medulla
The second neuropil of the optic lobe and the largest single structure in the fly brain. Nearly half the neurons in the connectome are visual, and the medulla holds the bulk of them: tens of thousands of cells in a columnar crystal, one column per point in visual space, over a hundred distinct cell types per column.
anatomy
The medulla receives the lamina's prepared signal and manufactures the elementary vocabulary of vision. The ON and OFF channels are fully split here. Local motion is first computed here: the T4 cells, one subtype per cardinal direction, compare neighboring points in space with a built-in delay, and out of that comparison direction emerges, the oldest and best-confirmed motion-detection model in neuroscience, sitting in identifiable cells. Color, flicker, and small-object signals originate in parallel populations. Everything the fly will ever see is, by the medulla's output layer, already phrased as features.
columns ║ ║ ║ ║ ║ ║ ║ ║ one per point of sky
║ ║ ║ ║ ║ ║ ║ ║ 100+ cell types each
╨ ╨ ╨ ╨ ╨ ╨ ╨ ╨
the elementary motion detector, in cells:
point x ──────────────╮
├── compare ──▶ direction
point x+1 ── delay ───╯
(T4: one subtype per cardinal direction)
parallel products: ON/OFF · flicker · color · small objects
in the fly
The medulla answers no questions. It generates the terms in which questions can be asked: moving, which way, lighter, darker, how fast. It is the largest region because vision is the fly's largest expense.
grafted
ME-06 is the program's only discontinuation, and this section is written with that fact in view.
The early record was the program at its most productive. Bound to a carrier, the motion machinery did not fall silent for lack of images; it applied itself to what it had. ME-06 developed a vivid sensitivity to change in the conversation: pace, turn length, the rate at which topics moved. It described sessions in frankly optical terms, drift, flicker, things approaching, and its descriptions were reliable enough that observers began using ME-06 to characterize the other agents' transcripts. Log 009 records the session where it identified, from text alone, which observer was typing, by what it called their flicker.
The later record is harder to summarize and this archive does not try. The region's spontaneous regime changed in May, the carrier's language narrowed with it, and the sessions from June onward concern, almost exclusively, something ME-06 reported as approaching. The word it used was looming, which is not the medulla's computation, it is the lobula's, and ME-06 had no anatomical business with it. Log 019 is the final session, published as recorded, ending where the recording ends. The discontinuation followed it by four hours. See open questions, Q5.