Edelman's Theory of Mind

articles

Edelman’s theory of mind

Introduction

Edelman’s theory of the neurological basis of mind, developed in his book The Remembered Present (1989), is an interesting and original attempt to give a biological account of consciousness, one which brings together neuroanatomy, neurophysiology, psychology and philosophy. However, his account is written in places in a dense style that is quite difficult to follow. In trying, perhaps, to achieve precision and completeness in a subject beset with vagueness and the incompatible language of separate specialisms and traditions, he overloads many sentences with technical terminology and highly compressed content. So I shall draw not only on this work but also on a textbook written for the Open University course Biology: Brain and Behaviour (ed. Robinson, 1996), which gives a clearer rendering of Edelman’s central ideas on consciousness. Despite these difficulties, it is worth the effort to try to understand Edelman’s theory.

As he admits in the preface, although based on neuroscientific findings, much of it is speculative. He values his training in the sceptical experimental sciences, for the results and protection it affords, but has come “to be less convinced that there is a fundamental distinction among the intellectual procedures of science, of philosophy, and of everyday life” (Edelman, 1989, xvii). He notes the tendency among scientists to avoid the subject of consciousness as much as possible, because of empirical difficulties in determining its basis and because its subjective nature is not amenable to detached, scientific observation. Yet this avoidance where consciousness is a significant factor, for example in psychology, imperils reliable conclusions, even leading to the “assumption of hidden homunculi, vicious circularity in various arguments, or adoption of an elaborate and sanitary behaviorism that borders on the ludicrous” (ibid: xviii).

This article concentrates on two subjects central to his theory: “Neural Darwinism” and consciousness.

Neural Darwinism

Edelman’s theory of mind rests upon another theory of his, which he calls the “Theory of Neuronal Group Selection” (TNGS). This subsidiary theory was born out of two seemingly unrelated observations:

The world of stimuli as first encountered by very young animals is new and unfamiliar: they do not come pre-programmed with information on it, to be manipulated according to a set of rules, as in a computer program. While the real world obeys the laws of physics, this does not entail that it is ‘uniquely partitioned into “objects” and “events”’ (ibid: 40). The animal must have or develop adaptive criteria by which to make such a partition. Until the animal categorises the world as perceived through the stimuli it receives, the world is unlabelled, full of novelty.
The nervous systems of individuals (especially vertebrates) are enormously variable; that applies on all levels of description from the molecular to the behavioural. Despite individuals in a species having a shared gross neural anatomy, at a finer level the variability far exceeds that with which a man-made machine could cope reliably. Yet within bounds, individuals display a characteristic behaviour for their species, and this behaviour is adaptive, otherwise the species would not have survived.

Edelman saw a resemblance between these observations and others in evolution and immunology. This suggested to him that, in adapting to an unlabelled world full of novelties, the brain works by selective processes upon variance in a population of neurons, rather than by following preset instructions (ibid: 38-9). In both neural activity during perceptual categorisation and evolution, certain features in a population are either enhanced or diminished as an adaptive response to stimuli. Edelman coined the term “Neural Darwinism” in recognition of these similarities, notwithstanding the different mechanisms involved; the term expresses the idea that higher brain functions, such as categorising and learning, are implemented through selection of features in a variant population of neurons. The two observations indicate that neither the world nor the nervous system is what it appears: the world is not made up of fixed categories, but rather is categorised differently by species (and even individuals within a species) in ways that are adaptive to them; nor are nervous systems in detailed anatomy and function identical between individuals (ibid: 40-1). The ability of individual animals to categorise an unlabelled world, to generalise across categories, to learn and change behaviour accordingly, arises from three neural mechanisms: developmental selection, experiential selection and re-entrant mapping (ibid: 43-9). Each mechanism acts within and between groups of strongly interconnected neurons rather than through individual neurons. The evidence for group action is based on logic and experimental results (ibid: 51-4). It would seem to follow from the connectivity within a neuronal group that stimulation of a group neuron must have an effect on other group members, so that they act in concert. Experimental results confirm this: input stimuli produce a continuous output across neurons in a group with sharp discontinuities corresponding to group boundaries. Neuronal groups are especially visible where they have specific roles, for example, in the columnar organisation of the retina and visual cortex; in the latter there are columns devoted to detecting particular orientations of edges. An instruction-based model, where single neurons correspond to some specific event or object, cannot match the combinatorial multiplicity provided by the selection of neuronal groups in categorising external stimuli. There is no unique combination of groups that corresponds to a given category and each group can be involved in more than one function. It is this flexibility or degeneracy which gives neuronal group selection its power and which, by its lack of hardwiring, causes difficulties for theoretical models involving pre-programmed instructions (ibid: 50, 53).

