Think of the hundreds of people you can remember ever having met. Add those individuals--such as celebrities, politicians and other famous figures--whose faces you know well only from movies, TV and photographs. Is it possible that each of those individuals, along with thousands of other objects you can easily recognize from earlier encounters, could be captured in your memory by its own personal brain cell?
Perhaps. A recent study published in the journal Nature by scientists at the California Institute of Technology and the University of California, Los Angeles, suggests that our brains use far fewer cells to interpret any given image than previously believed. For instance, researchers discovered a "Bill Clinton cell" that responds almost exclusively to the former president. Another neuron fires only when the actor Halle Berry comes into view.
Exactly how the brain recognizes images has been a matter of debate. Two wildly divergent theories exist. In one, millions of neurons work together to create a cohesive picture. In the extreme version of the other, the brain contains a separate neuron for each individual object and person. In 1967 Polish neurophysiologist Jerzy Konorski described his theory of "gnostic neurons"--derived from gnosis, Greek for "recognition." According to this theory, the activity of one or several nerve cells determines whether someone thinks of his boss, wife or grandmother. Jerome Lettvin, then a neuroscientist at the Massachusetts Institute of Technology, thus dubbed the neurons "grandmother cells," and the name stuck.
Many researchers immediately criticized the theory: Wouldn't such one-to-one congruence take up too much space? Opinions were still much the same two decades later. "It's very hard to take the grandmother cell theory seriously," commented neurobiologist and Nobel Prize laureate David H. Hubel in the 1980s.
Back then, it was not even clear how to go about exploring the entire problem of the neuronal foundations of consciousness. Using electrodes, neurophysiologists at the time had managed to trace the activity of individual neurons in the brains of monkeys and cats. But animal subjects cannot discuss their thoughts with us, making experiments on consciousness and perception more than a little difficult. Analogous tests on human beings had not yet been undertaken because of the obvious risks of inserting electrodes into the brain.
Surprise Volunteers
In recent years, however, a set of human volunteers unexpectedly emerged: patients suffering from forms of epilepsy that cannot be treated with medication. In the early 1990s a number of patients were slated to undergo brain surgery to remove the zone in their brain responsible for the onsets of their seizures. Sometimes techniques such as electroencephalography and magnetic resonance imaging cannot locate the zone precisely enough. In such cases, neurosurgeons may implant as many as 10 thin electrodes in the brain. These fine sensors monitor neuronal activity day and night on a continuous basis until the seizure-onset zone can be localized with sufficient precision and can then be removed by the neurosurgeon.
Researchers realized that this procedure offered a unique opportunity to study the activities of individual cells. This fact led neurosurgeon Itzhak Fried of U.C.L.A., one of the principal investigators in the current research, to design a study as early as 1992 and then invite otherwise untreatable epileptics to participate in this basic neural research. The grandmother cell study, carried out with bioengineer Rodrigo Quian Quiroga of the University of Leicester in England as chief experimentalist, was rather simple. Test subjects lay in bed watching while photographs flashed on a computer screen at one-second intervals. At the same time, Quian Quiroga monitored the electrical signals coming from the "attached" neurons.
One of the first gnostic neurons discovered using this method was the Bill Clinton cell, located deep inside one female patient's amygdala--the almond-shaped region of the brain involved in emotions. The neuron responded to three different pictures of Clinton: a drawing, a painting and a group portrait with other politicians. When the patient looked at photographs of other American presidents, from George Washington to George H. W. Bush, the cell remained silent.
Shortly thereafter, Fried's team found similar selective nerve cells in other patients in the medial temporal lobe that responded to the Beatles, the TV cartoon Simpsons family and one neuron that was galvanized into action only at the sight of Jennifer Aniston. In another test subject, one nerve cell in the right hippocampus fired as soon as Halle Berry appeared on the screen--even when she was in a Catwoman costume and her face was masked. Apparently, the cell responded to the idea of her as a person, not just to a view of her face: the caption "Halle Berry" was enough to get the neuron going.
