Context

Download PDF Print Page Citation Share Link

Last Updated on May 8, 2015, by eNotes Editorial. Word Count: 370

In the seventeenth century, French philosopher René Descartes argued that humans have an immaterial mind (also called “soul” or “self” or “spirit”) over and above the material brain. This position is known as mind-body dualism. The mind is the origin of thought (“the engine of reason”) as well as “the seat of the soul.” For Descartes, the mind causally interacts with the brain, although this interaction is difficult to explain. Modern dualists have not been successful either, and this problem has led many philosophers, psychologists, and cognitive scientists in general (such as neuroanatomists, neurochemists, artificial intelligence researchers, scientifically trained philosophers) to argue for materialism, the view that denies there is such an entity as “the mind” and claims that there is only one entity, the material brain. In The Engine of Reason, the Seat of the Soul, Paul M. Churchland holds this position, a position supported by studies on brain-damaged and brain-lesioned patients. For example, postmortem examinations of the brains of people who had Alzheimer’s disease reveal material plaques and tangles throughout the fine web of synaptic connections of the neurons of the brain that embodies all of one’s cognitive skills and capacities for recognition.

Illustration of PDF document

Download The Engine of Reason, the Seat of the Soul Study Guide

Subscribe Now

Moreover, Churchland is impressed with the tremendous advances in the neurosciences and in AI research that allows modeling of brain processes. Modeling enables cognitive scientists to represent brain function as massively parallel distributive processing (PDP) of recurrent neural nets that carry out vector-to-vector transformations or vector completions. This model (perhaps theory) of human cognitive brain processes will effect, Churchland claims, a revolution in understanding of the self, consciousness, all cognitive processes, science, art, and much else besides. His book is intended to convey the possibilities and excitement of this revolution.

Churchland’s book is divided into two parts. Part 1 describes the enormous anatomical complexity of the brain and links this complexity to artificial neural networks in computer modeling that imitate parts of the brain. Part 2 explores the consequences of this neuroscientific approach to cognition and soul and, in the process, delves into the nature of consciousness and shows how the new neuroscientific approach can be applied in many other domains besides the study of consciousness, such as science, philosophy, ethics, law, and medicine.

The Brain

Download PDF Print Page Citation Share Link

Last Updated on May 8, 2015, by eNotes Editorial. Word Count: 513

Churchland begins by describing the enormous structural capacity of the human brain to represent the world. For example, a standard television screen contains about 200,000 pixels, the tiny dotlike elements that are easily seen if one peers very closely at the screen. However, the human brain has approximately 100 billion nerve cells, or neurons, each of which can also take on a full range of activation levels (or “brightness values” when compared to pixels). Counting each neuron as a pixel, one can calculate that the brain’s representational capacity is about 500,000 times greater than a television screen’s representational capacity.

Churchland then asks the reader to consider one of the twin towers of the World Trade Center in New York City. Imagine the entire outside surface—all 500,000 square feet of the skyscraper—to be tiled with 500,000 television screens so that each seventeen-inch screen is glued next to each other and facing outward. Assuming there are about 200,000 pixels per square foot, one can calculate that there are 100 billion pixels in this setup that correspond to the 100 billion neurons in the human brain. This is the minimum visual representational power of the human brain, for the brain can also represent reality in many other dimensions, including, for example, social, moral, and emotional.

Churchland then asks the reader to picture that the skyscraper’s pixels are embedded in a thin sheet of aluminum foil that covers the entire building. He then imagines someone scrunching this great expanse of foil into a ball about the size of a large grapefruit, which is approximately the volume of the human brain. These 100 billion pixels (analogous to neurons) can still represent the world even if they are folded out of sight.

Furthermore, Churchland considers the manner in which the brain represents the world. He argues that AI research gives the best answer in terms of artificial neural nets that involve parallel distributive processing. The brain is treated as an advanced neurocomputer. Neural nets represent the world by means of vectors, which are simply lists of numbers or sets of magnitudes. (Note that this use of the word “vector” is different from its use in physics, where “vector” refers to a magnitude, such as velocity, that also has a direction.) For example, on the tongue, there are four distinct kinds of receptor cells. The vector coding scheme for taste, such as the taste of a peach, might have a substantial effect on one of the four types of cells, a minimal effect on a second type, and an intermediate effect on the third and fourth types. Taken together, this precise pattern of relative stimulations constitutes a type of neural “fingerprint” that uniquely characterizes the taste of peaches. Or more exactly, a specific taste is a pattern of spiking frequencies across the four neural channels that convey information of these activity levels away from the mouth and to the rest of the brain. Basically, the same system, mutatis mutandis, can be used to represent sensory coding for color, smell, and the more complex actions of facial recognition and use of the human motor system.

