Hollow Face Illusion

A hollow mask that bulges

If you turn a mask round so you're looking into its hollow back, with the nose pointing away from you, it ought to look like what it is: a bowl-shaped dent. It doesn't. The hollow mask looks like an ordinary face bulging out towards you, nose nearest, which is the exact opposite of the truth.

It gets stranger when you move. Walk past a hollow mask and the face seems to turn to follow you, an unsettling effect that has long been a staple of haunted houses. The picture above can only show you the still version, I'm afraid; to see the movement, you'll need a real mask, or one of the many videos of rotating hollow masks online.

Faces are expected to stick out

You've seen an enormous number of faces in your life, and nearly all of them were convex. So your brain has a very strong expectation that anything arranged like a face bulges outwards, strong enough to overrule depth cues it would normally trust, such as shading. A hollow mask's shading is actually consistent with a hollow shape, but your brain overrides it and fits the evidence to what faces are supposed to look like.

That makes this one of the clearest examples of perception working from expectation as well as from the light reaching your eyes, which is the idea behind predictive processing. The apparent turning comes from the same mistake. As you move past a hollow surface, its shading and outline change in a way that, for a convex face, could only mean the face was rotating to keep looking at you, and so that's what you see.

What weakens it

Harold Hill and Vicki Bruce measured how close people could get to a hollow mask before they saw it as hollow. The illusion was stronger when the mask was lit from above, as faces usually are, and when it was the right way up. Turned upside down, the mask still fooled people to some extent, which suggests a general preference for convex shapes as well as one for faces. Looking with two eyes instead of one also weakened the illusion at close range, because the difference between the two eyes' views gives direct evidence of the true depth (Hill & Bruce, 1993).

There's also a striking exception. Danai Dima and her colleagues found that while healthy volunteers almost always saw hollow faces as normal faces, people with schizophrenia could usually tell that they were hollow. Brain imaging pointed to weaker top-down influence, meaning the expectation that normally overrides the evidence wasn't pulling as hard (Dima et al., 2009).

Gregory's favourite illusion

Rotating hollow masks were novelties long before anyone studied them, but the British psychologist Richard Gregory made the illusion famous as evidence that perception is a kind of hypothesis-testing, not a passive recording (Gregory, 1970; Gregory, 1997). His view anticipated today's predictive-processing accounts of vision. If you'd like to explore further, the Thatcher effect is another illusion that depends on how specially the brain treats upright faces, and the Ames room is another case of the brain trusting its expectations over the real geometry.