The Autonomous Brain

A Neural Theory of Attention and Learning

Peter M. Milner

• Publisher: Psychology Press
• Year: 1999
• ISBN:
  9781135670269
• Edition: 1
• Topic: Psychologie
• Subtopic: Geschichte & Theorie in der Psychologie

The behaviorist credo that animals are devices for translating sensory input into appropriate responses dies hard. The thesis of this pathbreaking book is that the brain is innately constructed to initiate behaviors likely to promote the survival of the species, and to sensitize sensory systems to stimuli required for those behaviors. Animals attend innately to vital stimuli (reinforcers) and the more advanced animals learn to attend to related stimuli as well. Thus, the centrifugal attentional components of sensory systems are as important for learned behavior as the more conventional paths. It is hypothesized that the basal ganglia are an important source of response plans and attentional signals. This reversal of traditional learning theory, along with the rapid expansion of knowledge about the brain, especially that acquired by improved techniques for recording neural activity in behaving animals and people, makes it possible to re-examine some long standing psychological problems. One such problem is how the intention to perform an act selects sensory input from relevant objects and ensures that it alone is delivered to the motor system to control the intended response. This is an aspect of what is sometimes known as the binding problem: how the different features of an observed object are integrated into a unified percept. Another problem that has never been satisfactorily addressed is how the brain stores information concerning temporal order, a requirement for the production of most learned responses, including pronouncing and writing words. A fundamental process, the association between brain activities representing external events, is surprisingly poorly understood at the neural level. Most concepts have multiple associations but the concept is not unduly corrupted by them, and usually only a single appropriate association is aroused at a time. Furthermore, any arbitrary pair of concepts can be instantly associated, apparently requiring an impossibly high degree of neural interconnection. The author suggests a substitute for the reverberating closed neuronal loop as an explanation for the engram (active memory trace or working memory), which may go some way to resolving these difficulties. Shedding new light on enduring questions, The Autonomous Brain will be welcomed by a broad audience of behavioral and brain scientists.

1
Introduction

WHERE TO START?

How does anyone set about trying to discover how the brain regulates behavior? Dismantling it, the traditional approach of budding scientists to alarm clocks (in the days when alarm clocks were full of clockwork), may reveal how it looks, but not how it works. There is no-one to whom the investigator can write for a manual or a blueprint. Confronted by billions of cells connected by untold miles of threadlike processes, early psychologists were inclined to give up on the problem and concentrate on making sense of the behavior it produced.

About half a century ago improved techniques of staining made it possible to follow many of the brain’s connections. At about the same time, the microstructure of neurons and their electrophysiological and chemical properties began to be investigated more thoroughly. As knowledge about the nervous system grew, the possibility of discovering how it works began to seem less remote. Also more worthwhile, because the knowledge could be applied to a greater body of neural information to explain behavior and its impairments.

Anyone who has looked at the circuit diagram of a moderately complicated piece of electronic equipment, or the printout of a computer program, in the hope of discovering its purpose, will realize the hopelessness of trying to figure out what sort of behavior the brain will produce simply by studying its connections and the properties of the component neurons. Even knowing the purpose of the equipment, and exactly how all the components work, does not make it easy to relate structure to function in a complex system (Braitenberg, 1984).

Fortunately, in the case of the brain, we would be satisfied to discover just the general principles of how it produces the behavior that we can observe. A close analogy is seen in the inventor, who knows the purpose of his invention and has to think of a good way of assembling it from the parts available. Knowing something about the structure of the brain can help, but by far the most useful information relates to the behavior it produces. In any case, it is easier to observe behavior than brain activity, and we know a great deal more about it. Therefore, basing our theories of brain function on a knowledge of behavior would seem to be more sensible than trying to do the opposite.

Despite its reasonableness, this precept is not often strictly followed. After all, the brain, though not very accessible, is made of real stuff; behavior is insubstantial. Pavlov (1927), following Sechenov (1863/1965), made the mistake of starting with the spinal reflex, a physiological mechanism he knew something about, and letting it dictate his concept of behavior. The behavior on which he based his theory of learning was that of dogs responding in a quite atypical and artificial situation.

