Philosopher and psychologist William James has called the relationship between knowledge and the brain "the most mysterious thing in the world". It is therefore no surprise that many of the most creative and incisive minds in numerous often tenuously related disciplines have sought to blaze or at least illuminate a pathway to understanding that relationship. With the advent of the computer age a potent if sometimes unhappy metaphor for the activity of mind and brain became available, as did a new arsenal of tools for investigating and theorizing about that activity. The metaphor is that the brain, like a computer, is an information processor, and can be understood using the same kinds of methods used in information and computation theory. This approach to the study of the brain and the mind grounds the numerous areas of research that have come to be known as the cognitive sciences. The hope for many cognitive scientists is that as we develop newer and better ways of investigating the brain and expand the techniques and materials used in information processing, we will learn more about how the brain itself performs cognitive tasks.

The cognitive sciences, although over a half century old, have really come into their own over the past twenty-five years. One major reason is the emergence of a new and evocative method of computing first explored in the forties but not developed to its potential until the mid-eighties: parallel processing using networks of neuron-like units. So-called "neural networks" are free of some of the problems that made the brain-as-information-processor metaphor unappealing when all the information processors available were serial, symbol-manipulation computers like the ones we all have on our desks.

Serial machines are great at mathematics, word processing, communications applications, and graphics. They aren’t very good at recognizing patterns or categorizing unfamiliar input. Since serial machines follow explicit rules, they usually stop working altogether when faced with something for which those rules are unprepared. Brains aren’t like that. Among their more mysterious and useful abilities are recognizing patterns and dealing with novel experiences.

Neural networks, too, are able to handle new input and recognize patterns. Instead of crashing when they run into the unexpected, they (like brains) provide a response that is "good enough", i.e., they treat the new case like the most similar familiar case, a strategy that is often effective. When they do show a loss in performance due to unexpected or incomplete input, or even due to system damage, the performance loss is proportional to the cause instead of total.

What’s more, neural networks are far more realistic models of the brain than their serial, symbol-manipulating counterparts. Serial computers use vast numbers of voltage switches to implement various instructions that are stored as binary code. Neural networks are instead composed of neuron-like units that are interconnected. Each unit has a level of "activation", either a voltage or numeric value (depending on how the network is realized; most are modeled on standard serial computers but may be electrically or electronically implemented as well) that depends in part on the activation levels of the other units in the network. Real neurons in the brain are similarly interconnected, and each one has an activation level (its firing frequency) in part as the result of the signals it receives from those other neurons to which it is connected. Instead of following stored instructions, neural networks (and, presumably, brains) are able to provide the proper responses to input based solely on the configuration of their units’ activation levels and the connections between them. There are no instructions in a neural network, and this accords with what we know about the brain as well.

Even more recently, advances in neuroscience, perhaps the most vivid of which are new means of observing the living brain, have accelerated our ability to test theories that had previously been conjectural or limited due to the nature of the organ under study. In order to determine, for example, the function of a particular area of the brain, it used to be necessary to experiment on cooperative patients undergoing neurosurgery. One would stimulate an area of the brain with an electric pulse and record the patient’s motor or sensory response. Not only is this method limited by opportunity, but it also presents some danger for the patient. Positron emission tomography (PET) and other imaging techniques now allow real-time glimpses into the activity of living brains engaged in cognitive tasks without risk to the patient. One especially valuable advantage of PET and related techniques is their suitability for use with large numbers of subjects. Although the processes under observation are not fine-grained (PET scans measure brain activity indirectly by tracking blood flow), we now have the means to make more reliable generalizations about the physical organization of the brain and the areas in use on a given task.

The development of these disciplines has fueled changes in the approaches of anthropology, biology, linguistics, philosophy and psychology, just to name a few of the most prominent areas where cognitive science is practiced. In cognitive psychology, the amount of information processing required by a person to perform a task such as mentally rotating a geometric figure in order to compare it to another is inferred from the time it takes relative to other tasks. Cognitive anthropologists study the degree to which culture affects categorization and perception of the world. Philosophers of cognitive science debate such issues as how well the information-processor metaphor fits the brain and mind and whether our commonsense descriptions of cognitive activity have a place in science.

