|
Review for Journal
of the International Neuropsychology Society
Churchland sees the brain as “The Little Computer
|
|
I find this book as exasperating as a bright teenager or an Italian sports car; it has enough that is wonderful about it that you can’t just take it out back and blow it to bits. On the one hand, the book provides an excellent introduction to connectionism (PDP; parallel distributed processing) and its powerful applications to cognitive neuroscience and to philosophy of mind. On the other hand, these treasures are contaminated throughout by oversimplifications, propaganda, and egregious omissions, including one of the most important implications of connectionism: its generalizability to systems at any level of organization, not just to brains or “computers”. For neuropsychologists, is it worth the effort? And would you want a student to read it without parental guidance? Churchland intends the book for the general reader. The first 180 pages are a very readable primer on how parallel architecture adds great new powers to what can be done with serial computers; how PDP properties are similar to brain and cognitive systems; the information-carrying power of vector coding with examples from taste, color, face recognition; how networks can learn, and develop the capacity for inductive inference and perception of causality. This will be grand for the beginner; the serious student should go directly to the more complete (and technical) The Computational Brain (1992). Following this conceptual foundation is 140 pages of what Churchland sees as the practical and intellectual implications for consciousness and philosophy of mind, as well as for language, politics, morals, medicine, the law, and the relations between art and science. He says (p. 19) that his primary aim is “not to urge or establish any particular philosophical doctrine” but to present for “the thinking public ... the character and potential significance of the developing theory and recent experimental results”, so that readers may reconceptualize their own mental life in neurocomputational terms instead of in terms of, for example, an immaterial soul. Neuropsychologists may ask if he is really interested in the neuro part of “neuro-computational”. For decades Paul Churchland and his wife-colleague Patricia S. Churchland have been outstanding among professional philosophers of mind in stressing the benefits, even the necessity, of understanding something about brains. Patricia, already known for Neurophilosophy (1986) and The Computational Brain (1992, with Sejnowski), has recently edited The Mind-Brain Continuum, (1996, with R. Llinas). Therefore it is puzzling to see how little attention this book gives to the brain at the neurophysiological and neuropsychological levels. The focus is on the micro-cognitive, the computational, and the partisan-philosophic. The first macro feature one notices about the brain is its division into two hemispheres connected by an enormous bundle of fibers. For some reason Churchland does not discuss the hemispheres or their differences, the corpus callosum, the thirty-five years of clinical and experimental research on the split brain in regard to the unity or duality of consciousness, styles of thinking, language, and self. Could he have left them out because he thinks all this is relatively unimportant for the thinking public’s conception of their mental life, or for the connectionist mechanisms he is presenting? Considering Churchland’s grace at exposition, the omissions cannot be explained as dumbing-down for the general reader. The only mention of hemispheric duality or lateralization is in the preface where he thanks Pat; “After twenty-five years...I often feel we have become the left and right hemispheres of a single brain”. On page 155 is a stereo MRI image of Pat’s brain. Stereo glasses come with the book, tucked in the back cover. Much of Part 2 in the book concerns consciousness, and so it is surprising that there is no mention of the facts and implications of Libet’s 30 years of research on the brain mechanisms of consciousness directly stimulating unanesthetized human cortex (Libet 1993). Also missing is any discussion of the work on metaphor as fundamental to all of cognition, not just language; Lakoff’s paradigm-busting book, Fire, Women, and Dangerous Things (1987), makes it into the list of recommended further reading but gets no comment in the text. Who are the other players, on or off stage? What filters passed or culled them? The philosophers Nagle, Jackson, Block, Searle, Dennett, are included in order to be refuted (and Penrose on quantum-consciousness). Antonio and Hanna Damasio are thanked for the neuropsychology, and Llinas, Crick and Koch represent neurophysiology and are endorsed. Some that seem conspicuous (to me) by their absence are Sperry and Bogen, D. M. MacKay (Univ. of Keele), Jackendoff, Rosch, Baars, Rumelhart and McClelland, W. C. Wimsatt, D. Chalmers. Edelman gets only a one line mention. The index is three pages and trivial: the cerebellum is in, the corpus callosum omitted; Gershwin but not Geschwind; Brubeck but not Bogen. No sign of Zangwill, H. Simon, J. Anderson, Hebb. But there are entries for Paul McCartney, Charlie Parker, Mozart, Twiggy, Mike Tyson, Ronald Reagan, and Betty Crocker, all single mentions. Although the book is intended for general readers that does not excuse inaccuracy in the name of simplification, significant omissions in the name of brevity, or childish noninformative illustrations (e.g., fig. 8.3). I have always liked Churchland’s style; graceful, often gentlemanly and avoiding arrogance. He does lose hold of himself a bit, however, in discussing Searle (pp. 204ff.); “one foot on the dock and the other in the dory”, whose arguments he calls thinly disguised rehashes of thoroughly discredited pre-scientific ideas, or smoke screens, or a “Betty Crocker’s Theory of the Mind”.), and understandably, he goes ballistic when he refutes Dennett. Churchland’s enthusiasm is one of his strong points; he is truly excited by cognitive neuroscience and wants to share it. The risk is falling into too much “golly, isn’t brain-science grand!” such as comparing the number of possible synaptic configurations to the volume of the universe expressed in cubic meters (p. 5). Although he denies that he is urging a particular philosophical doctrine, occasionally the excitement overflows and the book reads like a missionary’s conversion tract for Eliminative Materialism (the formal philosophical position that folk psychology is too dumb to save, and will wither away, being totally replaced by something like neurocomputations). To me, perhaps the most powerful aspect of connectionism is its generalizability across levels; it applies