To the Editors:
The article on Neural Darwinism [NYR, September 15, 1986] gives a good account of Gerald Edelman’s work. I like the point of view very much since I have used the comparison between neural memory and Darwinism for many years. In my Hughlings Jackson lecture of 1979, “Learning as a Process of Selection and Amplification,” I put together the evidence that memory involves selection. Edelman has carried the idea much further and above all his great discovery of CAM molecules shows how groups of neurons may be formed. As yet there is little evidence about the actual presence of such groups. The article would have been strengthened by some mention of the evidence of Perrett and others that there are single cells in a monkey’s brain that respond when a face is seen. Some cells will answer only to the presence of one particular face (either human or monkey).
Edelman’s ideas about multiple maps in the brain are very appealing. There is good evidence for their existence but there is very little hard knowledge about whether they interact in the ways suggested. It would have strengthened the article to refer to the book of John O’Keefe and Lynn Nadel on The Hippo-campus as a Cognitive Map.
But of course there could have been reference to many other sources. The article is most useful as calling attention to the idea of memory as a process of selection, which I hope will become generally accepted.
University of Oxford,
To the Editors:
I have read Israel Rosenfield’s article on “Neural Darwinism” with pleasure and benefit. It is a very perceptive and clear summary of Edelman’s very important ideas. One matter may concern your readers. Edelman’s theory is presented as based on evidence and arguments derived from animal studies. Yet at the end the theory is invoked as an explanation for language and other human mental activities. It would seem that an important link has been skipped.
Francisco J. Ayala
University of California,
To the Editors:
The article by Israel Rosenfield entitled “Neural Darwinism: A New Approach to Memory and Perception” is an excellent piece of scientific reporting. We are, indeed, in a period of transition, when concepts held for hundreds of years about the essence of the neural and of behaviour—reflexes, inborn genetically determined behaviour patterns, and above all learning—begin to change towards the Darwinian concept of selection. Where, if at all, would the principle of instruction be used by nature, if not in the nervous system, where learning by trial and error on the most primitive level and teaching by the parents and the group in more complex animals would seem to be essential for survival? The apparent helplessness of the newborn human infant, who seems to be devoid of anything neural apart from a very few inborn reflexes (this is not so in reality, but appeared so to us since times unknown), forcefully suggested the classical concept of a “tabula rasa” and everything that it entailed. The emergence and unprecedented victory of the Darwinian concept of selection, its vindication by molecular biology, modern genetics, immunology, and even in ethology, or its more controversial extension into sociobiology, did not immediately change our basic understanding of the neural as a predominantly “instructive” phenomenon.
Certain naive insights, though, pointing to some kind of “selection” in neural functions were around much earlier. When Paul A. Weiss, the brilliant Austrian-born American neurobiologist, made his early experiments on tissue and organ recombinations, he interpreted his observations on the movement of foreign limbs transplanted into host animals—in species having a very high capacity for regeneration—as suggesting some kind of “resonance principle.” It appeared as if the supernumerary limb, when it became innervated by the host nervous system, were able to extract (or select), in spite of its heterotopic position, the impulses of the host nervous system addressed to its own analogous limb.
This was, of course, an absurd concept in view of our general knowledge on neural organization and function, and very soon Weiss himself abandoned the idea. Some years later, in the 1950s and 1960s, one of my pupils, George Székely (now professor of anatomy at Debrecen University Medical School, Hungary), and I were experimenting in somewhat similar direction. It was our aim—and here is the connection with Rosenfield’s article—to refute the claims of Roger W. Sperry—about a genetically predetermined specificity in nerve elements that would decide about what neuron in the developing nervous system would get into contact with which other neuron. I intensely disliked (probably purely intuitively) the Sperry concept and we tried to beat Sperry with his own weapons. (In order to avoid misunderstandings, it ought to be mentioned that Sperry received the Nobel Prize in 1981 not on the basis of his specificity theory, but for his ingenious split-brain experiments, one of the major discoveries of our era.)
We did not succeed, partially because Sperry was a superior experimentalist, but partially also because most of the experiments left ample room for interpretation in either direction. However, during this phase of our studies we came across quite a number of phenomena (still unexplained) suggesting very forcefully the idea that in spite of experimentally induced misconnections the nervous system is able in some way or another to redirect appropriate signals to the respective organ or tissue placed into a heterotopic position. (For a brief account including the bibliography of the original papers, see my article in Specificity and Plasticity of Neural Structures and Functions, Brain and Behavior, ed. M.A.B. Brazier, American Institute of Biological Sciences, Washington, 1961, pp. 49–66.) We did not pursue the matter because we lacked then the means to follow up the details, as it could have been done soon, albeit at the cost of inordinately large effort.