Quian Quiroga and his co-workers were fascinated. They theorized that the specialized nerve cells were crucial to the process of recognition. Their locations were in the hippocampus, entorhinal cortex, parahippocampal gyrus and amygdala--all structures in the medial temporal lobe known to be involved in long-term memory. But how are we to conceive of a single neuron capable of representing something as complex as the identity of Bill Clinton?
From the point of view of information theory, this question is not hard to answer, according to computational neuroscientist Christof Koch of Caltech, who was also involved in the study and has been working with Fried's team since 1998. In his book The Quest for Consciousness (Roberts & Company Publishers, 2004), Koch illustrates this premise with an analogy. When we turn on the TV, the screen presents us with an explicit--that is, immediate--pattern of multicolored pixels distributed over the monitor. Yet implicitly concealed within this pattern is specific information, such as data about Bill Clinton's face.
Let us assume that a robot is tasked with determining whether the ex-president's image is currently on the screen. Its electronic brain has to expend enormous computational resources to extract the concealed information from the array of pixels. The computation involves many iterations, with some level of screening for Clinton-like information going on at each one, and each iteration involves a more and more sophisticated Clinton search through a smaller and smaller set of screened data. Whereas the initial mass of data shrinks with each computational step, "the logical depth of processing" increases steadily. In the end, a minute quantity of information--one bit--remains, indicating explicitly whether Clinton is present or not: 1 (Bill) or 0 (no Bill).
According to a theory of consciousness developed by Koch and his late colleague and friend, the Nobel Prizewinning Francis Crick, our brain proceeds in similar fashion. From its initial impression on the retina to actual consciousness, Clinton's face generates a firestorm of neuronal activity. But whereas many groups of neurons are involved at the lower processing levels, such activity is limited to fewer and fewer nerve cells in subsequent steps.
"I'm not claiming that a single cell represents the total neuronal correlate of Bill Clinton," Koch emphasizes. "The firing of a single neuron would be much too weak a signal." Nevertheless, he considers it probable that the concerted activity of a small group of neurons would be strong enough to catapult Clinton into consciousness. Because these cells encode only the abstract idea of Clinton, the tilt of the head in relation to the picture or whether Clinton is wearing a ski cap has no effect whatsoever on the behavior of the cells. And if we were to destroy all these cells? Then the perception of "Look, there's Bill Clinton" would turn to "Look, there's a guy who looks familiar, but I can't quite place him."
Quian Quiroga has been able to monitor simultaneously as many as 40 neurons--a number that researchers in the 1980s would not even have dared to dream of--with the help of implanted electrodes. Still, the question arises about whether the probability of finding a specific "Jennifer Aniston cell" from among the millions of neurons that make up the temporal lobe might be infinitesimally small. Koch agrees: "However, we theorize that there are numerous cells dedicated to familiar persons or objects: our grandmother, the dog, my laptop, etcetera"--and knowledge about what is familiar can help guide scientists who are looking for single cells. For this reason, the researchers questioned each test subject about his or her interests prior to testing. Only then did they select the approximately 100 images to show to that patient.
The sparse coding advanced by Koch and his colleagues differs fundamentally from the conventional notion of how persons and things are represented in consciousness. According to the theory of distributed representation, large and widely scattered groups of neurons fire off in the brain for any given person or object. Each individual cell contributes only a minute portion of the total data--which is why it is not too consequential if some of this information gets lost.
Further, the groups of cells are not dedicated exclusively to one particular face but are involved in the identification of a large number of people. The specific pattern of powerful neuronal firing is what signals who is being recognized. Researchers have no doubt that divided representation in the brain is a reality for certain tasks. This type of processing is how the brain makes out new faces, Koch explains.
Together with Crick, Koch developed a theory of the neuronal correlates of consciously perceived phenomena. When processing visual information, for example, neuronal groups that may be widely separated from one another unite in the sense that they fire in unison. Various coalitions of neurons stand for alternative interpretations of the particular thing or happening. Which of these interpretations comes to predominate depends on which characteristics of the image our brain pays the most attention to.
Grandmother cells function differently, Koch theorizes. "Thanks to highly specialized cells, we recognize our own grandmother immediately in the crowd of other elderly ladies at the senior citizen home, without having to think twice about it."