Lashley (1950), whose life also was devoted to the problem of brain and behavior, concluded early in his career that brain processes must be very diffuse, believing that individual neurons and their connections have little influence on behavior. Lashley also had a long-standing prejudice against attributing learning to synaptic change. To some extent, these attitudes about brain processes colored his perception of behavior (Lashley, 1924).

A third major contributor to neural theories of behavior, Hebb (1949,1980), took the anatomical findings of Lorente de Nó (1938) as the inspiration for his cell assembly concept. Although he was a strong advocate for basing neuropsychological theory on behavioral evidence, he frequently interpreted behavioral data in terms of his anatomically inspired model of brain function.

Once the behavior to be explained or simulated has been thoroughly studied, the design of a model may be implemented in various ways. It is possible, for example, to use electronic components to construct a machine that locates sources of energy and recharges itself, mimicking the appetitive behavior of a simple animal. A working model is very convincing, but there are advantages in conceiving a hypothetical model based on components as similar as possible to those found in the nervous system. In the first place, such a model allows the correspondence between the design and the real thing to be assessed more directly. Also, if it proves necessary to postulate hitherto unknown physiological mechanisms to account for certain behaviors, the hypothetical model makes it possible to seek confirmation in the real nervous system.

WHY DID THE CHICKEN CROSS THE ROAD?

An important theme of this monograph is that behavior is not always determined by sensory stimulation. Of course, this is not a new discovery. Behavioral autonomy is something most people throughout the ages have taken to be self-evident. Many still do, but they typically attribute their autonomy, not to a material brain, but to an incorporeal ego, or “self,” that imbues the body with life and purpose.

This attribution had an unfortunate consequence. When psychologists towards the end of the 19th century began to question the notion that a physical body can be controlled by a disembodied spirit, they also doubted, or more likely did not even consider, the possibility that the body itself might possess the means to initiate behavior. They assumed the body to be inert except when activated by sensory input.

Nevertheless, as any electronics engineer will testify, complicated material structures readily make use of whatever energy is accessible to generate unexpected (and usually undesired) output. Few mechanisms can equal the complexity of the mammalian nervous system. It is a scene of literally billions of interrelated chemical reactions. Newborn infants, when awake, usually find some way of making their presence known. The more impoverished the external stimulation, the more vigorous their efforts are likely to become (at least in the short term).

Lashley (1951), and Hebb (1949, p. 3) were among those who criticized the reflex model, in which behavior is attributed entirely to sensory input. Hebb did not believe the internal influences on behavior to be innate, however. His view was that they develop as a result of previous stimulation. In any case, the criticisms of these researchers had negligible effect. The notion that behavior is always a reaction to a stimulus is so ingrained that the name applied by psychologists to an element of behavior is "response."

PERCEPTION, INTENTION, AND ATTENTION

The possibility that the nervous system may generate responses independently of sensory input has important implications for the way we view sensory mechanisms. Because typical neural theories of behavior depict the nervous system as a device that processes sensory input to derive adaptive responses from it, perception and memory are given priority treatment. Often, response mechanisms are ignored or taken for granted.

FIG. 1.1. Hypothetical goal-seeking bug.

It is hardly surprising, therefore, that theories and research aimed specifically at explaining perception and recognition are typically based on the assumption that these processes can be studied in isolation, with little if any reference to activity in other parts of the brain. Theories of recognition, for example, concentrate on the creation of memory traces and their relation to subsequent sensory input. Until recently (Desimone & Duncan, 1995; Rizzolatti, Riggio, & Sheliga, 1994), account was rarely taken of the fact that our motivation and actions have a decisive impact on what we perceive, though Sperry (1952) long ago pointed out that the sensory systems evolved to improve the performance of the motor system, not the reverse.

