"Through a Computer Darkly: Group Selection and Higher Brain Function" 36, No. 1 (October 1982)
"Neural Darwinism: Population Thinking and Higher Brain Function"
"Group Selection and Phasic Reentrant Signaling: A Theory of Higher Brain Function"
"Group Selection as the Basis for Higher Brain Function" ed.
"Neuronal Group Selection in the Cerebral Cortex"
"Cell Adhesion Molecules"
"Expression of Cell Adhesion Molecules During Embryogenesis and Regeneration"
"Interaction of Synaptic Modification Rules Within Populations of Neurons" (February 1985)
"Selective Networks and Recognition Automata"
In 1895, Sigmund Freud made his last attempt to explain the neurophysiological basis of the way the brain functions. His essay on the subject, “Project for a Scientific Psychology,” was never published during his lifetime. We have learned much about the brain since 1895, yet no equally ambitious attempt has since been made to examine the broad implications of neuroscientific research for the functioning of the brain and for psychology. Recently, Gerald M. Edelman, director of The Neurosciences Institute at The Rockefeller University, has proposed a new theory, one that gives us powerful reasons to revise our ideas about how we think, act, and remember. Although this theory is not directly based on Freud’s work, it confronts several of the problems with which Freud wrestled throughout his creative life.
Central to Freud’s work was the connection between memory and the psychology of everyday life. He considered memory to be a permanent record of past events, a record that was anatomically separate from the brain mechanisms that are responsible for our ability to make sense of the world around us. As he wrote in the final chapter of The Interpretation of Dreams,
[T]here are obvious difficulties involved in supposing that one and the same system can accurately retain modifications of its elements and yet remain perpetually open to the reception of fresh occasions for modifications…. [Therefore] we shall distribute these two functions on to different systems.
On December 6, 1896, Freud wrote to his close friend Wilhelm Fliess,
As you know, I am working on the assumption that our psychical mechanism has come into being by a process of stratification: the material present in the form of memory-traces being subjected from time to time to a re-arrangement in accordance with fresh circumstances—to a re-transcription. Thus what is essentially new about my theory is the thesis that memory is present not once but several times over, that it is laid down in various species of indications.
In the same letter he writes, “If I could give a complete account of the psychological characteristics of perception and of the [registrations of memory], I should have described a new psychology.”
Freud was acutely aware that recollections are often imperfect and fragmentary, and that they can and do alter perceptions. His theory attempted to explain how what he took to be perfect stores of memory were so transformed, arguing that memories cannot be released in their permanent form because the satisfactions and pleasures once associated with youthful impressions can no longer be experienced directly. Hence they reappear in dreams, but disguised and reworked. Ideas, Freud argued, become separated from associated emotions (affects) and disappear from consciousness. The emotions become attached to apparently unrelated ideas, disguising their real meaning. And we often appear to forget the memories themselves. Repression, screen memories, latent dream content, the return of the repressed—all were mechanisms elaborated in Freud’s theory to account for the ways in which fixed memories, however distorted and incomplete, can manifest themselves and affect our present view of the world. Freudian theory attempts to account for an apparent paradox: if we believe that memories are, by their very nature, permanently stored in the brain, why are they rarely recalled in their original form? It is the inaccuracy of recollection that Freudian psychology evokes so well. The reasons for this apparent inaccuracy may, however, be quite different from those that Freud suggested. In fact, the assumption that memories are in any sense part of a fixed record may be wrong.
If memory is a fixed record, neurophysiologists still cannot say precisely where and how memories are stored. The hypothesis of a fixed record may have been formulated prematurely, without sufficient attention to the means by which we recognize objects and events. We are probably much better at recognition than we are at recollection. We recognize people despite changes wrought by aging, and we recognize photographs of places we have visited and personal items we have misplaced. We can recognize paintings by Picasso and adept imitations of Picasso. When we recognize a painting that we have never seen as by Picasso or as an imitation, we are doing something more than recalling earlier impressions. We are categorizing: Picassos and fakes. Our recognition of paintings or of people is the recognition of a category, not a specific item. People are never exactly what they were moments before and objects are never seen in exactly the same way.