As often happens in the history of science, some controversial issue cannot be settled unless somebody comes along with an entirely new approach and technical repertoire. This is exactly what happened with the discovery by G.M. Edelman of the CAM mechanisms, correctly explained by Rosenfield. I would be quite happy with this solution; Sperry’s specificity theory is soundly refuted; or to express the point more correctly, we can see now how the emergence of the miraculous connectivity of the nervous system might eventually be explained. The non-expert readership should not however be given the impression that CAM mechanisms could fully account for the specific connectivity as we can observe it in the numerous known “unique” neurons and even unique connections in some invertebrates, and the very high specificity in the vertebrates in terms of neuron populations. The specificity of prefunctional addressing (c.f., Patricia S. Goldman-Rakic, “Modular Organization of Prefrontal Cortex,” Trends in NeuroSciences, 7:419–429, 1984) is much too good to be fully explained today. I happened to be one of the few neuroscientists who was arguing for sufficient “randomness” in connectivity against strong resistance and occasionally ridicule by the neuroscience community (who were too much “brainwashed” by the elegant Sperry concept). (See my article “Specificity Versus (Quasi-)randomness in Cortical Connectivity,” in Architectonics of the Cerebral Cortex, eds. M.A.B. Brazier and H. Petsche, Raven Press, 1978.) It was only during the last few years that the development in tracing connections of individual neurons made it possible to assess in almost quantitative terms how “random” (or better: indiscriminate) neuronal connections may be on the microscopic scale (i.e., within spaces of few hundred micron diameter). Certainly, this “non specificity” in connectivity is ample enough to satisfy the requirements of the theory of G.M. Edelman for a “degenerate” (in analogy to how triplet code in molecular biology is degenerate) repertoire of connections in neural networks.
The theory proposed by G.M. Edelman under the title “Group Selection as the Basis for Higher Brain Functions” (or in another more explicit version “Group [Degenerate] Selection and Phasic Reentrant Signaling: A Theory of Higher Brain Function,” the full bibliography is given in Rosenfield’s article) is probably the most impressive and most convincing “selectionist” theory so far. Additionally it is intellectually appealing through the beauty of the concept; its aesthetic qualities, i.e., in the supreme elegance of modeling automata of networks capable of some limited pattern recognition by selection generating de novo associative memory, without the need for a priori instructions. (Edelman and Reeke, Proc. Natl. Academy Sci USA 79, 2091–295, 1982.)
In spite of its unquestionable appeal, we should be aware that this theory is only one among several contemporary brain theories that place the principle of selection at their very center (for example the theory of J.-P. Changeux in France: “L’homme Neuronal,” Fayard, Paris, 1983; there is an English translation available under the title: “The Neuronal Man”). But Changeux’s theory has some advantage in being based much less on abstract reasoning than on direct neurobiological (molecular neurobiochemical) observations. In addition there are several other brain theories in which selection plays a more modest role, however, they cannot even be mentioned here. Although modern neuroanatomy supports Edelman’s theory on practically every account—occasionally beyond the expectations and explicit claims of its author—no refutation of any localized memory trace and specific memory molecules, and no resorting to mechanisms of changing and bringing continuously up-to-date selected groups of connected neurons will explain away our fundamental dilemma of having to account for, say, the 60 to 70 years of unbelievably vivid remembrance of early childhood events commonly experienced by the elderly. Even Edelman’s theory requires some change in the strength of connections (the efficacy of synapses)—only more distributed than in the early naive concepts. Hence they must have some material basis; and we are back, albeit at a higher level, in our original predicament.
Finally one word of caution: Darwinian revolution and the principles of variability and selection are magnificent and fundamental insights for the whole realm of the living. But is it not that the pendulum of our thinking is swinging out towards one of its possible extreme positions? Or are we approaching a level of insight that will stay with us for good? I do not think so. Let me cite one example: In a remarkable book Dougal Dixon (After Man, Granada Publishing, London, 1981) portrays the animal kingdom as it will live on this planet in 50 million years, long after the disappearance of man and his destructive interference with the environment. Dixon applied in a superb way the principle of Darwinian selection and showed adaptations to the assumed new environment of the animal genera that have the greatest chance to survive. He was aided in his efforts by a number of excellent artists. There is only one flaw in this new animal kingdom: the creatures, although magnificently adapted, are without exception very, very ugly. What went wrong? (Although I try my best to overcome my conservative instincts and tastes.) The Darwinian mechanism, as it is presently understood is “sterile” in a certain sense and lacks the “creative” drive (please, do not mix this up with some “creationalist”: anti-intellectual, antiscience, and paradoxically also antireligion rubbish) that we can experience in nature and also in the positive efforts of man. The theory (of Edelman) is beautiful, as good as anything possible for the time being, but let’s wait and see and learn something from modern physics also in things philosophically relevant….
