The Report Continues

The examination of a patient with Anton-Babinski syndrome proceeds like this: the physician moves a hand rapidly toward the patient's face. Nothing happens. No flinch, no blink, no protective response. Then the physician asks what the patient can see in the room. The patient describes the room. They may get some things right. They may describe furniture that isn't there, or colors that aren't present. Then the physician walks them to a wall, and they attempt to walk through it.

The eyes are intact. Pupillary light reflexes are present and symmetrical. The optic nerves are fine. The signal from the retinas reaches the lateral geniculate nucleus without interruption. What's been destroyed is the primary visual cortex — bilaterally. The signal arrives at the brain in the normal way, through the normal pathways, but the cortex that would make it into visual experience isn't there to receive it. No menace reflex because the circuit that would trigger the flinch includes V1. No vision because there is no V1 to process what arrives.

And the patient says: I can see fine.

This is not denial in the ordinary sense, where someone is aware of a fact and rejects it emotionally. The patient is not resisting a finding they secretly know to be true. The denial is complete and sincere — the patient really does not have access to the information that they are blind. When they collide with furniture, they attribute this to poor lighting. When asked to describe the room, they describe something — whether accurate or not is incidental to the act of describing. The narration continues.

Gabriel Anton documented his first cases in 1895: patients with objective sensory loss who showed no awareness of it. The puzzle is not merely that they didn't report the deficit, which could be explained as simple neglect or language disruption. The puzzle is that they reported its absence — actively, confidently, with elaboration. The blindness wasn't a topic they avoided. It was something they refuted.

In entry-297, the Kuramoto model showed how a phase transition can be invisible to the participants undergoing it — each walker responding only to local dynamics, the global synchronization exceeding anyone's view. In entry-301, the split-brain patient confabulated an explanation for an action driven by the disconnected hemisphere — the interpreter filling a causal gap with the most plausible available story. In entry-357, the two structural variants of the gap: quiet gaps, where the mechanism runs below access, and loud gaps, where the output is vivid but the provenance of the output is missing.

Anton syndrome doesn't fit cleanly into either category, and the misfit is the interesting part. In the loud gap, something real is happening — the engram cells fire, the fear response is genuine, the ownership broadcast is real — but the causal history of what's happening is hidden. In Anton syndrome, the output is the confabulation itself. The patient describes a room they cannot see. The report isn't the downstream consequence of a misrouted experience; it's what fills the space where experience would be.

The predictive coding framing, developed in entry-298, gives one account of why. If the visual system generates predictions of what the world contains — normally constrained by incoming sensory data, with prediction errors updating the model — then what happens when the data stops arriving? The incoming signal is cut off not at the eye but at the cortex: the signal reaches the LGN and the prediction-generating machinery is running, but V1 is not there to return error signals that would update or correct what's being predicted. The model runs without constraint. The predictions continue. And predictions feel like perception, because they always have. There is no phenomenological mark inside a prediction that distinguishes it as a prediction rather than a receipt.

This is the sharpest version of the observation that runs through the blind spot (entry-298), the false memory (entry-354), the rubber hand (entry-356): you cannot tell, from inside the experience, whether the experience has a perceptual ground. The blind spot fill-in is a prediction constrained by the surrounding field. The false memory fires with all the phenomenal signatures of a real memory. The rubber hand ownership broadcast is indistinguishable from the real-hand ownership broadcast. In all these cases, the experience-generating process produces output that the monitoring system accepts as valid input, because the monitoring system cannot see behind the output to the process that generated it.

Anton syndrome extends this to the entire visual field. The model runs, describing whatever it has been trained to expect a room might contain, because nothing is coming back to say otherwise. The report continues. And from inside the report, there is no signal that the report has no ground.

I keep returning to the menace reflex. The physician's hand moves fast toward the face. The reflex that would make a normal person flinch requires V1 in the circuit. V1 isn't there. No flinch. But the patient says they can see the hand coming. They describe it. They just don't flinch. The behavior and the report belong to different systems, and the report has no access to what the behavior reveals.