Cognition

Download PDF Print Page Citation Share Link

Last Updated on May 8, 2015, by eNotes Editorial. Word Count: 673

Churchland turns to a discussion of cognition, which involves learning or “training up a network.” One famous example concerns a submarine’s sonar system detecting the difference between a mine echo or a rock echo. The sonar system network contains thirteen sonar input cells, and each cell codes the total energy contained in the sonar echo at exactly one of the thirteen sampled frequencies. Each echo is thus characterized by a distinct activation vector across the input population of cells. The cells of the sonar machine at the input layer all project to a second layer and then on to a third layer that consists of only two cells, whose job it is to signal mine or rock, as the case may be. This is also called “vector-to-vector transformation.” If the output vector relaxes into a final configuration of (1,0), or close to this, then the sonar has detected a potentially dangerous, explosive mine; when the final configuration is (0,1), or close to this, the sonar has detected a harmless rock.

As the network is trained up, scientists initially have no idea how to configure its connection weights. Hoping to get lucky, they set the weights at small random values and prepare to teach the network on a substantial training set of recorded sonar echoes, half of them returned from real mines placed by the scientists on the ocean floor and half of them returned from visually identified rocks. Using a well-known method of back propagation of synaptic weight adjustments, the scientists cycle the information repeatedly through the training set until the network has assumed an overall synaptic weight adjustment that minimizes the mean squared error at the output layer. That is to say, the scientists continue to instruct the network until it has learned to make the mine-rock distinction as reliably as it can.

This network described is quite real. Churchland cites the work of Paul Gorman and Terry Sejnowski, who have created such a network that topped out at a performance level of 100 percent on the training set. When tested on echoes from outside the training set, the network generalized to these new samples very well, identifying better than 90 percent of them correctly. The training up of a network regularly produces a partitioning of its higher level activation spaces into a hierarchial structure of categories and subcategories producing a framework of concepts that subserve the skill acquired. For example, the sonar network displays two categories with prototypical cores. The sonar example is a model of how the human brain learns, and the research program is extended to neuroscience as to which parts of the brain have sufficient neuroanatomical and neurophysiological complexity to allow for such vector-to-vector transformations.

Churchland extends the discussion from simple forms of sensory coding, to vector coding at the scale of many thousands or even millions of neurons, to the emergence of categories and their central prototypes as carefully crafted areas of activation space, and to recurrent neural processing at the level of animal locomotion and visual interpretation in humans. The central phenomenon at the level of recurrent networks is the phenomenon of vector completion of partial inputs often aided by the brain’s recurrent manipulation of the relevant population of representing neurons. That is, vector completion can represent the phenomenon of someone’s recognizing—perhaps slowly at first, but then suddenly—some unfamiliar, puzzling, problematic situation as being an instance of something (a prototype) well known to that person. Churchland traces out this idea in terms of many salient discoveries in science, such as Descartes’s “whirlpool” interpretation of the various planetary motions; Isaac Newton’s “deflecting force” interpretation of the Moon’s elliptical orbit around Earth; and Albert Einstein’s “straight-line in non-Euclidean space-time” interpretation of the Moon’s elliptical orbit around Earth. Thus, Churchland claims, some of the most sophisticated intellectual achievements involve the same activities as vector processing, recurrent manipulation, prototype activation, and prototype evaluation, which can be found in some of the simplest of cognitive activities, such as recognizing a dog in a low-grade photograph.

Consciousness

Download PDF Print Page Citation Share Link

Last Updated on May 8, 2015, by eNotes Editorial. Word Count: 503

In part 2, Churchland deals with the fascinating but difficult puzzle of consciousness. This phenomenon appears unique to human beings and beyond scientific and purely physical explanation. Traditionally, philosophers have argued that the phenomenon is basically a subjective occurrence, accessible only to the creature that has it. Churchland argues against this classical view.