To the Editors:
Israel Rosenfield’s article, “Neural Darwinism,” is excellent. Gerald Edelman brings fresh light to our efforts to understand the nature of memory and perception and their basis in the brain. The mainstream efforts of the past have been to pose the behavioral or psychological question and then ask what we know about the brain that could answer the question. So behavioral analysis may reduce simple learning to new stimulus-response connections and then ask how those connections are made in the brain. This is what Pavlov did, thinking in terms of cortical areas of excitation and inhibition that irradiate to each other. The more recent approach is to try to specify the new behavioral connections in terms of events that take place at specific synapses.
There is much value to this approach, but our ignorance is still so great that we need the benefit of other approaches. What Edelman has done is to begin with certain principles of organization of cell function, beginning with concepts from immunology and looking at how organization of the nervous system comes about in the course of neuronal development, that great experiment of nature. Then he arrives at the concept of “selection” for the organization that development yields, based on cell-adhesion principles. This leads to the concept of brain maps, not so different from Pavlov’s cortical areas in principle but based on modern, detailed information about neuronal maps and their plasticity in the face of sensory input. It is this substrate which yields learning and memory and helps us define what the principles of learning and memory should be, leading in a remarkable way back to what Bartlett wrote in 1932.
Bartlett was off the mainstream of the atomistic stimulus-response psychology that fit so well with synaptic neurophysiology. His concepts of learning and remembering are still with us today, however, and demand a neural substrate more complex than synaptology can provide. Edelman’s findings and conceptions about neuronal organization provide the needed substrate and require that we think about learning and memory in new ways that can deal with greater levels of behavioral complexity, and they do the same for what they require of how we think about perception as well. In other words, instead of just starting from the behavioral questions and looking for appropriate neural substrates, we can also start from what we know of cellular principles and neuronal organization and specify how memory and perceptions must be organized if they are to be yielded by such a neuronal substrate.
University of Pennsylvania,
Israel Rosenfield replies:
Professor Young did indeed suggest a relation between memory and a particular kind of selection many years ago, and this work has been referred to in Edelman’s own writings. As Professor Young notes, Edelman’s theory goes beyond these original ideas, among other ways, in relating the origin of variability in the nervous system to embryonic development, as well as in relating the formation of perceptual categories to selection.
Experimental results suggesting that single neurons respond specifically to particular faces are not in my view convincing, in part because they fail to account for simultaneous activity elsewhere in the nervous system. If, for example, the visual system of the brain can abstract “faces” and even a particular face, among its many ways of abstracting a scene, but cannot identify which abstractions (or categorizations) are “faces,” and which “a particular face,” and so on, then other neurons (or more accurately, neuronal groups) will clearly be necessary to relate these abstractions to the given context. In such cases entire patterns of brain activity would “represent” any particular face, and not the activity of a single neuron. It would seem that brain function requires a constant interaction among many neuronal groups and that recognition does not depend on the activation of any one neuron (or neuronal group), but on patterns of activity. Hence the importance of correlations among neural maps (“reentry”), central elements of Edelman’s theory for which there is now neurophysiological evidence. Furthermore, the experiments Professor Young mentions have not shown that the neurons that appear to respond to particular faces are not also responsive to other stimuli that may, or may not, be quite unrelated to faces. According to Edelman’s theory one would expect that neuronal groups would respond to many different stimuli—and for this phenomenon, called “degeneracy,” there is also neurophysiological evidence. A few examples of stimuli to which the neurons do not react would hardly prove very much, since they may be activated by stimuli not imagined by the experimenter.
Professor Szentágothai’s generous comments give a fair warning about the discussion of memory in my article. At the present time, we cannot fully account for the vivid recollections of past events. And Professor Ayala is justified in posing a question about the final paragraphs of my article. Neural Darwinism is a theory about perceptual categorization—the ways in which visual, auditory, and tactile stimuli can be formed into perceptual categories by the nervous system. As Professor Ayala implies, the relationships between the formation of perceptual categories and other functions such as language are not direct or obvious ones, and they are currently the subject of extensive research.