Dreams as the real road to the nature of cognition

Henrique Schützer Del Nero

INTRODUCTION

The discussion about mind predicates is in the core of this article. However, the form adopted to access the problem might be considered unusual. This is why I warn from the beginning that speculations are the substratum of reflection. I shall not talk about anything in depth but I shall give the reader a broad overview of a way of seeing dreams as the privileged locus of mind appearance and analysis. If this sounds at first glance too Freudian it is only in a very distant way. I am not adopting a psychoanalytic background to do this work. Instead, I am suggesting that the knowledge that comes from different fields, something that I would like to call the Cognitive Science dream, can mutually help the comprehension of the relation between mind and brain.

MIND AS CONSCIOUSNESS

Trying to define what the most important feature of the mind is I shall adopt consciousness as its deepest mark. But what is consciousness? An entity, a class of processes, a class of states, etc.?

We define consciousness as being the entity that subsumes the following "objects": volition, "freedom", and the phenomenological screen of sensations, feelings, thoughts and responsibility. To be conscious is to be at least in one of the above states, or to possess one of the above "objects".

We must distinguish between automatic and voluntary behavior. "Freedom" appears when we perform voluntary acts, as a sensation of being capable of acting differently the way we are doing, the sensation of being able to choose the ways to handle the problem.

The phenomenological screen is that class of continuous experiences that are exhaustively treated by the gestaltists and by all that adopt descriptive psychology as a tool to understand the mental. Sensations are tied more to the objects of perception, be they external or internal (e.g. pain). Feelings are closer to happiness, fear, anxiety, lust, etc. Thoughts are more elaborate sequences of mental objects, being almost sentences, with syntax, semantics and pragmatics in play. Real thoughts, and not the sensation of being thinking, are silent talks or writings. They don't lack syntactical form, and even when they are false judgments, intentional or not, it is quite unusual for them to lack control over the rules of formation of language, over the rules of verification of truth and over the contexts where the sentences acquire their full meaning.

Responsibility plays an important role because the ability to be the subject that performs a certain class of acts being an object of external judgment for them and being the recipe of internal guilt or pride for them is one of the deepest evolutionary conquests of human primates.

Then, we define the mark of the mental as being consciousness. And we define consciousness as being at least one of the following objects: volition, "freedom" and phenomenological experiences.

DREAMS AS CONSCIOUSNESS

As far as one knows dreams reveal at least one of the features above described to qualify consciousness. They present phenomenological experience, in spite of lacking voluntary control (almost always, because sometimes one is able to control the sequence of dream experiences and even to interrupt a dream episode), and the sensation of freedom. Responsibility in the case of dreams is a more complex issue because, depending on one's norms and personality, even the non-purposeful representations are subject to inflict guilt.

Briefly, dreams are phenomenological experiences, therefore mental and conscious.

PSYCHOSES AS CONSCIOUSNESS

Psychoses, broadly considered as affections of thoughts, feelings, sensations, judgments and experiences, are conscious also. If they lack properness in the semantical sense (e.g. the truth-verification of a delusion) and pragmatical sense, failing to evaluate the bizareness of a sequence of thoughts that is not plausible, only in very special sub-cases they lack syntactical coherence: the case of hebephrenic schizophrenia and the neologisms of some unusual forms of schizophrenic disorder.

Psychose are also conscious phenomena, being deviances of thought, feelings and responsibility. The relation between normality, consciousness and responsibility, from the legal and ethical point of view, gets impressive when the psychiatrist judges the degree one has to evaluate the consequences of his, or her, crime (imputability). Even in the Civil Law there must be a degree of sanity to enable one to contract. The mental in these cases is not only an animal predicate, it is what scaffolds the relations of culture and society.

NORMALITY

These item is not an thourough one but only a trial on definitions that are to be held to build our hypothesis. Normality is a class of states, a quasi-set, that is defined by being awake states, hence able to interact with the environment, and maintaining a class of behavioral and mental attitudes that are coherent in the syntactical, semantical and pragmatical sense.

Psychoses are defined as deviances in the normal class of states, so they have eo ipso to occur during the being awake state. What characterizes psychotic deviances is a rupture with syntactical (very unusual, as we said), semantical (very common in hallucinations) and pragmatical (common in other states due to the implausible contexts where statements are made) rules.

We say that the major part of mental disorders mainly affect the pragmatical character because one can suppose that a paranoid is an example of semantical deviance, making incorrect judgments about reality. What is abnormal is not to say "I am being persecuted" because there are many people that really are, but instead being unable to justify the framework that makes persecution plausible. This is why the misjudgment is a pragmatical one and not a semantical one.