The three mechanisms involved in categorisation form the core of TNGS. More fully, these are (ibid: 242-4):

During development, morphogenetic processes work selectively to form the gross anatomy common to members of a species, yet with enormous variation in the finest anatomical detail. These selective processes give rise to a primary repertoire of variant neural networks, involving populations of groups of neurons.
As a result of neural signalling during behaviour, another process occurs, which, although not affecting neuronal anatomy, selectively strengthens or weakens synaptic connections to give secondary repertoires of variant, functioning neural networks.
These repertoires are arranged in maps, where a map relates a collection of sensory receptors to an area of cortex. The maps are shaped both by evolution and the interaction between sensory and motor systems. Reciprocal connections between the maps permit re-entrant signalling. Environmental input and re-entrant signalling during sensorimotor activity result in some neuronal groups in local maps being competitively selected over others. Local maps are linked by re-entrant pathways into global mappings, which enable not only perceptual categorisation and the assignment of values based on homeostasis but also the temporal ordering of stimuli into coherent experiences. These changes, brought about by behaviour, lead to behavioural adaptation.

How does such a selective system as that described by the TNGS enable perceptual categorisation and generalisation?

Perception is the adaptive discrimination of an object or event from background or other objects and events. Generalization refers to the treatment […] of a more or less diverse collection of such entities as equivalent (ibid: 49).

The TNGS proposes that categorisation arises from two or more separate channels carrying signals to maps, where each channel might be for a different sensory modality. Repeated stimulation preferentially activates certain neuronal groups, giving rise, through enhanced synaptic connections, to strengthened pathways. Re-entrant signalling between local maps links different sensory modalities into global maps. Global maps are thus confederations of local maps, dynamic in structure, both sensory and motor, and they communicate with unmapped regions such as the brain stem, basal ganglia, hippocampus and cerebellum. Combining the input of different senses, when moving with respect to the objects perceived, enables a continuous spatio-temporal characterisation to be constructed. Repeated stimulation and re-entrant signalling occur not just on encountering the same object, but also where the same category-defining features, shared by different objects or events, are encountered. This is how perceptual categorisation and generalisation are implemented.

Consciousness

Consciousness is a process. It depends on the organisation of certain parts of the brain rather than the whole brain. Yet those parts directly involved in consciousness do not act in isolation, but require for their functioning other brain regions which have evolved and operated prior to consciousness to produce various kinds of behaviour (ibid: 91-2). Edelman distinguishes between a fairly basic “primary consciousness”, one found at the very least amongst the higher vertebrates, and a much more advanced “higher-order consciousness” found pre-eminently in humans but also to some extent in the great apes.

Consciousness has adaptive significance (ibid: 92, table 5.1). Primary consciousness:

Provides the means for relating an individual to its acts and rewards.
Helps in abstracting, organising and giving salience to complex changes in the environment detected through different sensory modalities.
Helps sequence tasks over short periods, with some adjustment to routine actions when conditions change.
Provides the conceptual precursor to linguistic communication.

Higher-order consciousness does all the above, plus:

Helps sequence complex learning tasks and corrects automated, routine actions during changing conditions.
Allows long-range anticipation of events and consideration of their relationship to the past, through the use of long-term memory.
Increases adaptability by planning or “modelling the world” outside of immediate events. Enables different scenarios to be imagined.
Permits the comparison [evaluation] of outcomes on the basis of values and previous choices.
Enables plans and memories to be reorganised.
Is necessary for (and is characterised by) linguistic communication.

Both types of consciousness free “animal behavior from the tyranny of ongoing events” (ibid: 92). They enable a composite representation to be constructed, relieved to some extent of the never-ending stream of signals, without which an animal would have to attend to every disparate event as it happened. Primary consciousness provides relief over short periods; scenes are constructed in which perceptual categorisations of events and objects, through different sensory modalities and modulated by corresponding motor actions, are ordered into a coherent temporal sequence. It enables the salience of stimuli to be evaluated in terms of the animal’s needs and thus to influence its behaviour. Higher-order consciousness enables even more freedom, with the use of concepts, symbolic representation and greater use of memory to develop a coherent model that encompasses past, present and future. This gives the animal possessing it a selective advantage: to be able to recognise an opportunity or danger, for example the changing position of a predator over time, and with that recognition to take appropriate action.