One possible explanation for this is that our capacity to remember is not for specific recall of an image stored somewhere in our brain. Rather it is an ability to organize the world around us into categories, some general, some specific. When we speak of a stored mental image of a friend, which image or images are we referring to? The friend doing what, when, and where? One reason why the search for memory molecules and specific information storage zones in the brain has so far been fruitless may be that they are just not there. Unless we can understand how we categorize people and things and how we generalize, we may never understand how we remember. Yet we do remember names, telephone numbers, words and their definitions. Are these not examples of items that must be stored in some kind of memory? Notice, however, that we generally recall names and telephone numbers in a particular context; each of our recollections is different, just as we use the same word in different sentences. These are categorical, not just specific recollections.
Clinical neurologists have long been aware that brain disease may lead to severe alterations in memory, but they have yet to analyze deeply the nature of categorization. In a rare abnormality resulting from brain damage and known as prosopagnosia, patients lose the ability to identify the faces of friends and well-known public personalities. But they can recognize faces as faces. And while they cannot identify their own car or their own coat, they do recognize cars and clothing as such. They apparently can recall general categories but cannot identify specific items. Something similar may occur in some forms of amnesia as well. Antonio Damasio and his colleagues at the University of Iowa Medical School described a patient with amnesia sitting in a room with the curtains drawn and unable to recall the season. When the patient looked out the window, he noted the color of the trees and the dress of a passerby, and exclaimed, “By golly, it must be July or August.” He could not recall the month of the year, but he could deduce it given appropriate evidence.
These studies appear to suggest that our ability to recognize general categories as opposed to the recognition of specific items such as Mary’s face or Alison’s hat depends on two different brain functions. But the ability to recognize Alison’s hat is, in part at least, based on temporal associations. We may have seen Alison wearing that hat last Sunday. The loss of the ability to categorize events in time can cause a nearly total loss of specific references. It is not the specific items that are no longer recalled, but their temporal order or their arrangement in succession that has not been formed or has been lost. When Damasio and his colleagues examined the man with amnesia about the calendar year, they found he had brain damage which made him unable to establish “temporal and spatial relationships between separate sensory information items.”1
Individual needs and desires, then, determine how we classify the people, places, and events that fill our daily lives. Moreover, the categories we use seem to depend on cross-correlations, or context. Yet many influential theories of mental function posit fixed entities that have an independent existence of their own. Freud, for example, described many ordinary objects as fitting into categories based on their resemblance to male or female sexual organs (phallic symbols, for example) and tended to view such categories as representing deeper drives that are universal within the human species. Many clinical neurologists and psychologists disagree with Freud’s notion of universal sexual drives; they nevertheless hold that information is organized into permanent categories in one or more memory systems within the brain, and that it can systematically be brought to consciousness in ways analogous to memory searches used in computers. The processes that are responsible for our recognition of categories, however, do not seem to depend on such fixed mechanisms.
There are good biological reasons to question the idea of fixed universal categories. In a broad sense, they run counter to the principles of the Darwinian theory of evolution. Darwin stressed that populations are collections of unique individuals. In the biological world there is no typical animal and no typical plant. When we say a salt molecule has a specific size we are giving a measurement which, allowing for error, is true for all salt molecules. But there is no set of measurements that will universally describe more than the one example of a plant or animal we are measuring. Qualities we associate with human beings and other animals are abstractions invented by us that miss the nature of the biological variation. The central conception in Darwinian thought is that variations in populations occur from which selection may take place. It is the variation that is real, not the mean. It was Darwin’s recognition of this profound difference between the biological and physical worlds that led to the rise of modern biology. The mechanisms of inheritance through genes create diversity within populations; selection from these populations allows certain organisms to survive in unpredictable environments.
Darwinian ideas have had a variable influence on psychological thinking, which has sometimes strayed away from biological explanation. Modern ethology, which studies the relation of animal and human behavior, has recaptured much of the Darwinian flavor that unfortunately left psychology when early learning theorists such as Pavlov seemed to be successful in explaining behavior without paying heed to the differences between animal species. But as important as their insights are, ethologists have not applied Darwinian thinking to the workings of the brain in each individual of a species.
