Andrew L. Coward has a lofty goal, which is nothing less than the "Holy Grail" of cognitive science, namely, to generate a comprehensive and detailed account of human cognition and behavior that directly emerges from our scientific knowledge of the physiological structure and activity of the brain.  Consequently, in "A system architecture approach to the brain: From Neurons to Consciousness", Coward painstakingly describes his vision of a cognitive system that is the expression of its physiological and neuroanatomical substrate (i.e., the human brain or at least what we currently know of the human brain) and that can emulate human performance.  His proposal, labeled "Heuristically designing and managing a network", is the logical endpoint of the author's methodical exploration of a vast collection of scientific knowledge, from engineering to neuroscience, an exploration that spans 22 years. The use of the present progressive tense in the label suggests a work in progress but the author claims a more definitive status for his work,  as illustrated by a US patent, an electronic implementation, and the current book.

To be clear, the cognitive science literature is literally brimming with proposals that attempt to account for specific cognitive/behavioral phenomena by referring to the physiological processes of the human brain and its neuroanatomical substrates.  Yet, the author claims that such proposals (e.g., artificial neural networks) are unsatisfactory. His claim originates from the observation that existing accounts are generally confined to particular processes, and, perhaps as a result, these accounts cannot adequately explain how the human brain can learn to perform a large array of differentially complex operations with a limited pool of physiological resources and a largely stable architecture.

Coward reminds the reader that the breadth of his proposal is not the only attribute that makes it superior to existing accounts.  His proposal, he claims, is different in several other ways. For instance, it is informed by the author's understanding of the technological challenges involved in designing and manufacturing electronic systems, an understanding that he has acquired from working for Nortel Networks. Obviously, Coward acknowledges that the brain and electronic systems are not isomorphic universes. Yet, he also recognizes that if both are treated as complex entities that perform a large array of more or less interrelated operations, some of the constraints that play a role in the construction and functioning of electronic systems can be relied upon to understand the functioning of the brain as a cognitive system. Such constraints (collectively called by the author "recommendation architecture") concern the boundaries within which hardware configuration and functional properties (e.g., processing resources and capacities) can be envisioned to produce a system that functions with the intent of optimally mimicking human performance. 

The book contains interesting ideas grounded in a remarkably extensive literature review. These ideas are clustered around Coward's claim that his model can satisfactorily explain human cognition, including higher-level operations, in terms of the physiology of the brain.  They are organized into two areas: one devoted to architectural concerns and the other to issues of information processing.  Overall, the envisioned model of the human brain/mind is comprised of modules and related sub-components, algorithms that are either unique to some devices or broad-spectrum, and functional constraints based on the ratio of features to resources. The component parts are expected to replicate anatomically and functionally separate areas of the cortex and of older portions of the brain such as the midbrain and forebrain, and imitate the topology and synaptic algorithms of neurons. Most of the functional features that define the modular architecture of the model exploit well-know concepts and issues in cognitive science such as learning by feedback, the principle of similarity in perception and memory, hierarchy and modularity of operations, processing biases etc. Their novelty resides in the way they are all fitted into the development of an all-encompassing model that is intended to be a replica of the most critical functions of the human mind/brain.

The key question that comes to mind after reading this book is whether Coward's model can indeed satisfactorily explain human cognition, including higher-level operations, in terms of the physiology of the brain. Very modestly, I would like to venture the opinion that considerable research must be conducted into the human mind and its physiological and neural substrate before an affirmative answer to this question ceases to appear premature. Similarly, considerable empirical testing would need to be devoted to Coward's theoretical and practical realizations of his vision of a cognitive system resulting from our current knowledge of the human brain. Consequently, more work awaits cognitive scientists before the present progressive tense in the label "Heuristically designing and managing a network" can be replaced with a past tense.  The first step in this endeavor is to make the system devised by Coward more transparent and amenable to empirical testing. The guiding principle of the advocated testing is not to be limited to "output isomorphism". Namely, if a realization of a cognitive system appears to perform as the actual living system, the conclusion cannot be that the realization is entirely representative of that system's operations and their related physical substrates. Nevertheless, simulation data should be published in their entirety and cognitive scientists should be allowed to examine them before the system is described as having fulfilled its goal of replicating critical functions of the human mind/brain. 

In summary, this is a book for readers who want to be challenged with a model that assembles familiar behavioral, cognitive, physiological, and biological knowledge into a novel proposal. Thus, it is for individuals with a solid background in cognitive science, including basic knowledge of cognitive psychology, neuroscience, and artificial intelligence.  It is to be noted that Coward's exposition is laborious, often marked by repetition of the main points and at times deprived of examples that could make such points transparent at the first encounter. Undoubtedly, this is a read for individuals who can adapt to a technical language whose lexicon and conceptual assemblages may sound foreign but whose purpose is to communicate a proposal that deserves to be put to the test.

© 2007 Maura Pilotti

Maura Pilotti, Ph.D., Department of Psychology, Dowling College, New York. 

Categories: Psychology