Dreams, in the sense above described to characterize "normality", are not normal or abnormal because they lack one of the main criteria to be defined as such: being awake. This is why we prevent classifying dreams as normal or abnormal ones. They sometimes violate some coherence but there is no overt behavior, nor control, nor freedom to characterize them as anomalous states.

The thesis being launched is that dreams are mental phenomena in its purest form and the lack of control and "freedom" they present, instead of being a deviant mark, has to be considered a different class of mental phenomena that might resemble an autonomous system with no environmental feed-back. In other words, what characterizes normality and the strict obediance of the logical rules is not the isolated system, but it is the environment. According to this consideration, consciousness, with its deep experience of freedom, of volition, and of control, is a matter that comes from the external world and not from the internal one. The internal one is evaluated only during dreams when coherence is often violated. During psychoses what happens maybe is a deviant feeding of the environmental parameters, something that might resemble autism, here meaning "isolation". Isolation in dreams represents a lack of sensorial and motor contact with the environment, and during psychoses, in spite of the individual being awake, this loss tends to occur in different degrees.

Briefly, if the mind is considered mainly as consciousness; if consciousness appears only if one of the following predicates is present: phenomenological experience, "freedom", volition-control and responsibility; then normality is a conscious class of phenomena that involves all the conscious predicates. Psychoses are awake states that affect semantical and pragmatical rules. Dreams are conscious phenomena that are neither normal nor deviant, because they are not awake states, in spite of being mental, therefore conscious, ones.

The discussion of these ideas will be important because trying to model a cognitive architecture, often imposes the question about the nature and the loci of mental phenomena. If judgment, properness, volition and "freedom" are not internal states, but are due to environmental feeding of the system; if dreams are mental phenomena, quite often bizarre, then we conclude that mental phenomena, conscious ones, dissociated from the permanent control of external variables might be bizarre. What gives us coherence is not the labor of our inner mental life, but a peculiarity that this inner mental life is constrained all the time by environmental parameters. What sort of volition, freedom, and so forth that is hetero-determined!

SYSTEMS AND MODELS

One of the most complicated issues in modern science is to define a cognitive system and to model it. The efforts that are commonly called Cognitive Science come from a class of proposals:

i) the strict identification between input and output held by behaviorists has to be abandoned.

ii) the notion of intermediary process, be it representational or not, has to be back as a class of variables that define a cognitive system.

iii) the efforts to define and model cognitive systems has a double advantage: better artifacts can be built, and the human mind may be understood in a more quantitative and formal aspect.

Cognitive Science, however, presents as a common ground only these three aspects, eliciting then a typical "diaspora" because there are many ways, from the conceptual point of view, to treat these proposals.

All the tensions between rival models and representational tendencies are not going to be investigated. I am only going to show that a class of systems that are really cognitive in depht must present consciousness. Consciousness as defined above must present at least one of the features described .

To build a cognitive apparatus, or to launch a genuine cognitive model imposes, to the proponent, the penalty of clarifying certain conceptual aspects involved. There must be a strict observation and separation between contents and forms. These two terms are very apt to elicit confusion, but I shall define what we consider the semantical aspect of cognitive systems and their formal aspect.

FORM AND CONTENT

When a relation between two sets is defined, we establish for the individuals that are part of each set, a rule that connects individuals from one set to an individual of the other set. If we say that there is rule that connects the number 2 in one set with the number 8 in another set, there are many ways we imagine this rule to be. For example we may have a rule of adding 6 to 2, or we may have one that makes the 2³, and so forth. Forms in the sense we are adopting here are rules that connect an individual from one set to individuals in another set. Contents are the individuals themselves.

If we say that f(x) = y and that f(x) = x³ , we define the "form" as being the rule that connects x's and y's, in this case taking each "x" multiplying it three times by itself, obtaining then the "y" that represents this operation. The form is the rule, the contents here are 2 and 8.

Modeling cognitive phenomena is a very complicated issue because problems of content and of form are involved at the same time.

There are two problems regarding cognitive models: first a position regarding mental or cognitive individuals must be adopted. Second: we have to discover or to propose the forms, or the rules , or the equations that relate individuals from each domain. Several are the domains: we have one that relates sensorial inputs and behavioral outputs; an other that considers the intermediate domain of processing: the neurophisiologycal one; another considers an intermediate processing domain that is not neurophisiological but "mental", or "intentional" or "conscious". Our class of problems, related to cognitive modeling is to adopt the "two sets" view: input and output ( a behavioristic style), or to adopt one of the following related sets: input-neurophisyological-output, input-mental-output, neurophisiological-mental or mental-mental.