to much more than computers or brains. The fundamental idea is that very simple units can be connected into a network entity by very simple sorts of relations, and this entity can learn, develop concepts and categories, induce, anticipate, adapt. The theory does not limit what the units can be; it is not about transistorized electronic circuits or neurons. There is no reason why such a network could not be composed of larger or smaller units, such as individual people, whole computers, families or other social units such as economic systems, or subcellular systems, as long as they have the right sort of connections for transferring information. The time scale of the “mental” processes of such an entity would of course depend on its rates of transferring information, but that is not a limitation on the type of process that it could carry out. It is a great tragedy that the wishful metaphor “neural nets” has been widely adopted as the name of the whole class of PDP systems, regardless of the level of organization of the units with which it is implemented. In spite of his missionary zeal for connectionism, Churchland under-generalizes the power of PDP and limits the concept to the neurocognitive level. There are some fundamental conceptual issues in this book which form an invisible subtext, an unspoken background of assumptions. These points are much too large to argue here, but need to be flagged because they can lead astray readers who are not familiar with professional philosophy. The first concerns what an explanation is. Churchland has devoted thoughtful useful chapters to this in his excellent Neurocomputational Perspective (1989). Unfortunately the present book gives the impression that a thing or event is explained if you list its components and their relations. But components have components of their own, and so forth down through unbounded levels of organization, unless you take as an article of faith that there is a “bottom” somewhere. Furthermore, in complex systems with enormous feedback and feed-forward connections between levels it becomes impossible to order the levels and decide which is component and which is context (Wimsatt, 1976). Wimsatt proposes that the notion of a causal thicket is more appropriate than the notion of hierarchical levels. I prefer a heuristic definition of explanation that stresses its relative, context-bound aspect; a satisfying explanation is an account of enough of the variance for your purposes (Galin, 1996). This leaves open the choice of units of analysis (as in physics’ explanations of light as a wave or as a particle or as both). In contrast Churchland urges us to reconceptualize our mental life in neurocomputational terms, saying, “...it is now modestly plain that the basic unit of cognition is the activation vector, ...and the basic unit of computation is the vector-to-vector transformation, ...and the basic unit of memory is the synaptic weight configuration” (p.323). He has chosen these microcognitive constructs to be the basic units of analysis for the mind. I argue that we must choose units suitable to our purposes, and that depends on the level of organization that we want to describe or engage with. The serious problem is that Churchland often slips into a tone of argument that sounds as if he had found the level and the units, and all other levels and units are in some way less fundamental, lacking in explanatory power or “reality” because they are derived or defective. He periodically denies this explicitly, but slips right back. Patricia Churchland also denies providing ultimate explanations in Mind-brain Continuum (p. 287) and claims that she is concerned more with the pragmatic strategy of inquiry than with statements of faith about ontology (the nature of what is). However, she too slips back into statements like “...the hypothesis that awareness just is some pattern of activity in neurons”, and “...patterns of brain activity are the reality behind the experience” (p. 293, emphasis mine). This is the bad ontological kind of reductionism (nothing-but-ism), not the good explanatory reductionism (an account of the implementing mechanisms). It is dangerous because it slides easily into what I call “pernicious reductionism” which holds that the parts are more real than the whole, and that the smaller the parts the more real they are (Galin, 1992). Why should the readers of JINS care about philosophers’ cat fights? For anyone interested in mind and brain these issues are much closer to home than that. In a forthcoming book, The Taboo of Subjectivity, B. Alan Wallace of Univ. Cal. Santa Barbara sets out three important distinctions. 1. Natural Science is a method of inquiry. 2. Scientific Naturalism is an ideology (i.e., religion), the dominant one in our culture. It is a theoretical framework of interpretations of the results of scientific inquiry. Like all interpretations it is necessarily built on metaphysical assumptions, in this case: absolute realism (there is a reality out there), objectivity (reality is independent of point of view), and knowability (everything can be known). 3. Scientism is the fundamentalist perversion of Scientific Naturalism which, like other fundamentalisms, mistakes its assumptions for infallible, ontologically given absolutes. I am afraid Churchland himself occasionally falls into Scientism, exemplified in his curious attack on Christianity, Islam, and Judaism lumped together as “a continuing tragedy” (p.293-294). But there is much to recommend the book. Perhaps even its flaws: when we are not learning something uplifting from it, we can use it as a mirror held up to our own excesses. REFERENCES · Churchland, Paul M. (1989). A Neurocomputational Perspective: The Nature of Mind and the Structure of Science. Cambridge, MA: MIT Press · Churchland, Patricia S. (1986). Neurophilosophy. Cambridge, MA: MIT Press · Churchland, Patricia S., and Sejnowski, T. (1992). The Computational Brain. Cambridge, MA: MIT Press · Churchland, Patricia S., and Llinas, R. (1996). The Mind-Brain Continuum. Cambridge, MA: MIT Press · Galin, D. (1992). The Blind Wise Men and the Elephant of Consciousness. Consciousness and Cognition, 1, 8-11. · Galin, D. (1996). What is the difference between a duck? In Cohen, J., and Schooler, J. (Eds.), Scientific Approaches to Consciousness: 25th Carnegie Symposium on Cognition. (Pp. 445-450). Hillsdale, NJ: Lawrence Erlbaum. · Libet, B. (1993). Neurophysiology of Consciousness. Boston: Birkhauser. · Lakoff, G. (1987). Women, Fire, and Dangerous Things. Chicago: Univ. Chicago Press. · Wimsatt, W. C. (1976). Reductionism, levels of organization, and the mind-body problem. In Globus, G., Maxwell, G., and Savodnik, I. (Eds.), Consciousness and the Brain (pp.205-266). NY: Plenum. End
|