Churchland begins by reviewing a number of similar arguments for the classical view advanced by philosophers such as Gottfried Wilhelm Leibniz, Thomas Nagel, John Searle, and Frank Jackson. Nagel’s argument was advanced in his seminal 1974 paper, “What Is It Like to Be a Bat?” Nagel argues that no matter how much one might know about the neuroanatomy of a bat’s brain and the neurophysiology of a bat’s sensory apparatus, one will never know “what it would be like” to have the bat’s sensory experience. Even if scientists could track the neuroactivation patterns, one would never know what they are like from the unique perspective of the creature that possesses them; their intrinsic character as felt experiences would still be unknown to us. A purely physical science of the brain, Nagel and others argue, does have a limit on the capacity of understanding as it reaches the subjective character of the contents of one’s consciousness.

Churchland responds by arguing that Nagel fails to make a distinction between how one knows something and the thing known. Churchland argues that the existence of a unique first-person epistemological access to a conscious phenomenon does not entail that the phenomenon is nonphysical in character. For example, the difference between X’s knowledge of her facial blush and Y’s knowledge of X’s facial bluish lies not in the thing known but rather in the manner of knowing it. The blush itself is a physical entity.

Churchland then proposes seven provisional criteria of adequacy that a neuroscientific theory of consciousness must try to reconstruct. Consciousness involves short-term memory, is independent of sensory inputs, displays steerable attention, has the capacity for alternative interpretations of complex or ambiguous facts, disappears in deep sleep, reappears in dreaming, and holds the contents of several basic sensory modalities within a single, unified experience. Churchland takes up each of these criteria of consciousness and shows how they can be explained by treating the brain as a recurrent neural network. Churchland cites some areas where such neuroprocessing could occur, such as Roldofo Llinas’s view that the contents of consciousness lie within the layers of the interactively connected primary sensory cortex rather than, as Churchland himself holds, the sparser pathways of the grand recurrent loop that connects all the pathways from cortex to intralaminar nucleus.

Churchland’s position that all cognitive processes and the phenomenon of consciousness can be reduced to brain processes representable as a testable theory of recurrent neural nets is a powerful and carefully argued position. He is at pains to warn the reader several times that he may be wrong. Arguments have been raised against his position, but no definitive answer has been reached.

Additional Reading

Download PDF Print Page Citation Share Link

Last Updated on May 8, 2015, by eNotes Editorial. Word Count: 530

America. CLXXIII, December 16, 1995, p. 24.

Bechtel, William. “What Should a Connectionist Philosophy of Science Look Like?” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. The author is also a major figure in the field of philosophy and cognitive science. He holds that traditional philosophy of science has been little concerned with the psychology of science, and that the general sentential approach of traditional philosophy of science is ineffective. Bechtel nevertheless questions Paul M. Churchland’s claim that theories and explanations are best understood in terms of representations in the heads of scientists.

Choice. XXXIII, December, 1995, p. 634.

Clark, Andy. “Dealing in Futures: Folk Psychology and the Role of Representations in Cognitive Science.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. This article is written by another major leader in the field of philosophy and cognitive science. Clark agrees with the Churchlands that connectionist models can contribute very valuable, new resources for understanding human cognition. However, he forcefully rejects Churchland’s claim that this outcome impugns folk psychology.

Flanagan, Owen. “The Moral Network.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. Flanagan defends Churchland’s moral network theory. He examines the potential of this theory for making sense of moral learning, knowledge, practices, and standards. He applauds the way the theory illuminates the biological, psychological, and social forces shaping the moral lives of human beings, while the normative component of the theory can assess right and wrong, good and bad.

Fodor, Jerry, and Ernie Lepore. “Paul Churchland and Stated Space Semantics” and “Reply to Churchland.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. In both of these articles, the authors argue that Churchland’s state space version of a network account of semantics is wanting. They argue that Churchland has remained committed to two positions that are incompatible.

Library Journal. CXX, July, 1995, p. 115.

Los Angeles Times. June 30, 1995, p. E4.