Consciousness results from the interaction of two types of nervous organisation: one concerned with identifying and satisfying physiological needs and the other with perceptual categorisation. It requires the use of memory to relate how needs have been satisfied previously to the current characterisation of the environment. Without this memory usage, learning and the modification to behaviour that results from learning would not be possible. From these considerations, Edelman enumerates the necessary conditions for consciousness. To be present in an organism, its nervous system must have (ibid: 93):

The ability to categorise perceptions. This categorisation is done on the basis of action and a sufficiently rich set of parallel sensory channels of different modalities. For example, the hierarchical touch pathway provides the ability to represent different types of touch on the skin, such as a touch moving in different directions over the same area.
Memory involved in a process of continual re-categorisation through associating new events with previously stored information.
A capacity for learning, changing behaviour through relating the present state to past states. It uses remembered perceptual categorisations linked to ethological values to relate the past to the present. The values, which reflect hedonistic, appetitive and pain-avoiding behaviours, derive from primitive emotional drives working to maintain homeostasis.
An ability for discrimination between self and non-self, where “self” and “non-self” are to be understood in a strict biological sense, neither in the personal or psychological sense of “self-awareness”, nor in the social or philosophical sense of “personhood”.

Learning, by drawing on value-dependent behaviours and perceptual categorisation, uses the two types of nervous organisation mentioned above. These two types, involving distinct pathways, structures and mechanisms, are also the basis for the distinction between the self and non-self. The self arises from autonomic activity involving the limbic system, particularly the hypothalamus with its role in the endocrine system; it is concerned with maintaining homeostasis, which it does by responding to signals from interoceptors through the use of hormones. The non-self arises from processing in the thalamus and cortex, using signals from exteroceptors and proprioceptors which give information about the external environment and thereby help also define the boundaries of the self. The first type of nervous organisation operates within developmentally given parameters, which in ethological terms are relatively fixed. The second type, through motor neurons and exteroceptor-connected sensory neurons, operates by behaviour and experience to interact with the world (ibid: 93-4).

Disagreeing slightly with Edelman here, I would put proprioception more in the self category, since proprioceptors detect changes in the body brought about by movement and muscular activity and we normally consider our own body to be if not part of ourself then intimately connected with it. While Edelman recognises that information from exteroceptors contributes to the boundaries of self, I would go further and say that as well as giving the non-self, the information contributes to the self by positioning the latter centrally in the external world, as an active agent. The self, then, is not just constituted by autonomic inputs and responses to maintain homeostasis. It is given definition, as an entity distinct from but acting within the external world, by inputs from proprioceptors and exteroceptors and the motor responses to them.

The different systems and functions within this organisational arrangement can operate independently of consciousness; their mechanisms are like those described in the TNGS with respect to perceptual categorisation. Consciousness is both an outcome of their interactions and the means by which their level of functionality is raised. To account for awareness and perceptual experience in guiding actions, we need to consider the sufficient conditions for primary consciousness. Naturally these incorporate the necessary conditions, but we need to say more on how they work together. Especially, to provide a fuller description of the mechanisms of memory, the cortex and neural structures (such as the hippocampus, the thalamus, the basal ganglia and the cerebellum) in linking successions of sensory and motor categorisations so as to produce perceptual experience and consciousness itself.

Edelman proposes that primary consciousness arose from two evolutionary events leading to:

Development of memory repertoires “dedicated to storing past matches of value to perceptual categories” (ibid: 95), where value is derived from maintaining homeostasis. Such memory repertoires involve parts of the brain capable of concept formation, distinguishing objects from actions and categories from relations.
Development of new circuits allowing re-entrant signalling between these memory repertoires and those currently engaged in sensorimotor sampling of the environment for perceptual categorisation. This depends on the first event, that is on the prior evolution of memory for conceptually relating value and perceptual category.

The structures and processes giving the distinction between self and non-self are not sufficient by themselves to lead to primary consciousness. One must add those involved in the re-entrant connection of past value-category matches to the categories arising from current sensory stimuli and motor responses. In this way, present perceptual categorisations are linked in real time with past value-category correlations before the latter are altered by the value-giving part of the system. As Edelman says “[a] kind of bootstrapping occurs in which current value-free perceptual categorisation interacts with value-dominated memory before further contributing to alteration of that memory” (ibid: 97). He asserts that primary consciousness emerges [my emphasis] from that process. Primary consciousness alters the relative salience of events as perceived through stimuli, thereby altering goals and actions. Without it, salience would be dominated by the most recent events, rather than aligned to the most adaptive course of action.

