The Mind Within the Net: Models of Learning, Thinking and Action

Manfred Spitzer is a psychiatrist in Germany who has conducted extensive research examining language and brain function in normal subjects and patients with schizophrenia. He has now written a book reviewing neurobiological and computer simulation studies of neural networks with an eye toward applying these results to the study of psychiatric disorders. The central assumption of this book is that any attempt to understand how the brain works must consider how vast arrays of neurons interact cooperatively to produce intelligent outputs.

A number of principles of network function are described, including parallelism (the concept that neural computation involves simultaneous interactions of large numbers of neurons), distributed coding (that networks represent information as patterns of activation across neuronal populations rather than activation of single neurons), and incremental learning (a method whereby neural networks “converge” on an optimum distribution of connection weights to perform complex cognitive tasks). These topics have been covered elsewhere (1). The uniqueness of Spitzer’s achievement is his weaving together studies from a range of disciplines—cognitive psychology, animal neurophysiology, and human neuroimaging as well as computer-based neural network simulations—that illuminate these principles in simple and reader-friendly terms. The result is a highly accessible story about the dynamic nature of mind and brain.

Several sections of the book are particularly impressive. These include the discussion of simulated neural network learning of syntax by Elman considered in the light of language acquisition in children. Another section reexamines studies of the visual cortex by Hubel and Weisel. A neural network simulation by Sejnowki and colleagues is discussed; this work suggests that the primary task of neurons in the primary visual cortex is not to detect edges and bars but to produce three-dimensional reconstructions of objects in space based on two-dimensional patterns of shade and light. This task is critical for perceiving objects in the real world but is “invisible” to standard receptive field mapping methods. Research by Merzenich and colleagues using monkeys and human subjects with cochlear implants is examined to illustrate the extraordinary capacity of cortical networks to reorganize themselves throughout life. Spitzer’s own research of semantic processing in normal subjects and in patients with schizophrenia is discussed in terms of spreading activation network models.

An increasing number of studies emphasize that psychiatric disorders arise from complex neurocircuitry distributed over multiple brain regions rather than a single, localized “lesion.” Studies of actual and simulated neural networks are likely to become increasingly relevant to these efforts. This book provides an excellent review of the critical concepts and, moreover, makes for lively, entertaining reading.