Lycan, William G. “Paul Churchland’s PDP Approach to Explanation.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. Lycan discusses Paul Churchland’s prototype activation account of explanatory understanding. Lycan argues that if explanation concerns either quasi-logical relations between sentences or the natural relations between the affairs those sentences represent, then Churchland’s objections to the traditional deductive-nomological model of scientific explanation are ineffective.

McCauley, Robert N. “Explanatory Pluralism and the Co-evolution of Theories in Science.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. The author is a leader in the field of the philosophy of cognitive science. He discusses Churchland’s account of intertheoretic relations in science and the implications for a scientific psychology as opposed to folk psychology. He argues that the earlier continuum model advanced by Churchland and his wife, Patricia Churchland, is vastly oversimplified, leading to unwarranted expectations about the elimination of psychology in favor of advanced neuroscience.

Nature. CCCLXXVIII, November 30, 1995, p. 455.

New Scientist. CXLVIII, October 28, 1995, p. 44.

The New York Times Book Review. C, July 9, 1995, p. 1.

The Times Literary Supplement. August 25, 1995, p. 5.

Bibliography

Download PDF Print Page Citation Share Link

Last Updated on May 8, 2015, by eNotes Editorial. Word Count: 532

Additional Reading

America. CLXXIII, December 16, 1995, p. 24.

Bechtel, William. “What Should a Connectionist Philosophy of Science Look Like?” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. The author is also a major figure in the field of philosophy and cognitive science. He holds that traditional philosophy of science has been little concerned with the psychology of science, and that the general sentential approach of traditional philosophy of science is ineffective. Bechtel nevertheless questions Paul M. Churchland’s claim that theories and explanations are best understood in terms of representations in the heads of scientists.

Choice. XXXIII, December, 1995, p. 634.

Clark, Andy. “Dealing in Futures: Folk Psychology and the Role of Representations in Cognitive Science.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. This article is written by another major leader in the field of philosophy and cognitive science. Clark agrees with the Churchlands that connectionist models can contribute very valuable, new resources for understanding human cognition. However, he forcefully rejects Churchland’s claim that this outcome impugns folk psychology.

Flanagan, Owen. “The Moral Network.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. Flanagan defends Churchland’s moral network theory. He examines the potential of this theory for making sense of moral learning, knowledge, practices, and standards. He applauds the way the theory illuminates the biological, psychological, and social forces shaping the moral lives of human beings, while the normative component of the theory can assess right and wrong, good and bad.

Fodor, Jerry, and Ernie Lepore. “Paul Churchland and Stated Space Semantics” and “Reply to Churchland.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. In both of these articles, the authors argue that Churchland’s state space version of a network account of semantics is wanting. They argue that Churchland has remained committed to two positions that are incompatible.

Library Journal. CXX, July, 1995, p. 115.

Los Angeles Times. June 30, 1995, p. E4.

Lycan, William G. “Paul Churchland’s PDP Approach to Explanation.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. Lycan discusses Paul Churchland’s prototype activation account of explanatory understanding. Lycan argues that if explanation concerns either quasi-logical relations between sentences or the natural relations between the affairs those sentences represent, then Churchland’s objections to the traditional deductive-nomological model of scientific explanation are ineffective.

McCauley, Robert N. “Explanatory Pluralism and the Co-evolution of Theories in Science.” In The Churchlands and Their Critics, edited by Robert N. McCauley. Oxford, England: Blackwell Scientific Publications, 1996. The author is a leader in the field of the philosophy of cognitive science. He discusses Churchland’s account of intertheoretic relations in science and the implications for a scientific psychology as opposed to folk psychology. He argues that the earlier continuum model advanced by Churchland and his wife, Patricia Churchland, is vastly oversimplified, leading to unwarranted expectations about the elimination of psychology in favor of advanced neuroscience.

Nature. CCCLXXVIII, November 30, 1995, p. 455.

New Scientist. CXLVIII, October 28, 1995, p. 44.

The New York Times Book Review. C, July 9, 1995, p. 1.

The Times Literary Supplement. August 25, 1995, p. 5.

Unlock This Study Guide Now

Start your 48-hour free trial and unlock all the summaries, Q&A, and analyses you need to get better grades now.

  • 30,000+ book summaries
  • 20% study tools discount
  • Ad-free content
  • PDF downloads
  • 300,000+ answers
  • 5-star customer support
Start your 48-hour free trial