Does evolutionary thought have anything to do with the explanation of the psychology of individual human beings? The theory of the brain Gerald Edelman proposed in 1978 sought to explain neurophysiological function as a Darwinian system involving variation and selection. Although his theory is confined to neurobiology, implicit in this work is a bold attempt to unify the biological and psychological sciences, one that strongly depends on the ideas of evolution and the facts of developmental biology.
Edelman had earlier studied the immune system. For years, scientists had wondered how the body produced antibodies against viruses or bacteria it had never encountered. Linus Pauling had suggested in 1940 that there was one basic kind of antibody molecule in the body. When the body was invaded by a bacterium, he had argued, the antibody molecule would mold itself around the intruder, thus acquiring a definite shape. Copies of the mold were made and released into the bloodstream where they would bind to the invading bacteria. The system learned, or was instructed by, the shape of a bacterium only after being exposed to it.
Pauling’s theory that there is just one kind of antibody protein proved to be wrong. In 1969 Edelman and his colleagues worked out the complete chemical structure of the antibody molecule, providing the important clue to what structures within the molecule are varied to produce millions of different kinds of antibodies needed to protect the body against foreign organisms. For this work he won the Nobel Prize in 1972 along with the late Rodney Porter of England. Their studies confirmed a theory suggested in the 1950s by MacFarlane Burnet and Niels Jerne that all animals are born with a complete repertoire of antibodies and that intruding bacteria select those antibodies that can effectively combat their presence.
Contrary to Pauling’s theory, the presence of the bacteria does not determine the nature of the antibody that is made, but only the amount. A limited number of genes, a few hundred or a few thousand at most, provide, through recombinatory mechanisms, the codes for the many millions of different antibodies. Specialized cells in the blood each produce one of the many kinds of possible antibodies that then become attached to the cell surface. An antibody molecule that happens to fit more or less closely a virus or bacterium floating in the bloodstream will bind itself to the virus or bacterium. This sets off a chain of events that causes the cell to divide and make thousands of copies (clones) of itself and more of the same kind of antibody. Other cells may carry antibody molecules that fit the virus or bacterium in different ways, and these cells, too, will bind to the virus or bacterium and produce clones. The body can only rid itself of a virus or bacterium if there is at least one good enough fit in its antibody repertoire. Usually there are several fits and some of them may overlap.
So the immunological system is not taught what antibodies it has to make to rid the body of a particular virus. The invading virus selects the appropriate antibodies and these will be different in each individual. An unfortunate organism may not have any antibodies in its repertoire that can bind the virus, and this could be fatal. Scientists were generally pleased with this solution to the immunological question because it was consistent with Darwinian principles of selection that have formed the basis of modern biology. Theories of immunology based on a process of learning or instruction were not.
Comparison of those findings on immunology to the theory of evolution suggested to Edelman that the brain too may function as a selective system and that what we call learning is really a form of selection. The theory he worked out is based on three fundamental claims: 1) during the development of the brain in the embryo a highly variable and individual pattern of connections between the brain cells (neurons) is formed; 2) after birth, a pattern of neural connections is fixed in each individual, but certain combinations of connections are selected over others as a result of the stimuli the brain receives through the senses; 3) such selection would occur particularly in groups of brain cells that are connected in sheets, or “maps,” and these maps “speak” to one another back and forth to create categories of things and events.
But is it the case that there is a mechanism that creates such diversity in each brain? In 1963, the Nobel Prize-winning neurologist Roger Sperry proposed that the billions of complicated connections in the brain were each determined by specific chemical markers on each neuron. On this view, particular genes presumably provide a code for each of the markers. Twins with identical genes therefore should have identical, or nearly identical, brains. Learning may not have been preprogrammed in Sperry’s model, but this model seems to imply that what any person could learn was limited by predetermined connections in his or her brain. If Sperry were right, many brain functions would be genetically determined, and to this extent organisms could be limited in their ability to adapt to new environments. Adaptive behavior and flexibility would have to arise by instruction; i.e., by external stimuli creating patterns in the brain, much as programs give instructions to computers. Yet we know that animals can adapt individually to different environments. The human brain has permitted survival in remarkably varied circumstances throughout history. Genetic determinism strains credibility because it makes it difficult to account for the enormous variability of thought and action.
