CHAOS IN THE CNS: THEORY AND PRACTICE

Walter J. Freeman

Unpublished manuscript prepared for the Dahlem Workshop on Flexibility and Constraint in Behavioral Systems. All rights reserved by the author. No citation, abstracting or any other usage is permitted.

Abstract

  • Deterministic chaos is characterized by complexity that is self-organized according to internal constraints. Examples occur at all levels of organization of nervous systems, from single neurons through neural networks up to whole brains. The search for the constraints is enabled by use of the theory of chaos to model neural functions. The level of greatest interest for modeling by cognitivists is that of the limbic system, because in animals from fish to man that is where the organization of voluntary behavior occurs. Chaotic dynamics can account for its flexibility and creativity. Our observations of the electrical activity of the brain during intentional behavior give us a window onto essential parts of the intentionality of the subject. The term is used here in one of two currently debated meanings, deliberately in order to advance the claim that biological measurements of brain function will help to resolve some centuries-old philosophical disputes about the nature of the fundamental processes of the mind.

    •

    The Limbic System

  • It is generally agreed that voluntary behavior (a term attributed to Sir Thomas Willis in 1664 - Clarke and O'Malley, 1968 - that he used to distinguish activity of the soul from that of the machine) is a product of the activity of neuronal networks in the brain; that it is an emergent, internally created activity as distinct from stimulus-initiated reflex behavior; that it is flexible and adaptive owing to the modifiability of synapses between the neurons; that it requires the genesis and prior evaluation of approaches to intermediate goals; and that the main anatomical locus for its production in vertebrates is the limbic system, on the basis that surgical damage or disease in its parts, including the olfactory system and hippocampus, severely alters capacities of animals for consciousness, memory, orientation, and emotional behavior. In higher primates these functions are augmented and regulated by the frontal lobes and thalamus operating through the limbic system, but my conception of voluntary behavior includes the behavior of rats, rabbits and cats, in which the frontal lobes play a modest role.

    •

    Memory

  • Some pivotal questions are: How does the organized neural activity that initiates and sustains voluntary behavior arise in the limbic system? What observable forms does it take? How does it depend on past experience? How does it achieve flexibility in adapting to unpredictably changing environments? An optimal starting point for answers is study of the neural mechanisms for memory, of its role in coding information, and of its role in classical and operant conditioning, which voluntary behavior incorporates and transcends. A principal focus of research has been the ganglia of invertebrates, in the hope that an understanding of the biological basis of memory through the anatomy and physiology of single neurons in simpler nervous systems might lead to the discovery of universal mechanisms of memory in the modifiable synapses (Alkon, 1992).
  • This approach is profoundly important for outsiders to see in order to understand the viewpoint of neuroscientists. The most basic tenet of their beliefs is the Neuron Doctrine, which holds that all transactions in brains are to be understood in terms of these basic units and the movements of chemicals across their enclosing membranes under the control of the genes and the molecular machinery in the cytoplasm. All of physics, chemistry, molecular biology and behavioral pharmacology can be focused like a brilliant spot of sunlight on a single synapse in the ganglion of a snail, and the refraction can illuminate the entire nervous system, like the grain of sand that contains the universe. Investigators seek the legendary "memory trace", the permanently modified membrane channel, or its protein molecule and regulatory gene that form the channel gate to regulate the flow of chemicals across membranes. These ultimate structures are conceived as the memory "code written in electrical and molecular alphabets [that] might some day help us to read our own minds." (Alkon 1992, p. 126) The search continues with increasing frenzy, despite the conclusions arrived at years ago by Sir Frederic Bartlett, who wrote: "...some widely held views have to be completely discarded, and none more completely than that which treats recall as the re-excitement in some way of fixed and changeless 'traces' " (1932, p. vi)

