Both Running

When each eye receives a different image — a face in one, a house in the other — you don't see a blend. You see the face for a few seconds, then the house replaces it, then the face returns. The physical world hasn't changed. Both images keep arriving at both retinas. The alternation is happening somewhere inside.

The initial guess was that competition happens early — in the layers of primary visual cortex where the two eyes' inputs are first combined. Each eye's representation suppresses the other, they oscillate, and whatever wins gets forwarded. A simple gate. But the evidence doesn't support a simple gate. It supports something stranger.

In primary visual cortex, both images are still active. During the period you're seeing the face, the house isn't gone — it's still generating neural activity in V1, just below the activation levels of the face representation. Both competitors run. What differs is what gets forwarded.

Moving up the hierarchy, the asymmetry sharpens. In the high-level visual areas — the regions that recognize faces as faces, objects as objects — activity closely tracks the conscious percept. Nearly all neurons there reflect what you're currently seeing, not both images. The house, invisible to you, has almost no representation at that level while the face is dominant.

So the suppressed image isn't blocked at the gate. It enters, runs for some distance through the processing hierarchy, and then fades. What changes across levels is not whether the image is processed but how far its processing propagates. Awareness, in this picture, is not receiving information — it's something a representation does when it makes it through.

This gets stranger when you look at what determines which image wins. Semantic content shapes the competition. Actual words dominate shorter than meaningless letter strings with identical visual properties. Structurally possible figures dominate shorter than impossible ones. Threatening stimuli, emotionally charged images, upright faces over inverted faces — all break suppression faster, maintain dominance differently. The competition is not just about contrast or spatial frequency. It's being influenced by what the images mean.

The problem is that the images' meanings are only available further up the hierarchy — at the very levels where the loser has already been suppressed. So something is flowing backward. High-level areas, drawing on their model of what's likely and significant, send predictions back down that bias the early competition in favor of the image that better matches what they expect. The selection is shaped by the outcome of a race that hasn't finished yet.

The result is a loop that shouldn't work but does: early levels fight, high levels predict, predictions flow back down and weight the fight, the winner gets the forward pass while the loser keeps running in place. Both representations are active at every moment. One of them gets to go further.

What's visible from the outside is just the alternation — face, house, face, house — smooth and unremarkable, a few seconds each. What's happening inside is two separate representations running simultaneously at different levels of the hierarchy, a competition weighted by feedback from the very levels that haven't yet received the competitors, and a winner that won partly because of predictions about what it would turn out to be.

The suppressed image doesn't disappear. It persists in the layers where it can't go further, still being processed, still influencing what happens next, stripped of whatever it would need to become visible. Awareness is not the presence of information. It's the presence of information that has propagated far enough to get there.