Manfred Spitzer
MIT Press, £16.65, pp 360 
ISBN 0 262 19406 6
———————
Rating: ★★★★
We are told that the 1990s are the decade of the brain. With the unravelling of the human genome, there are few great scientific challenges except, perhaps, understanding the origin of the universe and consciousness. This new book seeks to provide a greater insight into the workings of the mind through neural networks (for once, the “net” in the title does not refer to the internet). Neural models may help us understand phenomenology as diverse as why children need to have structured play and why schizophrenics may experience auditory hallucinations. The book brings together an eclectic range of materials from medicine, psychology, psychiatry, neurobiology, and neural network theory.
The book is extremely well written and easy to read with plenty of fascinating titbits. How can the word “fish” be written as “ghoti”? What has Beethoven's ninth symphony got to do with the size of a standard CD Rom? Does it make any difference if one learns to play a stringed instrument before the age of 12? I particularly enjoyed the historical perspective. For example, I was surprised to discover that it was the great English scientist Sir Francis Galton who first used the method of word association in a series of introspection experiments, well before Sigmund Freud and Carl Jung. Spitzer should also be commended for the way he presents real research data. Many popular science books explain the results of research in general lay terms without empirical data. The excellent figures for this book present clear summary data, with detailed legends providing any necessary explanation. There is also an impressive array of references and a useful glossary.
Appropriately, Spitzer leaves the final chapter to more philosophical thoughts. In this age of artificial intelligence, he is cautious about what neural networks can and cannot achieve. Can computer models really tackle the complexity of such phenomena as emotions and personality—the essence of being human? Ultimately, neural networks are models or simplifications of an organic process. As such, they enable us to test hypotheses and dissect out processes that could not otherwise be examined. These models do not have to represent biological reality. If a neural model could really reproduce how the brain works would it still be a model? I leave the last comment to a cited quote from James McClelland, a leading figure in neural network research: “All my models may be wrong, but some of the principles we have discovered using them may be right.”