    • The Hippocampus

  • A snail is interesting mainly to the extent that it leads to understanding human brains. The best pathway upward may lie through the limbic system of rats and rabbits, particularly its centerpiece, the oldest cortex, the hippocampus. When cut from the intact brain and kept alive in a glass chamber as a "slice", it is the natural stepping stone from the isolated ganglion with its few neurons to mammalian neocortex with its many million neurons in each cubic centimeter. It provides the best model of synaptic learning yet devised in the use-dependent changes of long term potentiation (LTP) and depression (LTD), which have yielded important insights into synaptic plasticity, as well as experimental verification of Hebb's (1949) rule for synaptic change in the formation of nerve cell assemblies. Its immense numbers of neurons and complex patterns of connection support the development of a variety of neural network models for parallel distributed neurocomputation (Schwartz, 1990; Alkon, 1992), which might lead to more realistic models for brain function in voluntary behavior.
  • In regard to applications with neuropsychiatric patients, Alkon concludes (p. 225): "The awesome tasks of treatment must include restructuring the memory banks, reorienting the mind itself." Herein one can begin to see why this enterprise may fail, if the synapse is conceived as an address in a computer memory, which can be re-written as part of a new ''program". We know that memory is fallible, and that the human psyche can generate images, ideas, and articulations that transcend the limits of computers and neuropsychiatry (Sacks, 1985). A noncomputational approach is required (Globus, 1992).

    • Information Processing

  • The two major classes of theory that neuropsychiatrists have used for the past century to explain the biological bases of human behavior are energy-based and information-based. In the 19th Century in accord with Newtonian physics the idea of 'nerve force' replaced the preceding "animal spirits" (Clarke and O'Malley, 1968) and was widely used to explain brain function. Spencer (1863) accepted as "...an unquestionable truth that, at any moment, the existing quantity of liberated nerve-force, which in an inscrutable way produces in us the state we call feeling, must expend itself in some direction, must generate an equivalent manifestation of force somewhere" (p 109), thus appealing to the conservation of momentum. Hughlings Jackson, the founder of modern neurology whose concepts of evolution, dissolution, and hierarchical control in the nervous system are still operant, continued : "...we speak of the dynamics of the nervous system.... A normal discharge starting in some elements of the highest centres overcomes the resistance of some of the middle, next the resistance of some of the lowest centers, and the muscles are moved." (1882, from Vol. II, pp42-44, 1958). Significantly he added: "Resistances will be considered later," but in a marginal note by hand wrote: "No more of this was published." 'Nerve force' as he used it proved not to be a force or an energy subject to the laws of thermodynamics, and therefore 'resistances' could not be defined or measured. There were no physical or physiological meanings for these terms applied to brain function and behavior.
  • Yet they continued to be used literally (as in the current use of brain imaging techniques for cortical localization) and metaphorically, as in Freud's "The Project of a Scientific Psychology" (1895): "What I have in mind is the principle of neuronic inertia, which ... finds expression in the hypothesis of a current, passing from the cell-processes or dendrites to the axon. ... The secondary function [memory] is made possible by supposing that there are resistances which oppose discharge ... in the contacts [between the neurones] which thus function as barriers. The hypothesis of "contact-barriers" is fruitful in many directions." (from pp 356-359, 1954). This was published two years before Foster and Sherrington introduced the term 'synapse'. The development of neurophysiology was assured, and it was accelerated with the discovery forty years later by Otto Loewi of neurochemical transmission, but the category confusion between 'nerve force' and neural current was embedded in the foundations of neuroscience.
  • By the mid-20th century a new class of theory emerged with the computer metaphor, in which 'nerve force' was replaced by 'neural information'. It was initiated by Craik (1943), but the most influential step was taken by Shannon and Weaver (Lucky, 1989), who introduced information theory by divorcing 'information' from 'meaning' (Dreyfus, 1979). His theory is appropriate for telephone engineering, where the terms 'bits', 'channel capacity', and 'codes' can be clearly defined, but its use in biology has been as loose and metaphorical as that of its predecessors, 'animal spirits' and 'nerve force'. No one has measured the 'channel capacity" of the optic nerve, or the number of bits per second it carries when signaling a loving mother or an abusive father. The applications in artificial intelligence have been equally questionable. AI theorists seek to attach 'meanings' to their rule-driven symbols in the same way that some physiologists seek to attach bits of memory or mind or soul to membrane channels, protein molecules, or synaptic vesicles. This is the trap into which those will fall, who are seduced by the Cartesian Subject/Object distinction, and who will then search for the operational "images" and "representations" of Objects in Subjective Minds.
  • Let there be no misunderstanding. When a person or an animal learns something, neurons do grow or shrink, and changes do occur in their chemistries. One of the better known examples was the discovery in the 1960's by Holger Hyden that RNA in neurons is increased during learning. Of course, neurons need more protein to get bigger, and RNA is required to produce it, but the idea arose that memories of learned tasks might be stored in strands of RNA in the same way that DNA carries genetic information, reflecting a conflation of hierarchical levels that is common in this field. Another example is a small protein that was discovered to increase in hippocampal neurons during learning. The protein is an enzyme in mitochondria, which provide the metabolic energy needed to drive the neurons. Presumably the educated neurons need more energy to perform their tasks, and one might say that the mitochondria store the 'memory trace' of that need, but this is ludicrous. The problems of understanding the neural bases of voluntary behavior seem not so much biological as conceptual and philosophical.

