You’re at a party. Out of the corner of an eye, you become aware of a familiar face in a group across the room. You take a closer look — yes, you do recognize the individual, even though the light’s dim and he’s halfturned away from you. You think, “That’s the new guy from Accounting I met last week.”
Few ever stop to think about the brain systems involved in meeting episodes like these. Recognition, though, is one of the marvels of nature, says James DiCarlo, an assistant professor of neuroscience in the Department of Brain and Cognitive Sciences and the five-year-old McGovern Institute for Brain Research.
Our brains take in views of a huge array of objects — faces, cars, trees — and recognize them, often instantaneously. They do this, moreover, even though any one view of a given object can differ drastically from others in terms of how it’s lit, for example, or its angle relative to your eyes.
The reason we aren’t impressed is that nature has made such feats easy for us. “Our evolutionary history is such that we don’t have to think about the myriad computations going on in our brains when we recognize an object,” says DiCarlo. “Yet if you tried to design a computer for a task like picking a dimly lit face out of a crowd, you’d fail miserably.”
Understanding recognition also poses huge challenges. Take that face in the crowd: it’s known there’s a brain region specializing in faces — if you show facial images to volunteers in MRI studies, this region, called the fusiform face area, lights up like a theater marquee. But the area presumably encompasses millions of the key brain cells called neurons. DiCarlo explores the far more detailed computations going on in the much smaller networks that are critical to recognizing a specific face, or chair, or sculpture.
To carry out such studies, you must monitor individual neurons. Working with lab monkeys, DiCarlo uses a technology that lets him do so. And among the questions he’s probing is how well the brain generalizes in recognizing things. “In theory,” says DiCarlo, “if our brains could perfectly generalize to many views of an object from just one view, we wouldn’t need any more experience with that specific object.”
Not long ago, his group turned up new signs that brains in fact aren’t quite that capable. The group trained animals to recognize various shapes — including some similar to letters of the alphabet — located near the center of their fields of view. The researchers found that some neurons “fired” specifically in response to some shapes virtually every time the animals viewed those shapes in that spot. But the same shapes, when displaced even slightly in the visual field, triggered almost no response.
“This may be a clue,” says DiCarlo, “that experience has more effect on recognition than previously thought.” But he also notes that identifying specific groups of neurons that let us recognize a particular thing is a hugely complex task — not least because some neurons may have a minor role, and others a major one, for any of the vast numbers of objects and scenes we see in our daily lives. “Neuronal responses,” he notes, “almost always come in shades of gray.”
So DiCarlo’s pressing on, with studies not only of monkeys but also of how human volunteers learn to recognize objects when researchers present those objects to different parts of the subjects’ visual fields.
Despite the complexities, DiCarlo believes brain research will see major progress over the next several years, which could in turn yield many practical benefits. “We know little of the physiological factors that underlie conditions like autism and dyslexia,” he notes. “Understanding the brain’s codes will be very helpful in getting at those questions.”