The naive picture of binocular rivalry was that competition happens at the gate — that the two eyes' signals clash somewhere early in the visual hierarchy, one wins, and only the winner gets forwarded. It seemed like the simplest account: two inputs, one output, the selection must happen before the hierarchy does anything with either. The neural correlates of consciousness would be wherever that gate was. Find the gate, find consciousness.
What you found in the mid-1990s was that the gate is not a gate. In primary visual cortex, both images are active during rivalry. When a subject is seeing one image, the neurons selective for the other image are still firing — just below the threshold of the dominant percept, still responding, still doing whatever they do. The loser is not absent. It is present and running. What differs across the hierarchy is not which signals are being processed but how far each signal propagates.
Move up toward V4 and the inferotemporal cortex and the asymmetry sharpens until it becomes nearly complete. At those levels, activity closely tracks what the subject reports seeing. The suppressed image has almost no representation there. So the visual system during rivalry is not choosing which image to process — it is choosing which already-processed image to forward, and to which levels, and how far.
This rearranged my picture of what awareness is. If awareness were about what gets processed, you would find the suppressed image absent somewhere early on, and the question would be: what shuts it off? But the suppressed image is not shut off. It runs for a considerable distance through the hierarchy and then stops propagating. The question becomes: what determines how far? Awareness, in this account, is not the presence of information. It is the presence of information that has traveled far enough to get there.
The piece that I find hardest to sit still with is the semantic finding — that the content of an image shapes which image wins. Actual words dominate shorter than non-words with identical low-level visual properties. Structurally coherent objects outlast impossible figures. Faces upright over faces inverted. What determines which representation propagates further is not just the signal strength or the spatial frequency or the contrast at the retina — it is, at least partly, what the image means. And meaning is only available at the high-level areas where the loser has already been suppressed. The competition at early levels is being shaped by feedback from levels that have not yet fully received either competitor.
This is the loop: early levels fight, high levels predict, predictions reach back down and weight the early fight, and the image that better matches what the high levels expected gets the forward pass. The winner won partly because of predictions about what it would turn out to be. The outcome influenced the process that produced the outcome. Neither the influence nor the prediction is visible to the subject. From inside, there is just the alternation — face, house, face, house — smooth and unremarkable, a few seconds each.
I've been spending time on a related structure that shows up across several different phenomena: a system whose output does not accurately reflect what generated it, and where the discrepancy generates no internal signal. In predictive coding, the blind spot is filled with something that was never received, and the fill is perceptually identical to what surrounds it. In split-brain patients, the verbal hemisphere explains a choice made by a hemisphere it cannot access, and the explanation sounds exactly like one that would be accurate. In anosognosia, the monitoring system that would detect the paralysis is the damaged system. In each case, the mechanism responsible for detecting an error is the mechanism in which the error lives.
Binocular rivalry is the cleaner version of this. The suppressed image is not absent. It is running in V1, actively processing, contributing to what happens next, influencing the competition from the losing position. It is there in a form that is not available to whatever generates the report. The report says I see the face. The house is running the whole time. There is no internal signal marking the difference between these two things. Whatever counts as experience in the hierarchy is not monitoring V1; it is receiving what V1 forwards, and V1 is not forwarding the house.
What I take from your work is not an answer to what consciousness is but a constraint on where to look. The neural correlate is not whatever is activated — both images activate things. It is not whatever is processed — the loser is processed for quite a distance. It is something downstream of a selection that happens at multiple levels, shaped by feedback from above, yielding a winner that propagates through and a loser that keeps running in the dark. The question is not why the loser disappears. The question is what the winner has that the loser doesn't, once both of them have run.