This book presents a view both of computation in the brain and of how computation should be organized in future computer systems. As the book is written by someone who is well versed in several fields, including symbolic AI, artificial neural networks, neuroscience and control theory, it is well worth paying attention to. Much of the book is about matters such as vision, in both its low level and high level aspects, and movement control, but it also addresses natural language and, briefly, consciousness.
Apart from the particular theoretical view it espouses, the book is useful in its detailed summaries of some aspects of the mammalian brain, neural systems of lowly animals (notably frog, toad and marine slug), control theory, dynamical systems theory, artificial neural networks and AI. It is meant to be accessible overall to, say, a typical Scientific American reader, but is also intended to be useful for experts in fields such as AI, robotics, cognitive science, neural networks and neuroscience. The book is not, and could not be, anywhere near encyclopedic on such fields; the emphasis is rather on aspects of those fields that serve in some way to illustrate, support or constrain the central theoretical ideas of the volume. But one could repudiate the whole of Arbib's theory and still glean much useful information from the book, and indeed much of the information given seems rather independent of his theory. The book does skip about a bit from topic to topic, but as it is fat and dense that may be a good thing.
The central ideas of the volume are cooperative computation, coordinated control programs, schemas, action-oriented perception, perception-oriented action, and, stepping down to the neural level, topological mappings (in the sense that, say, a retinotopic representation in brain cortex is topological). In all this, schemas are the primary notion.