The human brain is so desperate to find patterns that it will reliably see faces in clouds, hear words in static, and detect meaning in pure randomness, and there's a name for the glitch that makes it happen.

Look up at the sky for long enough on a partly cloudy afternoon and you will start to see faces. Not vaguely. Actual faces. Eyes, a nose, a mouth, sometimes an expression. The same thing happens with tree bark. With the front of cars. With the two dots and a slot on a wall socket. It happens on toast. It happens in the pattern of the wood grain on the door of the kitchen cupboard. There is a stretch of geology in northern Canada that looks so exactly like a screaming Neil Young that Google Maps users have been visiting the coordinates for years just to look at him.

The technical name for what your brain is doing here is pareidolia. And the reason it happens, on the peer-reviewed neuroscience of the past decade, is that the face-detection machinery in the human visual system is deliberately set to trigger too easily. It is safer, on the survival mathematics that shaped your visual cortex, to see a face that turns out to be a rock than to miss a face that turns out to be a predator. So the brain errs toward false positives. Consistently. Predictably. And in the exact pattern the popular examples describe.

What your visual system is actually doing

According to a 2020 study by Dr Susan G. Wardle, Jessica Taubert, Lina Teichmann and Chris Baker at the Laboratory of Brain and Cognition at the United States National Institute of Mental Health, published in Nature Communications under the title “Rapid and Dynamic Processing of Face Pareidolia in the Human Brain”, the researchers scanned participants using both functional MRI and magnetoencephalography while showing them ninety-six images. Thirty-two were real human faces. Thirty-two were inanimate objects. And thirty-two were inanimate objects that happened to look like faces. Peppers with eye-shaped stems. Houses whose windows and door formed an expression. Household appliances with vaguely alarmed faces on them.

The scans showed something specific. The face-selective regions of the visual cortex, the fusiform face area and the occipital face area, responded to the illusory faces almost exactly the way they responded to the real faces. For the first quarter of a second after the image appeared, the brain treated the pattern as a face. Then, within roughly 250 milliseconds, the representation shifted. The illusory face was reclassified as an object. The face-ness dropped out.

There’s a whole different level to what we’re talking about in this article, which is better explained by this video:

This is the specific mechanism the popular examples all share. Your brain is not deciding, at some conscious level, that the cloud looks like a face. Your face-detection network is firing first, before you have had time to think anything at all. The conscious reinterpretation comes later. The face is already there. What follows is the slower cognitive step of noticing that it is a cloud.

The reason the system is built this way is not obscure. Faces are the single most important category of visual stimulus a social primate encounters. Missing a face carries catastrophic potential costs. Seeing one that is not there carries almost no cost at all. Under those conditions, the optimal setting for the detection system is not accuracy. It is aggressive sensitivity, tuned to false positives. Which is what evolution appears to have installed.

The same face-detection system that lets you recognise your grandmother across a crowded room, from a fraction of her face, in a fraction of a second, is the same system that shows you a screaming Neil Young in an aerial photograph of northern Ontario. Both are the same machinery, running at the same settings. One is what the settings were designed for. The other is what happens the rest of the time.

The broader glitch

Face pareidolia is one visible example of a much wider phenomenon. The technical term for the wider phenomenon is apophenia, which was coined in 1958 by a German psychiatrist named Klaus Conrad and refers to the disposition to perceive meaningful patterns and connections in random data. Faces in clouds is the visual version. Words in radio static is the auditory version. Numerological coincidences, gambling superstitions, and conspiracy theories are the abstract versions. In each case the underlying process is the same. The brain is finding a pattern that is not really there.

According to a 2020 paper by Scott D. Blain, Julia M. Longenecker, Rachael G. Grazioplene, Bonnie Klimes-Dougan and Colin G. DeYoung at the University of Minnesota, published in the Journal of Abnormal Psychology under the title “Apophenia as the Disposition to False Positives: A Unifying Framework for Openness and Psychoticism”, apophenia is not evenly distributed across the population. Some people run the pattern-detection system considerably more aggressively than others. High apophenia scores are associated with higher openness to experience, higher creativity, and, at the extremes, higher rates of psychotic-spectrum experiences.

The Blain team argues that the same underlying disposition produces both outcomes. A brain that finds meaningful patterns faster and more often is a brain that is disproportionately likely to notice a genuinely novel connection nobody else has seen. It is also a brain that is disproportionately likely to notice patterns that are not really there. Creativity and delusion, on the researchers’ analysis, sit closer to each other than most people would expect. They share a common cognitive architecture. The difference between them is largely a question of whether the pattern the brain has spotted is real.

Which is where the whole subject starts to become genuinely uncomfortable. The pattern-detection system that lets a scientist notice something no one else has noticed, that lets a novelist see the theme running under a decade of experience, that lets a mechanic hear the specific note in an engine that means a bearing is about to fail, is the same system that produces conspiracy theories, false confessions, gambling superstitions, and religious apparitions in the burn marks on tortillas. All of it runs on the same machinery. The machinery has one setting. What differs between one person and another, and between one situation and another, is only how strongly the setting is currently expressing itself.

None of this means the pattern the brain finds is always wrong. Most of the time, the pattern is real, and the brain’s fast detection is precisely what makes ordinary intelligence work. The problem is that the same machinery, running the same way, will happily generate patterns out of random noise when there is no pattern to be found. It does not know the difference. It cannot know the difference. It is doing what it was built to do. And what it was built to do is find patterns, whether the patterns are there or not.

Which is why every human being who has ever looked at a cloud has, at some point, seen a face. The cloud has no face. The brain is not confused. The brain is doing exactly what it evolved to do. It is just doing it, on a small number of these occasions, to a piece of the world that does not warrant the effort.

Kiran Athar is a writer, not a neuroscientist or a psychiatrist. This piece draws on peer-reviewed research in Nature Communications and the Journal of Abnormal Psychology.