Each of these areas affects the other. In fact, the multidisciplinary nature of cognitive science makes it essential that a researcher in one field be able to make use of the results from another. Thus a linguist must be able to tell whether her new theory of language acquisition in children is consistent with the latest findings in cognitive developmental psychology, as well as whether or not such a process of acquisition could be realized in a human brain. Likewise, the researcher in neural network modeling must not only be aware of the constraints imposed on his models by neuroscience, but must also be sensitive to the philosophical questions raised by his methodology, since given our state of general ignorance about the relation between mind and brain, one must take care just which predictions one makes based on a model that must be imperfect. What has been needed in the cognitive sciences is a single resource accessible to those working in and studying each and all of the disciplines involved.

The MIT Encyclopedia of the Cognitive Sciences, edited by Robert A. Wilson and Frank C. Keil and boasting a list of contributors that reads like a who’s who of the affiliated areas of study, seeks to provide a comprehensive and detailed survey of the subjects that will prove useful to both students and professionals. This is no mean task, given the volume and diversity of work that must be presented briefly enough to fit, clearly enough to inform rather than baffle the reader, and faithfully enough to show both areas of fundamental disagreement and those of exciting convergence. For example, three consecutive entries are on mobile robots, modal logic, and modeling neuropsychological deficits. If one is to make use of the book, each of these entries must be accessible to a psychologist, an ethologist, a biologist, and a computer scientist.

For the most part, the book succeeds as well as one could hope. One reason is the inclusion of introductory essays providing overviews of six of the most important disciplines involved: Philosophy, Psychology, Neuroscience, Computational Intelligence, Linguistics and Language, and Culture, Cognition and Evolution. Each serves as a fine introduction to the development of the relevant field and a map of its most pressing concerns and the often divided opinions on them. The essays on disciplines outside the reader’s area of expertise are indispensable, and that which covers his or her own backyard is likely to be useful as well due to its bird’s-eye perspective.

The entries are likewise clear and informative and rarely presume familiarity with technical terminology or domain-specific concepts. This is accomplished by ubiquitous cross-references to other entries in the volume. Since one is likely to be led to an unfamiliar entry due to a cross-reference from a more familiar one, it can be dizzying to discover that mastering the new concept necessitates understanding five others, each of which leads to still more. Still, such is to be expected in so complex and challenging an area as the cognitive sciences, and the rewards are easily worth the effort for the serious reader. Since no single entry can afford to be exhaustive, further readings and references are provided for those who want to explore a topic in more depth.

Two mild warnings are necessary for readers who may expect this work to contain neutral, disinterested reports of the topics covered, presented for the consumption of the layman. First, each entry is written by an expert, often the very person who has made a theory or area of study notable. For example, Tim van Gelder writes about the dynamic approach to cognition for which he has often argued, John Searle explains his own famous Chinese room thought experiment, and James McClelland authors the entry on the connectionist cognitive modeling that he pioneered. Each of these topics is considered controversial by many. Nevertheless, no bias seems to infect the collection as a whole in either the choice of entries or their mode of presentation; despite passionate disagreement within the cognitive sciences, no voice of note appears to be unheeded. The second warning is that the entries are accessible to non-experts familiar with cognitive science, but not always to those with no previous familiarity with at least one of the fields involved. This is unlikely to affect most readers, for a book of this size, at this price, is not one that will be picked up by the curious layman, and is intended for those studying and working in the cognitive sciences.

All in all, the editors have done an excellent job providing students and professionals in the cognitive sciences with a reference book that anyone who wishes to understand and even contribute to the ongoing exploration will find of great use. The publishers at MIT Press have just about cornered the market as far as books on the cognitive sciences are concerned. With The MIT Encyclopedia of the Cognitive Sciences, they add to their record of success and may save the educated non-expert hundreds of dollars by providing a single, concentrated overview of where the relevant cluster of disciplines has been, where it is at, and where it still needs to go.

Robert Litzenberger is a PhD student in Philosophy at SUNY Stony Brook, writing a dissertation on the philosophy of psychology.

Categories: General, Philosophical