In the past decade or two, we have learned an enormous amount about the workings of the brain from neural network models. Manfred Spitzer's book enables the non-specialist reader to explore the range and significance of these discoveries. He begins with the basics of neurophysiology and chemistry, contrasting brains with computers, and thus introduces the crucial concept of parallel distributed processing. Then he explains the mathematical basis of artificial networks which, despite their relative simplicity, can carry out pattern recognition and learning tasks. From this foundation he proceeds to progressively more complex network models which seem to mirror structural principles embodied in subsystems of the brain. The physiology of interconnections within the brain is also gradually explained, so the book gives us a running comparison-and-contrast between human brains and neural networks. Throughout the exposition, each new concept is illustrated by at least one application to specific brain functions (or dysfunctions). Many of these applications, especially in the final section, are medical.
Manfred Spitzer is a psychiatrist (Professor and Chairman of the Psychiatric Hospital at the University of Ulm, Germany). This invites comparison with other medical practitioners who have written for the general public, such as Oliver Sacks, Peter Kramer and Antonio Damasio. Spitzer deals with a number of mental dysfunctions such as schizophrenia, depression and Alzheimer's disease, but he eschews the anecdotal approach which is prominent in the writings of his better-known peers, preferring instead to focus on principles suggested by neural network models and their implications for diagnosis and treatment. This impersonal approach is likely to keep the book off the best-seller lists -- which is a pity, because it captures a vital part of what made the 1990s the ‘decade of the brain’.
While Spitzer avoids needless jargon, he also avoids ‘dumbing down’ his subject. He makes it clear throughout that models are only one component of scientific investigation, useful because they can be tinkered with in ways that living brains cannot, and because they allow us to simplify and isolate certain functions that are fully and inextricably integrated in living brains. The book demonstrates the interplay among models, observation of living brains and patients, and neurophysiology, each of them suggesting new lines of investigation to the others. The most detailed examples of this interplay are Spitzer's reports of his own research into phantom-limb phenomena and language disorders among schizophrenics.
Much of the book deals with language issues, and this is where Spitzer comes closest to being controversial. Neural network models, even fairly simple ones, have succeeded in learning certain features of human languages without being ‘programmed’ with any syntactic rules. Spitzer presents this as evidence against the Chomskyan hypothesis that some syntactic ‘rules’ or structures are innately represented somehow in human brains. One of his claims is that Elman network models allow us to account for the ability of children (and the inability of adults) to spontaneously develop complex creole languages in an environment where they hear only pidgin language. But here Spitzer may be exaggerating the explanatory power of neural networks. Elman networks, by Spitzer's own account, ‘extract’ structures from their input; whereas children raised by pidgin speakers apparently receive no input from which the syntactic features of a creole could be ‘extracted’. Indeed Pinker in The Language Instinct (Spitzer's own source) cites children's ability to create creole languages as the strongest evidence for the very hypothesis that Spitzer denies—that their ability to generate these complex structures and ‘rules’ must be in some sense innate.
The book is not entirely free of errors on the proofreading level. Probably the most serious is the labelling of a key diagram early in the book (p. 26) which illustrates the crucial concept of synaptic weights. These weight factors can be positive or negative, but the signs are omitted from the weight labels in the diagram, making it difficult to follow the explanation in the text. Generally, though, the diagrams and captions (and the glossary) are very helpful, and the proofreading errors few and minor. The style is admirably lucid -- perhaps surprisingly so, considering that Spitzer did his own translation from the original German (Geist im Netz), with acknowledged help from the MIT Press editors.
Spitzer calls his final two pages ‘User's Manual for Your Brain’, which at first glance seems whimsically grandiose. But he clearly believes that a better understanding of the network structure of the brain can be of practical value to anyone. His book does a good job of building such an understanding from the neuronal ground up, and I would freely recommend it to anyone with a serious interest in how the brain works.
[ gnoxic studies (home) ]