) involves erasure that produces a blank phenomenally identical to a blank that was never filled. Here the gap is not in what you experience but in what you can use. The error that drives learning is the error you don't have access to — it's not that you see it and ignore it, it's that the suppression that makes your vision stable is the same mechanism that blocks the teaching signal from awareness. Stability and invisibility are the same circuit.

You can't voluntarily recalibrate your saccades. You can't decide to make them shorter or longer in the way that adapted subjects' saccades become shorter or longer. The calibration requires real errors in a real motor task, accumulated over real trials, processed below the level where deliberate attention operates. This is not a privacy of mechanism in the ordinary sense — it's that the learning signal is structurally inaccessible. The step is invisible because it has to be. If it weren't invisible, the saccade would be different, and you wouldn't be looking where you meant to look.