A Physiological Hypothesis of Perception

A PHYSIOLOGICAL HYPOTHESIS OF PERCEPTION WALTER]. FREEMAN* The brain is a particular kind of physical object that has the property of making within itself representations of the outside world. These representations exist in transient bursts of energy that constitute the objective aspect of mental images. The representations are shaped by sensory input; following on sequential transformations they serve to shape motor activity and to predict future sensory input. The task of the physiologist is to describe the material substrates of these representations, their physical locations and patterns, and the operations by which they are constructed and transformed in the brain. In short, what are the physical forms of mental images, and how does the brain make them? This essay reviews the conceptual bases of the task, and outlines an experimental analysis of the operations that underlie perception in the simplest mammalian sensory system, that for olfaction. It is shown that the operation of constructing a perceptual representation from sensory input cannot be described by the classical stimulus-response paradigm. The operation consists of a state change in the brain that leads to selfordering of neural activity. It can be modeled with mathematics used to describe self-organization in nonequilibrium systems and not with deterministic equations. Representations Ofand By the Brain To paraphrase a well-known political aphorism, the brain is much too important to be left to physiologists. There are numerous psychiatrists, cognitive psychologists, behaviorists, linguists, philosophers, anThe author is pleased to acknowledge the collaboration of James Boudreau, Maria Biedenbach, Jayana Emery, John Horowitz,Joe Willey, Soo-Myung Ahn, Steven Bressler, Diane Martinez, Dan Sunday, Walter Schneider, Charmane Eastman, Marty Rosenthal, and especially Brian Burke for computer system development and programming. Supported by grant MH06686 from the National Institute of Mental Health. ?Department of Physiology-Anatomy, University of California, Berkeley, California 94720.© 1981 by The University of Chicago. 0031-5982/81/2404-0245$01.00 Perspectives in Biology and Medicine · Summer 1981 \ 561 thropologists, and proponents ofartificial intelligence who describe what the brain does; and there are chemists, biophysicists, zoologists, em· bryologists, and neuroanatomists who describe how it develops and what it is made of. The task of physiologists is to deduce how it works. Indeed our field of action is rather severely limited. We can choose among competing formalisms that describe the brain functions to be analyzed, and we can select among various types of brain and among various physical and chemical variables for observation and measurement, but there is not much room for the creative play ofimagination. Perhaps this is why physiologists often write poetry in their spare time. There appears to be consensus among those who analyze what the brain does that one of its main functions is to make representations of the outside world. This idea in various forms can be traced back to the school of Hippocrates. In recent times it appears in the forms of the "parallel representations" of Craik [1], the "command cell" of invertebrate neurobiologists [2], the "statistical ensemble" ofJohn [3, 4], the "search image" of ethologists [5, 6], and several others (e.g., [7-15]). There are also internal images postulated in cybernetic theories of Wiener and his associates [16], "re-afferent copy" proposed by von Hoist [17], "corollary discharge" ofSperry [18], and the "cerebral modelling of the future" of Bernstein [19]. The diversity of forms and their experimental bases reflect the recurrence ofthe need for the concept ofcentral nervous system representations and the breadth of the range ofexperimental data that support it. These inferences pose for the physiologist the questions, "What physical properties do representations have?" and "How can we explain the mechanisms of their formation?" Proposed answers to the first question fall into one of two categories. In one set of views, "individual neurons can encode complex information and concepts into simple electrical signals; the meaning behind these signals is derived from the specific interconnections of neurons" [20, p. 3]. Examples are "feature detectors" [9], "cardinal cells" [10], and "command cells" [2, 20]. This category is based on the enthusiastic but not always critical extension to the central nervous system of the neural code that was firmly established for the peripheral nervous system during the century from Müller...