    • Neurodynamics

  • My own views have been changed by new data that are inconsistent with the Neuron Doctrine. My students and I asked the questions, what is the spatiotemporal pattern of electrical activity that is induced in the olfactory system, when an odor conditioned stimulus is given to an animal? Which neurons are reliably and exclusively activated, with what time course, and in what locations? By the hypothesis of Adrian (1950) that was supported by use of metabolic labeling (Lancet et al., 1982) we predicted that clumps of cells in the bulb would be selectively activated during inhalation, comparably to the spatial selectivity of cortical responses in other sensory modalities.
  • We first examined the activity of single neurons by their action potentials. The results were unsatisfactory (Freeman, 1975), because the variability of response patterns was as great between repeated measurements with the same odor as it was between the measurements with different odors. We concluded that the sample size was inadequate by a factor of 1,000 to 10,000. We turned to the recording of local dendritic potentials, the electroencephalogram (EEG - Basar, 1980) of the bulb, with 64 electrodes forming a window onto the bulbar surface. We estimated that our array accessed about 20% of the bulb in the rabbit. In this way we employed spatial ensemble averaging over neural populations on single trials (Freeman, 1987).
  • The results showed as predicted that the information serving to classify odors was in the spatial patterns of activity averaged over each inhalation. However, each rabbit had its own unique spatial pattern with the familiar scent of the background, and as it learned to discriminate each new odor, its spatial pattern changed to a new form. The information that enabled us to classify these neuroelectric patterns was truly distributed, in that both high and low levels of activity were important, comparable to light and dark, and no part of the bulb was more or less important than any other. Most strikingly, with each change in training, such as addition of a new odor to be discriminated or a new unconditioned stimulus, all of the identified spatial patterns were changed for these and for other odors (Freeman and Grajski, 1987). Given that the central patterns were repeatedly changing, even though the stimuli and the behavioral responses were kept the same, we concluded that the bulbar patterns were dependent not merely on the stimuli as we observers knew them, but on the context, the previous experiences, the states of arousal and attention, the expectancies of responding to the stimuli, namely on the intentionality of the subjects, and, by inference, their goals and meanings, which are ultimately unknowable by us. The term is used here not in the post-Cartesian sense developed by Brentano and Husserl, in which mental images and representations "about" objects have an independent existence, are "caused" by stimuli, and "cause" goal-directed behavior in respect to objects (Searle, 1983), but in the more fundamental Aquinian sense that was re-introduced by Heidegger and his followers (Dreyfus, 1993; Wakefield and Dreyfus, 1990), in which the knowledge base and goals of a subject construct the behavior irrespective of consciousness or representations in the mind. Our studies of brain activity in relation to behavior have led us to the conclusion that "representations" in the sense entertained by Searle and others do not exist in brains (Skarda and Freeman, 1987; Freeman and Skarda, 1991), and therefore cannot serve the roles assigned to them by cognitive scientists.

