If Your Brain Predicts What You See, Why Isn’t Vision a Hallucination?

Walk into a dim bedroom at night and a coat hanging over a chair can, for one unpleasant second, look remarkably like a person.

Then you switch on the light.

The figure disappears. The coat is still exactly where it was. Nothing in the room has changed except the quality of the visual information reaching your eyes.

That small experience points to a much bigger question. If the brain uses expectations to help interpret what we see, why don't we simply see whatever we expect to see? Why does the coat eventually become a coat rather than remaining an intruder?

The answer seems to lie in the constant negotiation between what the brain expects and what the eyes actually provide.

Your Eyes Are Not a Video Camera

It is easy to imagine vision as a kind of biological recording system. Light enters the eye, the retina captures an image, and the brain somehow displays the finished picture.

But the visual system is much less uniform than a camera sensor.

Fine detail is concentrated in a small central region of the retina called the fovea. Move away from that center and visual acuity falls quickly. Peripheral vision is much better at giving us broad information about shapes, movement, contrast, and texture than at providing the kind of fine detail available at the center of gaze^[7].

Yet that is not how the world looks to us. We do not experience a tiny island of sharp vision floating inside a blurry field. The scene feels continuous and richly detailed.

Part of that impression comes from the fact that our eyes are constantly moving, sampling different parts of the scene. But visual perception also depends on context, previous experience, and information already available elsewhere in the brain. What we see is therefore not simply a frame delivered intact from the retina.

And sometimes we can watch that process change in real time.

When a Meaningless Image Suddenly Becomes a Face

Mooney images are a good example. They are simple black-and-white pictures made from large patches of light and dark. The first time you see one, it may look like little more than random blotches.

Then someone tells you there is a face in it.

Suddenly the image reorganizes itself. A forehead appears. A nose becomes obvious. What looked like meaningless patches a moment earlier becomes a face, and afterwards it can be surprisingly difficult to return to the original state of not seeing it.

The stimulus has not changed. Your knowledge has.

Brain-imaging work suggests that recognition also changes how the same visual input is represented in early visual cortex^[6]. Once an ambiguous image is recognized, patterns of activity in early retinotopic areas become more similar to those produced by the corresponding clear image.

That matters because it shows that recognition is not simply added at the very end of perception, after an untouched visual picture has already been constructed. What we know about an object can influence how the incoming visual pattern is represented relatively early in processing.

But this creates an obvious problem. If knowledge can reach that far into perception, something must prevent expectation from taking over completely.

What Happens When the Brain Gets It Wrong

One influential way of explaining this balance is predictive coding^[1].

In this framework, the visual system is not only reacting to incoming stimulation. Higher levels of the system generate expectations about what is likely to be present, while incoming sensory information is used to test those expectations.

The important part is the mismatch.

If the incoming signal fits the brain's current prediction reasonably well, there is little need to revise the model. But if the sensory evidence does not fit, the discrepancy becomes informative. Predictive-coding theories describe this discrepancy as a prediction error.

Some neuroimaging findings fit this idea particularly well. Expected visual stimuli can produce lower overall activity in primary visual cortex while, at the same time, producing sharper neural representations of the stimulus^[3]. The result is not simply "less processing." The representation may actually become more efficient when the input matches what the system was prepared to encounter.

Mismatch produces the opposite kind of problem. In animal experiments, neurons in visual cortex respond strongly when expected visual feedback from an animal's own movement suddenly fails to match what is actually seen^[2].

This is one reason predictive coding is useful as a model of perception. A prediction is allowed to guide interpretation, but it does not get the final word. Incoming sensory evidence can contradict it.

The coat can look like a person for a moment. The light eventually corrects you.

Why Darkness Gives Expectations More Room

Not every sensory signal deserves the same amount of trust.

A clear, high-contrast object in daylight gives the visual system much more reliable information than a vague silhouette seen through fog. Predictive-processing accounts describe this difference in terms of precision: roughly, how reliable the system estimates a particular source of information to be^[5].

You can think of precision as changing how loudly different pieces of evidence get to speak.

In good lighting, the visual signal is usually strong enough to constrain interpretation tightly. You may expect your dog to be standing in the garden, but if there is clearly a cat sitting there, expectation does not have much room to negotiate.

Darkness is different.

Edges are weaker. Color information deteriorates. Fine details disappear. Several different objects may become compatible with the same rough pattern of light and shadow.

Now previous experience matters more. A shape at the end of the hallway might be a coat, a doorway, or a person. If you are already uneasy, "person" may briefly win.

Then the light comes on, uncertainty collapses, and the interpretation changes almost immediately.

This does not mean the brain literally chooses fantasy whenever vision becomes noisy. It means that ambiguous evidence leaves more room for prior knowledge and expectation to influence which interpretation seems most plausible.

What If the Eyes Disagree?

There is an even stranger way to expose this competition.

In binocular rivalry, researchers present a different image to each eye at the same location—for example, a face to one eye and a house to the other^[4].

You might expect the two pictures to merge into a permanent hybrid.

They usually do not.

Instead, perception fluctuates. For a while the face dominates. Then the house appears. Sometimes pieces of both are briefly visible, but experience tends to alternate between competing interpretations rather than settling into one stable mixture.

Predictive-coding models offer one way of understanding this. The visual system is faced with two incompatible explanations of the same location, and neither can permanently account for the incoming evidence. Perception therefore shifts as the balance between the alternatives changes.

Binocular rivalry does not prove that predictive coding is the only explanation for vision. But it makes one thing very clear: perception is not simply the sum of whatever signals happen to arrive at the eyes.

Reality Still Pushes Back

Calling vision "constructed" can make it sound as though the world outside the brain has somehow become irrelevant.

It has not.

The construction is constrained continuously by light arriving from real objects. Expectations help the brain interpret ambiguous information, but sensory evidence can resist those expectations, contradict them, and force perception to change.

That is why a shadow can become a person in the dark but usually stops being one when the room is illuminated.

Vision is neither a perfect recording of the world nor a private hallucination projected onto it. It is something more interesting: an ongoing attempt to explain sensory evidence while the evidence keeps pushing back.

Your brain makes predictions.

The world corrects them.