We think that the cognitive enterprise has been dealing with mental-mental models, because if one accepts the representationalistic aspect of cognitive phenomena, one is pushed to dissociate the neural substratum from the mental predicate. This is what gives us the token-identity thesis that affirms only that there is a many to many relation between the physical and mental individuals. The pursuit of a strict equivalence between these two sets is denied by a large class of articles that preclude intentional idioms to be translated unto physical ones. The problem of intentionality is the major obstacle to the strong reduction, making all the efforts to relate mental variables to physical ones a matter of speculation. There is an a priori impossibility to relate the mental with the physical in this sense, then Cognitive Science, when it accepts the representational point of view, precludes the models to deal with the neural substratum.

Dreams however, are very interesting phenomena because there is at least one class of physical variables that can be tied to the mental one in a strict relation of equivalence: x is dream content only, if and only if, x is REM state. There are no equivalences between types of dreams and types of REM neurophisiologycal variables, and this is why many can consider this a form of token-identity. I accept that this is still too weak but it is more in terms or the relation between the neural and the mental that many other proposed equivalences. I shall explain: if the identification between neural states and mental ones is very difficult, allowing a class of equivalences that are typical of a classical discipline called Neuropsychology; if the identification between mental states and neural states is precluded in another argument by the impossibility to reduce intentional idioms to physical ones, then the relation between dream contents and REM sleep can be considered: either a matter of Neuropsychology or a closer relation between the mental and the physical because dreams in this case would lack "intentionality".

Of course, intentionality is a dangerous concept with a multitude of implications. I am not going to deal with all the subtle aspects of it. Instead, I am going to suggest the following reasoning:

i) The mental cannot be translated to the physical because it presents intentionality (Brentano's thesis [1] and Fodor's position [2] . The argument is so strong that even Clifford Hooker in his magnificent "Towards a General Theory of Reduction" [3] accepts the thesis of non-reducibility in terms of intentional idioms).

ii) The mental is conscious

iii) Dreams are conscious

iv) Dreams lack control, volition, responsibility and freedom

v) Conscious in the normal awake state exhibits control, "freedom", responsibility, etc

vi) Consciousness exhibits intentionality

vii) Consciousness is directly expressed in idioms that are intentional idioms

viii) Dreams are remembered and expressed in intentional idioms

ix) During the dream state what the subject experiences is only the phenomenological character of mental and conscious phenomena.

x) Dreams can be related to the neural substratum by a class of strict equivalence: every dream state is a REM state.

xi) If dreams are considered intentional states then the relation between mental and neural that can be grasped from this pair is a typical subject of the Neuropsychologycal style, i.e. a class of neural-mental models.

xii) if dreams are considered only in their formal aspect they represent a neural-neural model, i.e it is pure neurophysiology

xiii) if dreams instead present conscious experience but lack of control, and if "awake" normal states present control , volition, etc,

xiv) if wakefulness is immediately related to language and if dream states are mediately related to language then

xv) we can suppose that intentionality is a mark of consciousness as Brentano said but related to that part of consciousness that involves control, immediate usage of language, freedom, and contact with the environment. In this way, dreams might be considered as lacking intentionality in spite of being conscious, because of phenomenological experiences.

If this is legitimate reasoning, dreams represent a privileged locus of analysis in terms of modeling cognitive phenomena, because they are mental in its full form but they are not intentional and they do not make direct usage of language.

If we suggested above that "freedom", control, volition and responsibility are hetero-determined, i.e. they are determined by the continuous feeding of environmental parameters, we can here suppose that intentionality and language in its direct form is a matter of hetero-determination.

Mind in its autonomous sense can be grasped only by the analysis of dreams qua contents of experience, quite often bizarre, and qua forms as REM sleep.

The strongest part of these considerations is that if we model awake cognitive phenomena we cannot escape from the traps of intentionality, then all the models, be they symbolic or connectionist ones (to allude to the rival computational models in Cognitive Science) are mental-mental models. Neural-mental models may be developed in a more conceptual way than Neuropshychology, touching upon some terrible problems, by considering dreams as the purest mental forms, here deprived of intentionality and of language, in its immediate sense, allowing us to consider the real status of coherence, properness, volition, control, freedom and responsibility.

I must go back to the sort of considerations I started regarding the relation between forms and contents.