For Edelman, the important question was not, as Sperry had argued, how specific structures in the brain are made according to markers on each neuron, but how, given a particular set of genes, enough variability would be created within the constant overall structure of the brain to account for the adaptability of humans and higher animals in an unpredictable environment. The deeper issue that had to be explored to understand this was the relationship between genes, on the one hand, and, on the other, the form and structure—or “morphology”—of animals. Notwithstanding all the work on the genetic code, biologists still cannot predict the shape of an organism from the information in its genes. If dinosaur genes rather than dinosaur bones had been found we would never have known they were dinosaurs and could not have said what they looked like. Why don’t genes tell us about morphology?
The answer to this question, according to Edelman’s findings, lies in early embryonic development. As an embryo develops, cells divide, move from place to place, and ultimately become specialized. A cell’s fate, whether it becomes a liver cell or a nerve cell, depends on where it happens to be at the moment that specialization begins, as well as where it has been. Cell shapes and movements will inevitably vary in each individual, making it impossible to predict exactly where a particular cell will be at a given time. Therefore the genetic mechanisms that determine what a cell will become must somehow be sensitive to its location at a particular time. If each cell’s specialization were completely predetermined by genes, even one misplaced cell could create havoc and the organism’s subsequent development would follow an abnormal course. A supporting beam placed between two walls that arrived just moments too late would be of no use once the walls had collapsed.
How does a pattern emerge in the embryo from early activity of the cells? Biological systems are not built from preformed parts. Instead cells are cemented together in the embryo and the individual cells or groups of cells are then shaped into structures that serve more complex functions. In determining the shape, it is the signals across the boundaries of such structures that count. How this works may be suggested by an analogy.
Imagine an architect given the task of shaping a brick wall into the façade of a building, with windows, ornaments, and a main door. One way our imaginary architect may proceed is to indicate where the holes for the windows would be drilled by drawing the borders on the brick wall. The holes drilled and the window frames added, he might then add indications for the shutter hinges on the window frames. In general, shaping a relatively uniform collection of interconnected material units, such as bricks, into a well-differentiated structure requires determining the boundaries of the structures that have to be carved into the units or added to them. And those boundaries can only be determined once all the bricks (or other material subunits) are delivered or are in place. One brick might be substituted for another without changing the pattern. Above all, the boundaries will determine the function of the bricks, such as supporting a window or a door.
As tissue grows in the embryo, borders are formed demarcating the different functional parts of the organism; but obviously there is no architect. The borders are established between different groups of cells by different intercellular cements or glues, known as cell adhesion molecules or CAMs. Three kinds of CAMs were discovered by Edelman and his colleagues in the late 1970s; and more have since been found. Two of the CAMs appear on the surfaces of cells very early in embryonic development. One is called L-CAM because it was first discovered in the liver, and the other N-CAM since it was originally found in association with nerves. L-CAMs on the cell surface will only stick to other L-CAMs on the cell surface and the same is true of N-CAMs; L-CAMs and N-CAMs will not stick to each other. For these primary CAMs, cells can adhere to each other only when they have the same kind of CAMs on their surfaces.
The structure of the cell adhesion molecules themselves is determined by particular genes. And in early development of the embryo other genes regulate when the CAMs are produced. However, the exact amount and “stickiness” of the CAMs (which vary continuously as the embryo develops) depend on where the cells carrying them are and where they have moved, and the individual cell’s position is not under direct genetic control. Therefore the arrangement of groups of cells that are linked together by one kind of CAM will vary even in genetically identical individuals. These groups of cells send signals that turn CAM genes on and off, as well as turning on and off the action of genes that specify cell specialization. The entire sequence by which differentiation occurs is therefore determined not directly by genes but indirectly by the combined action of genes and signals from cell groups that activate genes, and is therefore called epigenetic.
Cells linked together in collectives by L-CAMs form borders with other cells linked together in other collectives by N-CAMs. The borders result because the two kinds of molecules do not stick to each other. Edelman and his colleagues have shown that cells on one side of the border will subsequently change into more specialized cells of one kind, and those on the other side of the border will become specialized cells of another kind. Throughout embryonic development, borders are formed between cells that are linked together with different CAMs, and following border formation the cells specialize. As the cells differentiate further, new CAM borders are formed before new changes are introduced by signals from the cell groups to the genes that activate both CAMs and the genes that specify cell specialization. Depending on the past history of the two cell groups, signals exchanged at the border between collections of N-CAM and L-CAM cells will determine the subsequent formation of very different kinds of cells on each side of the border. (Recently a chemical signal that activates CAM production has been identified.)
