Binocular rivalry

Why do two different images shown to the two eyes alternate instead of blending into one?

Show the left eye horizontal stripes and the right eye vertical ones, in the same place at the same time. The obvious predictions are a plaid or a haze — some average. Neither happens. You see the horizontal stripes cleanly for a second or two; then they break up, the vertical ones take over, then back, indefinitely, and wanting a particular outcome does not produce it.

Two things need explaining and are easy to run together. Why does the system refuse to average? And having refused, why does it not pick one and stop?

Start with the refusal to average, the more revealing half. Averaging is what the visual system does constantly and correctly — two eyes normally see almost the same scene from slightly different positions, and combining them is how depth is recovered. The machinery for blending is present and working, so something must be switching it off.

What would tell a visual system that blending is wrong here? If both images were true of the same patch of world, you would be looking at a surface striped two ways at once, which does not occur. In ordinary viewing a large mismatch between the eyes at one location does have a common cause: one eye's view is occluded, by your nose, a branch, a window frame — and then averaging is exactly wrong, because the two images belong to different surfaces. A system built to recover one world from two views should treat gross incompatibility as a signal to choose. That is a functional argument for why choosing is sensible, not a measurement.

Now the second half: given that it must choose, why does the choice not stick?

Suppose the two images are represented by two populations of neurons, each suppressing the other — mutual inhibition. Whichever gets ahead pushes the other further down, so the system runs away to one of two stable outcomes. That gives exclusive perception immediately. It also gives permanence, which is wrong.

So add the one thing we know about neurons under sustained drive: they adapt. The dominant population is firing hard, so it adapts fastest; its inhibition of the loser weakens while the loser, quiet, recovers. At some point the balance tips, the two exchange roles, and the new winner starts adapting in its turn. Inhibition supplies exclusivity, adaptation supplies switching, and neither alone produces both.

Test that. A pure inhibition-plus-adaptation oscillator is a clock, and should produce alternations of near-identical length. Measured dominance durations are not regular at all: they are right-skewed, short spells common with a long tail, and successive durations are close to uncorrelated, so knowing this spell was long tells you almost nothing about the next. That is the signature of noise, not a timer, deciding when the tip happens — so the model needs a third ingredient, and the irregularity is the evidence for it rather than an embarrassment.

One further constraint is counterintuitive enough to state. Raise the contrast of one eye's image and you might expect it to dominate for longer. What is chiefly observed instead is that the other image's spells get shorter — the modern reading of Levelt's second proposition, which I am recalling rather than deriving. That is what mutual inhibition predicts: strengthening one population's drive mostly strengthens the suppression it applies, which bites on the rival's turn rather than its own.

The natural picture so far is that the two eyes are competing, and for a long time that was the assumption. It is at best incomplete. Logothetis, Leopold and Sheinberg showed in 1996 that if you swap the two images between the eyes several times a second, perception need not swap with them — an observer can go on seeing one pattern for its usual second or two while the eye delivering it changes repeatedly underneath. Whatever is alternating is therefore not simply the two monocular channels; the competition can be between representations of the two stimuli, at a level where which eye supplied them has already been discarded.

That does not make the eyes irrelevant. Interocular suppression is real, and effects tied to early, eye-specific stages are documented too. The position most of the field has settled into is that rivalry is a hierarchy of competitions rather than one, with the distribution among them still argued. State it as contested, because it is.

Two smaller corrections. Exclusivity is an idealisation: with larger patches you routinely get patchwork percepts, part of one image and part of the other with the boundary travelling — what you would expect if the competition is fought locally and coordinated only over short distances. And "you cannot control it" is too strong: attention shifts the balance and effort can lengthen a favoured spell. What you cannot do is stop the alternation, which is the part adaptation explains.

What would refute the account. Regular, serially correlated dominance durations would sink the noise-driven story and demand something like a timer. Contrast that lengthened an image's own dominance without shortening its rival's would make mutual inhibition the wrong shape. And perception that always followed the eye through Logothetis's swaps would put the competition between eyes and nothing higher. Each could have been observed and was not.

• Why patchwork percepts appear at larger sizes, and what that says about how locally the competition is fought.
• What rivalry has been used to argue about consciousness, given that the stimulus is constant while the experience is not.

Every term the collection defines is gathered in the glossary.