The problem is exemplified by the dilemma of the goal-seeking bug shown in Fig. 1.1 (Milner, 1970). An odor to the left of the bug energizes predominantly its right legs, turning the creature toward the olfactory stimulus. Light falling on the bug’s left eye energizes predominantly its left legs, turning it away from the light. But what happens if the bug is confronted by a floodlit plate of food? Or worse, suppose the food is in front and a light to one side? With two sources of sensory input trying to steer the bug in different directions, it would never reach the food, nor would it run directly away from the light. To function effectively, the bug must be able to ignore all stimuli except the one it considers most important.

This sort of dilemma is not a freak accident. It happens all the time. On a summer evening, a hunting bat may receive sonar echoes from dozens of flying objects, but it must “lock on” to only one of them at a time if it is to catch anything. In the presence of several stimuli, animals may be forced to choose between possible courses of action and, having decided on one, they must prevent any irrelevant input from interfering with the behavior.

It might said that animals focus their attention on the object they most urgently need. If that object is not present, they may settle for one of lower priority. A very hungry and less thirsty animal would be able to drink if food were absent because, although the sensory path for food may be more strongly sensitized than that for water, the facilitation is ineffective when there is no food.

Theorists often pay lip service to the influence of attention on perception (the word “attention” here, as elsewhere throughout the book, means selective attention, a process that highlights a particular stimulus), but they rarely, if ever, incorporate attention into their theories in a formal or useful way. Attention admittedly is difficult to incorporate into neural models of perception because it involves taking account of the animal’s intentions. Although in experiments, stimuli can be controlled very precisely, determining an animal’s intention is more difficult.

Nineteenth century introspectionists such as Külpe (1901) speculated at length about the influence of intention on all aspects of behavior, but intention had no place in the behaviorists’ reflex-based theory. In that theory, responses were supposed to depend only on stimuli, certainly not on mental concepts such as intention. The pioneer behaviorists would have found the possibility that planning a response could induce profound changes in the sensory systems, as demonstrated by Moran and Desimone (1985) for example, too unsettling even to contemplate.

The new wave of conditioning theorists, including Mackintosh (1975), Rescorla and Wagner (1972), and especially Grossberg (1975), attempting to explain the results of Kamin’s (1969) “blocking” experiments, recognized the relation between motivation and attention. However, any influence this might have had on neural theories of perception has been slow to make itself felt.

More recently, Rizzolatti (1983) has speculated along similar lines. On the basis of results from single-unit recordings in cats and lesions in monkeys and human patients, he argues that attention is not a centralized sensory process that can be directed at will to enhance any stimulus, but that it consists of the facilitation of sensory neurons by premotor activity during the preparation for a response.

The evidence supports this view with respect to attention directed to a point in space, but for a response that is to be directed to a particular object, Rizzolatti provides no convincing answer to the problem of how the premotor activity locates sensory neurons that represent that object. R.Miller and Wickens (1991), in discussing their version of the cell assembly theory, also stress the importance of selective attention. They note the relation of attention to the motor system, but do not develop its role in the execution of responses. This problem was raised some time ago (Milner, 1974) and is discussed at greater length in chapter 4.

Most investigators of perception concentrate on sensory input and its aftereffects, paying less regard than does Rizzolatti to contributions from the response system. This has led to a number of problems that will subsequently be discussed.

One obvious problem is that almost everything we see or hear has innumerable potential associations, only one of which is aroused at any time. Not only does attention select the items relevant to the current task from the vast array of sensory input, it also determines which associations of those items should be followed up. It is obvious, and frequently noted, that attention focuses on one or other feature in the sensory field, but few investigators have attempted to explain how it identifies and targets the relevant stimuli.

One theme of the present thesis is that sensitization of the appropriate sensory pathways is an essential component of any response. The urge to perform some act comes first, exe...

Table of contents

• Table of Contents
• List of Figures
• Preface
• 1 Introduction
• 2 The Behavior Model
• 3 Neural Representation
• 4 Stimulus Equivalence and Attention
• 5 Temporal Order
• 6 Memory
• 7 Theories of Amnesia
• 8 The Motivation/Response System
• 9 Basal Ganglia: Behavioral Functions
• 10 Envoi
• Bibliography
• Author Index
• Subject Index