The function of the border between cell groups with different CAMs thus depends on the context—the surrounding cells, and the past history of the cells. In general, moreover, the rules governing CAM response are similar both for the neurons of the brain and for other body structures. Because the borders of cell collectives depend on dynamics of movement, there will be individual variations that are not determined just by genes and whose diversity will ensure that different brains would have different structures. Yet the general patterns formed and the broadly similar sequences of embryonic development would account for the fact that the individual brains of members of a species resemble one another.
The importance of the borders formed by the CAM-linked collectives was shown in a spectacular series of experiments in Edelman’s laboratory on the emergence of a single feather in the chicken. At each stage in the feather’s morphogenesis, borders were found between groups of cells linked together with different kinds of CAM. Following the formation of the borders each group of cells changed, those on one side into one kind of cell, and those on the other into another kind of cell. For example, the final feather, with its regular pattern of branching (a pattern that varies from feather to feather), is carved out of a cylinder of tissue. The distinctive feather pattern arises after alternating groups of cells (i.e., L-CAM-linked cell groups alternating with N-CAM-linked cell groups) make their appearance. Groups of cells with N-CAM linkages then die, while the alternating groups with L-CAM linkages become hardened, or keratinized. The result, a feather. The edges of the barbs on the feather were the borders between the L- and N-CAM cell groups. Even more dramatically, when changes were induced experimentally in the linkages made by one kind of CAM, cell groups that were linked by a completely different kind of CAM were altered.
From the workings of such epigenetic mechanisms we can see why knowing the entire genetic repertoire of an animal would not alone permit us to predict its final detailed morphology. Identical twins are never absolutely identical. And just as every feather has a different pattern of branching, every brain would be expected to have a different pattern of connections.
The demonstration that there are molecular reasons why no two brains could be identical is central to Edelman’s view that the brain functions as a system based on selection—the CAM mechanism creates diversity in the anatomical connections of an individual’s brain. The context and history of cellular development thus largely determine the structure of the brain; and therefore context and history are also important in brain function. Because development, structure, and function are related, it would not be surprising if the functional activities of a connected group of cells in the brain depended both on the activities of neighboring cell groups and on the past history of the particular group itself.
To show that brain function, like structure, also depends on context and history and not on localized functions and fixed memories is the burden of Edelman’s theory of neuronal group selection. What emerges is a new approach to the biological basis of psychology.
A major claim of this approach is that the unit of selection in the brain is a neuronal group, a set of interconnected neurons that function together. The patterns of connections that are established among neurons vary from group to group because of the changes in dynamics of the CAMs during development. The brain thus contains large numbers of different neuronal groups.
Neuronal groups are connected to one another as well as to the sensory receptors for light, touch, and sound in the eyes, skin, and ears. In general, neighboring groups of neurons in the brain receive input from neighboring sensory receptors (for example, on the second and third fingers of the hand); two neighboring groups in the brain can in fact receive input from the same sensory receptor. Although the inputs can overlap, the responses of each group to the stimuli will be different. Because each group of neurons has its own pattern of internal connections, which differs from those of other groups, each group will respond differently, even to identical stimuli.
The activities of a group of interconnected neurons would acquire significance not just because of the anatomical connections and physiological mechanisms on which the functioning of that group depends, but because of its context and the history of its received signals as well. If this is true, then a given “memory” cannot be stored in a specific place in the brain, since neighboring activities would of necessity change, and therefore the “context” of any neuronal cell group is never constant. If one were to assume that a memory was in fact stored as it is in a computer, altering this process would irreparably destroy it.
The embryonic processes I have described involving CAMs are central in creating a large repertoire of different neuronal groups. But after birth the principle of selection changes. Instead of alternations of CAMs, changes in the strength but not the pattern of connections occur; such changes determine the paths over which neural signals will flow.2 Environmental stimuli may cause one group to respond with greater activity than other groups receiving the same input. And when this happens, laboratory experiments show that the connections between the neurons in that group (the synaptic junctions) can be strengthened.