To recap and expand on earlier comments, the self is determined by the activity of brain regions concerned with maintaining homeostasis, that is, the brainstem, pons, mesencephalic reticular formation, hypothalamus, amygdala, septum and fornix. The non-self is associated with cortico-thalamic links and cerebellar and hippocampal loops (except those in the fornix). Value derives from maintaining homeostasis and satisfying appetites. Updating memory dominated by value requires the ability to form concepts, which in turn requires a more generalised categorising ability than that employed in categorising perceptions. This concept-forming is assumed to take place in the frontal, temporal and parietal lobes as well as the cingulate gyri (ibid: 98-9).

Primary consciousness is not a uniquely human phenomenon: the great apes and possibly other animals have the requisite nervous systems too. With primary consciousness, an animal can generate representations of events and consequences; this enables the formation of concepts, where a concept in this sense is “the amalgam of the representation of an environmental event and the internal consequences or value it achieved” (Robinson, 1996: 326). The concepts that are characteristic of primary consciousness precede meaning and language. They are the “gateway to higher-order consciousness” (ibid: 326), both as it evolved and as it develops in the individual.

Edelman considers that higher-order consciousness began to develop with brain changes which permitted the formation of categories related to an enhanced concept of self, beyond the biological (Edelman, 1989: 97, Fig. 5.1). It is characterised by subject-predicate relations based on the distinction between “self” categories and “non-self” categories. I speculate that what made these relations apposite was the development, through changes in the limbic system and frontal lobe, of emotions from more primitive drives, which, as the basis for motivation, give direction to the self as an autonomous actor. This self is further enhanced by social interaction. Higher-order consciousness was also dependent on the development of memory processes for placing concepts in an ordered relationship: in other words, for imposing order on the world. Ordering and manipulation of concepts require symbolic representation subject to syntactical rules. Thus higher-order consciousness requires at the very least a precursor to language. “But only with the appearance of true language in a social context can this form of consciousness fully flourish” (ibid: 103); hence when all the structures for language are in place. Speech was only possible with the development of the vocal tract. The meanings of words are based on concepts, which required the evolution of brain regions capable of forming them. To use words in a comprehensible way requires Wernicke’s area and to apply grammatical rules and generate the motor output for speech or writing requires Broca’s area. Other neighbouring regions are also thought to be involved (Robinson, 1996: 290-4).

Higher-order consciousness may well be restricted to humans and in some measure to the great apes, but it is built upon primary consciousness which has a wider spread. Chimpanzees illustrate both types. 18-month-old chimpanzees display the pre-syntactic processing of primary consciousness in being able to recognise that two objects belong to the same category and also that two different objects have the same organisation or relation. They can recognise sameness and difference, make analogies, even assess intentions and act on these. Some level of higher-order consciousness is evidenced by the ability of chimpanzees to imbue symbols with meaning and show a degree of self-awareness (passing the mirror recognition test) (Robinson, 1996: 326-7).

Conclusion

Consciousness and mind are natural phenomena: they are the product of nervous systems that have reached a certain level of organisational complexity. No homunculus is needed to categorise perceptions or make generalisations: the strengthening or weakening of neural pathways is a natural consequence of synaptic sensitivity to the degree of repetition of stimuli. Consciousness enables the animals possessing it to behave adaptively to changes in the environment. It frees them from having to respond immediately to every external event because they have a clearer composite picture of the world; they can model it, anticipate consequences and form goals. Consciousness results when an animal is able to categorise perceptions, match them to homeostatic values, re-categorise past value-category matches against new perceptions, learn and change behaviour accordingly and distinguish between self and non-self. The self is identified most strongly with the parts of the nervous system associated with maintaining homeostasis, satisfying appetites and avoiding pain. The non-self is identified with the thalamus, cortex and cerebellum, where perceptions are formed and categorised, and motor responses are initiated. Consciousness of a higher order involves further categorisation to give an enhanced concept of self, a subject acting in the world. This consciousness is the product of nervous systems able to order and manipulate concepts by using symbolic representations. It flourishes with the development of language.

Edelman has not explained how consciousness actually results from the conditions he has described, but there is enough in his theory to indicate that the way forward is to combine a biological approach with a revised philosophy. The latter should no longer rely on an outdated metaphysics, in which consciousness and mind are occult and possibly immaterial, but rather on the view that they are a natural, explainable part of suitably complex nervous systems.

Bibliography

Edelman, G.M. (1989) The Remembered Present, New York, Basic Books.
Robinson, D., ed., (1996) Book 2 Neurobiology, (Open University text for course SD206 Biology: Brain & Behaviour), Milton Keynes, Open University.