    • Emergence of Complexity

  • This outcome led us to review our understanding of the concept of emergence and its close ally, complexity, in the context of vertebrate neurodynamics. A summary can be constructed by describing the process of recovery from surgical anesthesia, which is designed and used to suppress precisely those activities that are of central interest here. At the deepest level of anesthesia the interactions among cortical excitatory neurons are suppressed, and the response to a stimulus is a simple postsynaptic potential. Its rapid decay back to the baseline after every amplitude of input shows that the activity of a noninteractive collection of neurons manifests a zero level point attractor. This is the simplest solution set for a dynamical system.
  • Early during recovery the excitatory neurons excite each other, which induces and and sustains a background activity that is often called "spontaneous". This activity manifests an emergent property of the collection of neurons, which by interactions form a population. The population has a nonzero point attractor as well as a zero attractor. When the excitatory neurons also excite the inhibitory neurons and vice versa, the background activity becomes oscillatory. This manifests a limit cycle attractor, again an emergent property of the population of participant neurons owing to the inclusion of both types of neurons. The dynamic complexity increases with the degree of functional connectivity. Different populations have differing frequencies of interaction. When such groupings interact over longer distances as recovery continues, they cannot agree on a common frequency, and the oscillations become aperiodic. Thereby they manifest a chaotic attractor.
  • Despite this increasing complexity upon recovery to this level, intentional behavior does not yet return. Several more stages of emergent complexity must take place, that involve the interactions of multiple sensory, motor, and neuromodulatory systems comprising the brain. These further steps are virtually unknown, because the complexity of the functional interconnections and of the spatiotemporal activity patterns are beyond our present competence to describe and measure them. There is nothing in principle, however, to prevent us from doing so at some time in the future, other than the self-referential paradox of the mind coming to comprehend itself.

    • Propositions

    Some conclusions from these data are as follows.
  • 1. Brain function is hierarchically organized. Sensorimotor activities are carried out by individual neurons in networks, whereas perceptual activities in voluntary behavior are organized by large masses of neurons. The one is observed by recording microscopic action potentials, and the other by recording field macroscopic field potentials in EEGs (Basar, 1980).
  • 2. The action potentials are sensory-driven and motor-driving, and they manifest respectively the stimuli and the responses, but the macroscopic activity manifests the "meaning" of stimuli for the subjects. Meaning is the context-dependent neural patterning that is created by masses of nerve cells unique to each subject and that relates a stimulus to action toward a goal. It cannot be reduced to either nerve energy or information, hence the phrase "information processing" is useless for describing perception (Skarda and Freeman, 1987) and intentionality in either of two current meanings (Dreyfus, 1993; Searle, 1983).
  • 3. The behaviorally related content of brain activity is expressed in spatial patterns of amplitude of oscillatory EEGs. The time-dependent wave forms of the "carrier" waves are aperiodic (Freeman and van Dijk, 1987), which means that they are not reducible to discrete frequency components such as "40 Hz" (Gray, et al. 1989) and are therefore locally unpredictable. The spatial coherence rules out a stochastic origin. Hence the spatial patterns are products of deterministic chaotic generators (Freeman, 1991).
  • 4. Chaotic systems have the property of creating novel patterns in spatial and temporal coordinates (Shaw, 1984). The hypothesis follows that the chaotic dynamics of the neural populations comprising the limbic system is the source of the neural activity that controls the goal-directed movements in voluntary behavior. Hence the constraints on voluntary behavior are predominantly internal, not external.
  • 5. Past experience is embedded in modified synapses within the limbic system. An act of remembering is not a retrieval of stored information. It is the construction of a pattern (Bartlett, 1932), when an appropriate stimulus or preceding pattern constrains the limbic system into one of its learned basins of attraction (Freeman, 1991), thereby releasing a creative dynamic process, for which the outcome is never twice precisely the same. The variability is not stochastic; it is deterministic but locally unpredictable, as must be the case for behavior that is known to have the same characteristic.
  • 6. Learning by trial and error and by insight take place when a novel stimulus with reinforcement leads to the emergence of an unpatterned chaotic state during an orienting response, a "What is it?" reaction (Skarda and Freeman, 1987). The chaotic activity provides the substrate from which a new nerve cell assembly can form, leading to a new attractor and its basin of attraction. This process constitutes a bifurcation that may underlie adaptive voluntary behavior.
  • 7. Each time a change is made in the olfactory memory store by adding a new stimulus or changing the reinforcement contingency of a stimulus, the central patterns for other stimuli also change, reflecting the seamless fabric of a true associative memory system. It is this nonrepresentational dynamical system revealed by electrical recording that gives access to the intentionality of the subjects, in the fundamental sense developed by Heidegger and pursued thereafter by his students Buber and Tillich, of the flow of the environment through the mind/brain in the continuous construction of voluntary, purposive behavior.

    • References