Edelman and his colleagues have worked out a set of possible rules that might govern these synaptic alternations. Molecular changes take place within neurons and at the synaptic junctions so that the neurons tend to be activated by similar stimuli on subsequent occasions. Particular variants within the brain’s population of neuronal groups are selected by the stimulus. Indeed, a group that responds to a stimulus might do more than this. As its connections are strengthened it might alter the strength of its links to other groups and, by competing with other groups, integrate neurons from them into its own response activity. The strengthening of the synaptic connections creates what Edelman calls a secondary repertoire: this is made up of neuronal groups that respond better to specific stimuli because they have been selected and their connections strengthened.
In their responses to stimuli, neuronal groups could be likened to a set of radio receivers, each tuned to a small band of frequencies. One radio might receive frequencies in the 1600 to 1700 kilocycle range, and another might receive frequencies in the 1550 to 1650 kilocycle range. Depending on the broadcast frequencies in the area, some of the receivers will respond to one broadcast, others to several broadcasts, and others to none at all. Analogously to an animal moving its head or body, moving from New York to Peking, for example, would change the response patterns of the individual receivers. The receiver’s purpose depends on where it is: in Peking it receives Radio Peking, in New York WCBS, and in Moscow Radio Free Europe, Radio Moscow, and a lot of jamming simultaneously.
Like the radio receivers, a given neuronal group can respond to more than one stimulus—what is called a degenerate response. In our analogy, a given radio receiver might pick up Radio Peking better than Radio Moscow, but it can pick up either station. (Of course, stimuli are not organized into coherent pieces of information like radio broadcasts. At higher levels in the brain, the stimuli must be organized in ways that will be meaningful and useful for the organism.)
The brain performs this organizing operation by using maps made up of neuronal groups. A map is a collection of neuronal groups in the brain which are arranged in a way that preserves the pattern of relationships either between a sheet of sensory receptors (such as those in the skin of the hand) and a sheet of neural tissue in the brain to which the sensory stimuli have been transmitted, or between two sheets of neural tissue. Groups are arranged in maps that “speak” back and forth to one another so as to create categories of things and events. Different kinds of maps are found in different parts of the brain, and an analysis of how such maps interact is an essential and final part of Edelman’s theory.
Because the brain has to be prepared for unpredictable events it must map stimuli in a variety of ways. Brain maps sort incoming stimuli by similarity (same frequency of sound, same intensity of sound, etc.) as well as by a mixture of properties. The main evolutionary principle at work here is that stimuli are organized into patterns that will help the organism cope with its environment.
In 1870 two young Germans, Gustav Theodor Fritsch and Eduard Hitzig, first discovered that the brain maps motor stimuli. They reported that touching discrete areas on a dog’s brain activated specific parts of the body. By the end of the century motor and sensory maps were a well-established feature of neuroanatomical teaching. These maps were assumed to be permanent and more or less identical in all members of a species.
It was therefore surprising when in 1983 Michael Merzenich and his collaborators at the University of California at San Francisco discovered that the sensory maps showed considerable variation in the brains of normal monkeys. The brain maps in a particular monkey’s brain varied over time. And there was considerable variation in corresponding maps in different monkeys as well. Subsequent experiments demonstrated that the maps became rearranged, even within short periods of time, following injury to a nerve supplying sensory input from one of the monkey’s fingers.
Merzenich’s work gives powerful support to Edelman’s claim that particular combinations of neuronal groups are selected competitively from the general population of neuronal groups by sensory input. Since the nerves from the different areas of the skin of the hand are connected eventually by overlapping branches to the same receiving areas in the brain, the part of the skin surface represented in a particular brain by a group of neurons depends on selective competition. Neuronal groups in one area of the brain may, for example, receive overlapping input from both the back and the palm of the hand, and the stimuli from the palm may more effectively select particular neuronal groups, establishing a dominant representation of the front of the hand in the map in that part of the brain.
Should the incoming nerves be damaged, reducing or eliminating the input from the palm (as in Merzenich’s experiments), the groups that could respond to the stimuli from the back of the hand will then be able to be expressed in the absence of competition from the neuronal groups in the palm. Thus, based on the inherent variation among neuronal groups, a new representation emerges in that area of the brain. The continuity of the new map’s general activity nonetheless can still represent, in some abstracted form, the activity at the sensory receptors.
