Book Review
Apprentices of Wonder: Inside the Neural Network Revolution
by William F. Allman
Reviewed by Barry Kort
Apprentices of Wonder is written for the educated lay audience. It is not a textbook on neural networks from a practitioner in the field. Rather it is the engaging story of people and their projects, written in the style of Pamela McCorduck (Machines Who Think), Tracy Kidder (Soul of a New Machine), Grant Fjermedal (The Tomorrow Makers) and James Gleick (Chaos: Making a New Science). William F. Allman is Senior Editor at U.S. News and World Report and an award-winning science writer. The book jacket captures the essential character of Allman's engaging style:
Apprentices of Wonder
takes you into the exciting new science of the twenty-first century: the study of a revolutionary kind of artificial intelligence, called neural networks, that mimics the complex activity of the human brain. Here is the story of a scientific revolution–the drama of the men and women whose work in brain/mind science rivals in scope and impact the emergence of relativity theory and quantum mechanics in physics.Science writer William F. Allman take us behind the scenes and into the labs of the world-renowned pioneers of the neural network revolution, showing them in action as they reinvent the mind, from its electronic rumblings to its metaphysical rantings. We watch neuroscientist Gary Lynch map the neural circuits for smell in rats; trace the steps by which Jay McClelland, David Rumelhart, Geoffrey Hinton, and others created thinking machines that learn from experience; witness Carver Mead test the first silicon retina and listen as Terry Sejnowski's machine, NETalk, learns to read aloud. Drawing from a wide range of research and researchers–from whiz kids to Nobel laureates–this bold new band of psychologists, physicists, computer scientists, and philosophers is exploding traditional thinking about the mind and forging a new understanding of the underlying connection between mind and machine that may one day solve the mystery that has challenged man since his first self-conscious thought: What is consciousness and where does it come from?
Using puzzles, problems, and brainteasers, Allman illustrates the power and limitations of our brain's remarkable cognitive abilities and explains with startling clarity the highly complex technologies–and wondrous architecture–of the mind and the dazzling mind-bending ideas of some of its most brilliant apprentices. With the verve and human drama of Chaos and The Soul of a New Machine, Apprentices of Wonder takes us on a fascinating journey to the farthest frontiers of the mind and the revolution now taking place in our thinking–about thinking.
The hyperbole of the book jacket notwithstanding, I did indeed find Apprentices of Wonder to be an appealing, informative, entertaining and readable excursion into neural networks. The story reaches back to the analysis of perceptrons by Minsky and Papert, and their finding that a single-stage network could not implement the exclusive or (XOR) function or other nonlinear transformations. Their negative finding discouraged other researchers who were then experimenting with multi-layer neural networks. The resurgence came primarily from the work of John Hopfield and David Tank whose positive results and penetrating analysis stimulated fresh interest.
Allman engages the reader with a sprinkling of instructive brain teasers, designed to illustrate the diversity of cognitive skills from language parsing to syllogistic logic to spatial reasoning to combinatorial logic to reasoning by analogy. He keeps the reader alert with self-referential sentences like "Wee con undrestin wrds efen wen theh ar missspld. Or fill in the blanks when l t rs are missing."
Of all the neural network research, Allman's favorite appears to be Terry Sejnowski's NETalk, that modest 309-neuron network which learned to read aloud elementary school text in a day's worth of training. Sejnowski's NETalk may not be as profound as hippocampal research (mentioned briefly) or the reverse engineering of the vestibular occular reflex (barely hinted at), but it does make a good story. Had I not been attracted by the gee whiz machines, I might have missed out on the neuroscientist's exploration of the wetware.
Lest I mislead the reader with my personal bias, here is the Table of Contents of Apprentices of Wonder:
INTRODUCTION - FROM NEURON TO PSYCHE CHAPTER ONE - MAKING UP A MIND
The New Connectionist Revolution
The Symbolic Mind
Rethinking How We Think
CHAPTER TWO - THE PUZZLE MACHINE
The Mechanics of Thought
Putting the Brain to Work
Bringing the Brain Back In
Our Irrational Mind
The Insight Machine
CHAPTER THREE - THE GREAT DIVIDE
Brain Science versus Mind Science
The World Within
A New Engine of Thought
The Brain-Mind Split
A Mind in a Machine
The Changing of the Guard
A New Concept of Mind
CHAPTER FOUR - WETWARE
The Anatomy of Memory
Probing the Mind's Machinery
Wiring the Brain
Making Memories (or What Memories are Made Of)
A Brain Circuit for Smell
The Evolution of Mind
CHAPTER FIVE - WHAT MAKES A BUNCH OF NEURONS SO SMART?
The Science of Complexity
The Mathematics of Mind
A Wrench in the Gears of the Universe
Miniature Universes or Miniature Worlds
Finding the Best Solution Among Many
The Creative Computer
CHAPTER SIX - THE SELF-TAUGHT COMPUTER
Neural Networks Learn to Learn
The Neural Net Revolutionaries
The Interactive Mind
Neural Nets' Troubled Past
A Simple Neural Net
The Keepers of the Flame
The Building of the Neural Net Revolution
The Power of Multiple Layers
Making Hidden Connections
CHAPTER SEVEN - MACHINE DREAMS
Putting Neural Nets to Work
The Neural Net Industry
The Matchmaker
Neural Nets in Space
Computing with Light
Engineering the Brain
Computers that Listen
The New Engineering
CHAPTER EIGHT - WARRIORS OF THOUGHT
Winning the Hearts of Mind Researchers
The Empire Strikes Back
The Language of Thought
Theoretical Language versus Human Language
The West Coast View
The Iconoclast Linguist
Clues from Color Perception
Fundamental Categories
Language and Connectionism
The Culture Connection
CHAPTER NINE - THE PRIZE PUPIL
NETalk Learns to Read Aloud
Studying the Brain from Both Sides
Inside the Neural Net
Hidden Representations
The Ghost in the Machine
EPILOGUE - THE APPRENTICES OF WONDER
I heartily recommend this book, especially for those neural network specialists who despair of explaining their work to nontechnical family and friends. I know of no other source which makes the field of neural networks so readily accessible to the public readership.
Barry Kort
This book review originally appeared on UseNet and was reprinted in Artificial Intelligence, Volume 53, 1992.