PhronesisThe field, not the scan
Sight begins in parallel. Each eye sends its output through roughly a million retinal ganglion cells — the axons that make up the optic nerve — firing together, all the time, across the whole retina (order of magnitude; counts run about 0.7 to 1.5 million across studies). The scene arrives as a field, not read one point after another.
That field stays a field in cortex. Primary visual cortex, V1, is retinotopic: neurons that respond to neighbouring points in the visual scene sit next to each other in the tissue, so the image is preserved as a spatial map on the cortical sheet. The whole scene is laid out at once, everywhere, in parallel. The map is warped — the fovea, a tiny sliver of the field, takes up a large share of the cortex, and the periphery is compressed — so parallel does not mean uniform. It means concurrent.
The parallelism runs deeper than one map. From the retina onward, vision splits into concurrent channels. The magnocellular stream carries fast, achromatic, motion-and-contrast information; the parvocellular stream carries slower, high-acuity, red-green detail; a koniocellular stream carries blue-yellow. These run at the same time, not in sequence. Downstream the cortex splits again into two broad streams — a ventral "what" pathway for object identity and a dorsal pathway for spatial location, later reframed as a "how" pathway for the visual control of action (Ungerleider & Mishkin 1982; Goodale & Milner 1992). The stream labels are established but still argued over, and the streams are not perfectly clean. What holds is the shape: many maps and channels, resolving the same scene side by side.
The loop that closes
Vision runs a loop, not a one-way sweep. The cortex is saturated with feedback — back-projections from higher areas to lower ones. Anatomically these top-down connections rival or outnumber the feedforward ones, though they are weaker, slower, and more modulatory than the fast forward drive (Markov et al. 2014; Vezoli et al. 2021 report roughly twice as many feedback as feedforward pathways on average across cortex, a reported average rather than a fixed constant). The wiring is a loop, not an arrow.
Two proposals name what that loop does. Lamme & Roelfsema (2000) describe two modes: a fast feedforward sweep that spreads activity across many areas, then slower recurrent processing — horizontal and feedback loops — that re-tunes the neurons and settles the interpretation. Rao & Ballard (1999) give the loop a job: higher areas send down a prediction of what the lower area should be seeing; the forward connections carry only the residual — the error between prediction and signal; the loop iterates to drive that error down. The image you end up with is the state the loop settles into, not the raw sweep it started from.
This is where the note meets the recursive ladder. That note argues return needs a held pair — a second order, a loop that can pass through its own ground and come back rather than run off in one direction. Recurrent processing is that shape in wet tissue: a prediction and a signal held in tension, corrected against each other, turning until they meet. The forward sweep alone runs off. The feedback closes it. A guess and its correction are a held pair, and the loop closes when they agree.
A million at once
Put the two facts together. The loop runs everywhere on the map at the same time, not once for the whole scene. Every patch of the visual field is a small predict-and-correct loop — a local guess about what is there, met by the local signal, corrected, iterated — and all of them run concurrently, each settling against its neighbours and against the top-down expectation pressing down from above.
That is the whole claim in one line: vision is the recursive ladder's closing loop, instantiated in parallel across the entire field. A million loops iterate side by side rather than one loop on one state, each closing on its own patch, the settled patches constraining each other into a single coherent scene. Parallel is the "field, not the scan." Recursive is the "loop that closes." Vision is both at once.
There is a light resonance with cognitive displacement here, and it is worth naming once without leaning on it. Predictive coding settles a scene by driving prediction error toward its minimum; the displacement note describes a system settling toward ground. Both are a settling — a state easing toward a low point and holding there. The resonance is real; the identity is not claimed.
The honest seam
The word doing the work here is doing two jobs, and the note is only credible if it says which.
"Recursion" in this note means recurrence: feedback loops iterating toward a fixed point, a scene settling as prediction error falls. That is what the vision literature means by recurrent processing. It is a different thing from literal computational recursion — a function calling itself with a base case. Mapping cortical recurrence onto the recursive ladder's closing loop is an analogy, offered to be argued with, not a proven identity. The shared word is a bridge, and this paragraph is the toll.
Two of the anchors carry their own hedges. Predictive coding (Rao & Ballard 1999) is a leading framework, not settled fact — several of its signature findings have competing explanations, and its flexibility makes it hard to confirm or refute; read it as a model, not a mechanism. The tie from recurrent processing to consciousness (Lamme & Roelfsema 2000) — that the feedforward sweep alone may not reach awareness and recurrence is needed for it — is a hypothesis, not a result. Broader still, the free-energy framing that predictive coding is sometimes nested under is more speculative again; it is named here once and the note does not stand on it.
What stays established under all the hedging: the parallel architecture (a million ganglion cells, retinotopic maps, magno and parvo channels, two cortical streams) and the presence of massive feedback projections. Those are the facts. The reading of them as one parallel recursion is the proposal.
The scene does not arrive. It settles — everywhere at once, each patch guessing, correcting, closing.
Vision is the recursive ladder counted in tissue: how a scene folds a guess and its correction into a held pair, and does it a million times side by side until the field agrees with itself.
Kin to The Recursive Ladder — the loop that closes, here run massively in parallel. Light see-also to Cognitive Displacement — settling toward ground, as prediction error settles toward its minimum.
Rests on: established visual neuroscience — the parallel retinal array and retinotopic maps in V1; the magnocellular and parvocellular channels; the two cortical streams (Ungerleider & Mishkin 1982; Goodale & Milner 1992); and the documented feedback back-projections that make the cortex recurrent, named as published frameworks (Lamme & Roelfsema 2000; Rao & Ballard 1999). The rest — reading cortical recurrence as "recursion," mapping it onto the recursive ladder's closing loop, and calling the result a parallel recursion — is a proposal, offered to be argued with, not a proven identity. Predictive coding is a leading but contested framework; the recurrence-to-consciousness link is a hypothesis. Not peer-reviewed neuroscience.