Information in the brain is distributed among many maps, and according to Edelman’s theory, there must be incessant reference back and forth among them for categorization to occur. Sounds, for example, can be categorized as speech, noise, and music; or they can be used to locate things in space. Recent research shows that such localization requires a number of interacting maps. Owls, like human beings, use sounds to locate moving animals, for instance a mouse they might attack. The important sensory clues are the different arrival times of a sound at each ear, and its intensity. Since the owl’s brain cannot directly map the different times of arrival of a sound at each ear, two initial sensory maps represent the frequencies heard by the owl: one maps those heard in the right ear and the other maps those heard in the left ear. These representations are then combined in another map where the arrival times (called sound disparities) of a given frequency in one ear are compared to those in the other ear.
The sounds made by a mouse in a field can by this means be categorized according to the disparities in sound. These disparities can be used to help determine the sources of the sounds. Specific neuronal groups within a map may be activated by a difference in arrival time between the two ears of, for example, one thousandth of a second. In itself the activity of the neuronal groups will not tell the owl’s brain the source of the sound. But the entire pattern of activity of the map, the ways in which other neuronal groups are activated as well, will represent in the owl’s brain the location of the source of the sound. This pattern will have to be extracted in a further mapping, which could for example characterize certain patterns of activity indicating that sources of sound were, say, at 30 degrees to the left, others at 60 degrees to the right, etc. Finally the mapping that has derived the place of a source of sound is connected to a visual map of space, created from the owl’s visual receptors. The visual map is thus related to a map that recategorized auditory sensory input to place sounds in space. By relating the two sensory modalities, the owl’s brain has created a general map (auditory and visual) of space and the owl can respond to a variety of sensory inputs.
A particular pattern of activity will lead to a motor response in which the owl dives for its prey. If the owl is successful, it will associate that mapping and that pattern of activity with the particular motor act of attacking. If it fails, however, it will try other responses until it finally succeeds in capturing its prey. This was shown in the series of experiments by E.I. and P.F. Knudsen in which young owls were raised with one ear plugged, thus shifting the perceived location of sound relative to its actual location. Within four to six weeks these owls learned to localize sound accurately. The owls adjusted to the altered mapping of sound apparently by rearranging their internal mappings. Recognition therefore depends on mapped and remapped patterns of activity.3
No single map contains all the information necessary for the owl’s movements, and as I have said there must be a constant reference back and forth from neuronal groups in one map to neuronal groups in the other by means of so-called reentrant connections, i.e., nerves traveling in both directions to link the maps. According to Edelman’s theory, this is how the brain creates its categories and generalizations. Of course, the owl brain may also use the initial mapping of sound frequencies for higher maps that do not represent spatial disparities, but rather the actual sequence of sounds (to identify the kind of animal making the noise, perhaps). This will eventually create other kinds of categories for sound information.
The brain has many different kinds of maps and ways of mapping other maps that categorize “inputs” in many ways. The purpose of the maps is to create perceptual categorizations that permit the animal to act in appropriate ways.4 The environments in which an animal might find itself will of course change and so the perceptual categories must also change. But this is exactly what the multiple mappings are best suited for: the maps interact with one another and constantly recategorize information. And by referring the more abstract mappings back to the primary sensory maps that have a continuous relationship with external stimuli, the brain can effectively keep track of its various regroupings of the sensory inputs.
That mappings can be related to one another without any preestablished instructions has been demonstrated by Edelman and his colleague George Reeke, Jr., who built a new kind of automaton, based on the principles of selection, to simulate the mapping activity of the brain. The automaton abstracted from the mappings of visual inputs a variety of categorizations, such as for letters of the alphabet, without having been given specific instructions to do so. This provided further evidence that the interacting maps are essential for categorizing perceptions in a selective system.
A spy sitting in a music hall might want to locate the person he just heard say, “Nine o’clock tomorrow,” and he may also want to enjoy the singer’s “Casta Diva.” One set of brain maps will locate the person who said “nine o’clock,” while another set of maps will permit him to hear the “Casta Diva” for his own pleasure. Sounds have been categorized in different ways by his brain in accordance with his adaptive needs: business and pleasure.
