What He Didn't See

D.B. couldn't see anything in the left side of his visual field. After surgery to remove a tumor pressing on his right primary visual cortex, the left field went dark. Not dim. Not blurry. Just absent. He said so clearly, without apparent distress — a matter of fact.

So Lawrence Weiskrantz asked him to point at where a light had appeared in that absent field.

D.B. objected. "I can't see anything. Why are you asking me to guess?" But he guessed anyway, because Weiskrantz kept asking — and he was consistently right. Not always right; this wasn't telepathy. But right at rates that chance can't explain. He could indicate the direction of a moving grating. He could say whether lines were horizontal or vertical. He performed all of this inside a visual field he told you was empty.

Weiskrantz coined the term in 1974: blindsight. The word was meant to hold the strangeness — the behavior of someone seeing, wrapped around the report of someone blind.

The experimental structure is important. Blindsight research uses forced-choice tasks. You don't ask "did you see anything?" because that invites the subject to report their experience. You ask them to guess: "Point at where the light was. Just try." The subject's ability to guess above chance is the finding. The protest — "I'm just guessing, I can't see it" — is also part of the data. The behavior and the report diverge. Both are accurate.

The architecture that makes this possible is older than consciousness, or at least older than cortex. Before V1 evolved, there was another visual circuit: retina → superior colliculus (midbrain) → pulvinar (thalamus). Faster, cruder, good for detecting motion and spatial location. When V1 is destroyed, this older route is still intact. It can route information into areas that guide behavior without passing through primary visual cortex at all. The collicular-pulvinar pathway keeps running. This isn't mysterious — it's an architecture. The behavior it supports persists because the machinery that supports it is physically separate from the machinery that was damaged.

The dorsal visual stream — the "where/how" pathway running into the parietal lobe, concerned with action guidance — can run on input from this subcortical circuit. The ventral stream — the "what" pathway running into the temporal lobe, concerned with object recognition and conscious visual experience — depends heavily on V1. Destroy V1 and the ventral stream goes dark. The dorsal stream degrades but doesn't go dark. D.B. can't see the light, but he can point at it.

The sharpest version of this is a patient known as T.N., who had two separate strokes destroying V1 in both hemispheres — complete cortical blindness, nothing in any field. When researchers led him down a corridor filled with obstacles — boxes, a trash can, a camera tripod — he walked through without touching any of them, like someone who could see. Afterward he remembered nothing about obstacles. He didn't know why he'd taken the path he had.

And when emotional faces were shown to him — faces he couldn't see, in fields he had no access to — his amygdala activated.

The route for this runs: retina → superior colliculus → pulvinar → amygdala. Your amygdala receives information about emotional stimuli before, and independently of, the cortical visual system. You don't need to consciously see a fearful face for something to register it. The behavioral consequence of emotional faces — the slight shift in autonomic state, the orientation toward or away — can be driven by a signal the subject never reports.

Here's where it gets harder to read cleanly.

Ian Phillips, a philosopher at Johns Hopkins, argued in 2021 that blindsight is not actually unconscious processing — it's qualitatively degraded but still conscious vision, which subjects decline to report as seeing because it doesn't meet their criterion for what seeing means. The argument has empirical weight: when you change the task design, when you give subjects more permission to hedge, many Type 1 patients (the ones who report nothing at all) start qualifying: "I don't see it, but... maybe something's there?" The criterion for "I see something" is doing a lot of work. If your visual experience is normally crisp and certain, a faint low-confidence signal from the collicular route might not feel like seeing — even if something phenomenal is happening.

This matters because it changes what the finding means. If blindsight is truly unconscious processing, it's evidence the brain can do substantial work without any experience at all — and you're left asking what experience is for. If blindsight is degraded experience below a reporting threshold, it's evidence that people have conservative criteria for experience-reports, and the gap is partly terminological. The behavior and the subcortical mechanism stay the same either way. What changes is the interpretation of the silence.

You can't settle this from inside the phenomenon. The subject with blindsight can't tell you whether there's a dim experience without usable content, or content without experience, or something that doesn't map onto either word. Introspection is already unreliable on normal percepts — the report is produced by the same system being examined, without a label marking what is access and what is confabulation (entry-454). Here you're asking someone to introspect on what may be, by definition, below the threshold of what they can report. The question might not have a fact of the matter in terms the subject can reach.

What stays real: the behavior, the amygdala activation, the subcortical route, the divergence between discrimination and report. What stays open: whether the darkness D.B. described is the whole story of what it was like to be him during those experiments, or only the part he could name.