Later that evening the spy may have forgotten the time he overheard mentioned by the person he was shadowing during the concert. Annoyed, he hums the melody of “Casta Diva” and suddenly recalls the nine o’clock assignation. Or he might recall it when he sees the announcement for a nine o’clock movie. This suggests that memory is not an exact repetition of an image of events in one’s brain, but a recategorization. Recategorizations occur when the connections between the neuronal groups in different maps are temporarily strengthened. Recategorization of objects or events depends upon motion as well as upon sensation, and it is a skill acquired in the course of experience. We recollect information in different contexts; this requires the activation of different maps interacting in different ways from those of our initial encounter with the information and it leads to its recategorization. We do not simply store images or bits but become more richly endowed with the capacity to categorize in connected ways.
Memory as recategorization is one of the deep consequences of Edelman’s theory of neuronal group selection. In a remarkable book published in 1932, the English psychologist Frederic C. Bartlett sketched out the view to which Edelman’s work has given a precision and a physiological justification. In Remembering Bartlett wrote:
Remembering is not the re-excitation of innumerable fixed, lifeless and fragmentary traces. It is an imaginative reconstruction, or construction, built out of the relation of our attitude towards a whole active mass of organised past reactions or experience, and to a little outstanding detail which commonly appears in image or in language form. It is thus hardly ever really exact, even in the most rudimentary cases of rote recapitulation, and it is not at all important that it should be so.
It is this quality of memory that Freud, too, sought to capture. Believing that memories must leave permanent traces, and unable to see how a perceptual structure could remain open to new perceptions if it were altered by previous stimuli, he constructed a theory quite different from the view of brain function presented here.
Unable to accept that fragmentary memories may well be fragmentary, Freud assumed memories were fixed in the same way that Newton had assumed time was absolute. Einstein eliminated absolute time and thereby presented a larger view of space and time. In dispensing with fixed memories and replacing them with memory as categorization, Edelman’s theory represents a radical departure from previous thought and may well open the possibility of a broader and deeper view of human psychology.
Each person according to his theory is unique: his or her perceptions are to some degree creations and his or her memories are part of an ongoing process of imagination. A mental life cannot be reduced to molecules. Human intelligence is not just knowing more, but reworking, recategorizing, and thus generalizing information in new and surprising ways. It could be that inappropriate categorizations from damaged maps may cause psychoses, just as the inability to correlate the succession of objects or events in time may be largely responsible for the loss of specific memories in the case of amnesia already mentioned.
Of course, language is acquired in society, but our ability to use it, to constantly reconceive the world around us, is at least in part a reflection of the multiple mappings and remappings that appear to be central to brain function. Such a view reinforces the idea that no two brains can be, or ever will be, alike. Edelman’s theory of neuronal group selection challenges those who claim that science views individual human beings and other animals as reproducible machines and that science is little concerned with the unique attributes of individuals and the sources of that uniqueness. Humanism never had a better defense.5
A.R. Damasio, P.J. Eslinger, H. Damasio, G.W. Van Hoesen, and S. Cornell, “Multi-Modal Amnesic Syndrome Following Bilateral Temporal and Basal Forebrain Damage,” Archives of Neurology, Vol. 42 (March 1985), pp. 252–259. Damasio et al.’s interpretation is quite different from the one I have suggested. ↩
This is analogous to the process by which antibodies that have been selected are then produced in large numbers, or cloned. The two processes resemble each other in effect, but the mechanisms are different: in the brain the strengths of synaptic connections are increased; in the immune system the number of cells is increased through cloning. ↩
E.I. Knudsen and P.F. Knudsen, “Vision Guides the Adjustment of Auditory Localization in Young Barn Owls,” Science 230 (November 1, 1985), pp. 545–548. ↩
The origin of perceptual categories by neuronal group selection is in some ways analogous to the origin of species by natural selection in evolutionary time. Much as unpredictable events over a long period of time may result in the selection of certain characteristics in organisms, unpredictable environmental events in an animal’s lifetime may result in the selection of certain neuronal groups leading to the formation of perceptual categories. ↩
I would like to thank Leif Finkel for his invaluable assistance in the preparation of this article. ↩