It is a common problem in Cognitive Science not to give a strict definition of the sort of sets that are being considered. Of course, authors give us these kinds of individuals, but the implications of the class of individuals that are being recruited often are omitted. For example, consider an article that is modeling a kind of reinforcement behavior using a neural net for this. What kind of "individuals"are in one set and what kind are in the other? Of course we know that in one set there are "behavioral" variables and in the other set there is the substitution of the neural units and connections for these individuals. attractors, local minima, and so forth can be considered as "individuals"of the second set and can be related to "individuals"of the first set, e.g. a quick reinforcement in the presence of food. What is not clear in terms of the epistemological implications of these sorts of models is that: a) if behaviorism is denied why do these authors still insist on modeling behavioral variables? b) if cognitive science obliges the use of "representation"in which sense is "representation" used here? c) if neural nets are supposed to resemble the Central Nervous System (CNS) why are they modeling behavioral variables? d) if the neural nets resemble the CNS are they trying to establish any sort of equivalence between old behavioral concepts and new cognitive tools?. No, these models are pearls of contra-sensical conceptual investigation because neural networks in this case are being used only as mathematical tools to treat invariant statistical properties from which we don't have full understanding of. They are not "cognitive"in this sense, in spite of appearing in "cognitive literature".

The breakdown of behaviorism and the style of input-output models and the growth of Cognitive Science may be attributed to three reasons:

a) quantum indeterminacy precludes a strict determinism (Heisenberg uncertainty principles),

b) a certain class of non-linearities, those that are sensitive to the initial conditions, preclude deterministic behavior ( in the sense of predictability) for a system

c) there is an impossibility of translation of mental idioms, when intentional , to physical ones. [4]

Due to the above three problems, not explicitly mentioned, and maybe sometimes not even perceived, the project of modeling inputs and outputs were doomed to failure and intermediate processes needed to be recruited. The problem began at this point because with the growth of symbolic architectures in the late 40's and 50's there was a strong appeal to treat the intermediate process as symbolic, representational, intentional object that would prevent the identification of cognition with the neural substratum. Neural networks, connectionist models came later and introduced a mess that sometimes is elliptical, called sub-symbolic: we are now a little bit closer to the CNS but not so close. This is wrong: whenever one accepts that cognition is representation in the intentional sense (here intentional is used in Brentano's sense), that cognition is made of propositional attitudes, one is always modeling mental individuals and mental relations (according to our classification of forms and contents). Even when one is modeling a complex problem of higher mental functions, if one adopts a representational intentional way of looking to the mental, the formalism that underlies may be a neural net, can have some relations to the CNS, but it will be never a model that relates the neural to the mental because the usage of intentional representations precludes this from the beginning.

To prevent these problems, trying to really model neural and mental individuals and rules together, there are three strategies:

a) to eliminate the mental, considering mental attributes and objects as non-natural types or categories ( eliminative materialism)

b) to abandon the substantive aspects of the computational and mathematical formalisms, considering them mere tools to handle data. (something that might be called Computational Cognitive Psychology and Computational Neuropsychology)

c) to reconsider the notion of intentionality and of representation: either by eliminating the notion of intentionality and using representation as only "re-presentation" something closer to the mathematical use of relation [7] or by dividing mental-cognitive phenomena that are intentional and those that are not . This last way may be done for a certain mental feel, e.g. pain, or can be done for a large class of mental-conscious experiences: dreams, that in our view are not intentional because they would lack control, volition, responsibility and immediate access to the language.

DYNAMICAL SYSTEMS AND CHAOS

Let us briefly consider the class of problems that were pointed out above about normality, psychoses and dreams. Assuming that there must be a difference between form and content. Now I shall deal with a certain class of formal tools: the dynamical systems that have sensibility to the initial conditions, hence presenting, for a certain class of parameters, unpredictability of the state the system will assume in the space of states (the space in which the variation of the states of the system along time are represented, given usually by differential equations-for continuous time systems- or by difference equations - for discrete time systems).

During dreams there are plausible sequences and often bizarre associations. During normal wakefulness there are plausible sequences and during psychoses, bizarre sequences. During dreams our CNS is not fed by environmental parameters, however, during wakefulness it is and that during psychoses there may be a bad functioning of the sensoring mechanism by which the environment "pushes" the system into coherence.

What class of suggestive formalism can be recruited to explain this behavior? I think that the Theory of Dynamical Systems and the appearance of chaos, a very charming subject in spite of being badly understood can bring on inspiration.

Deterministic chaos is the name given to certain systems that, for a certain class of parameters, behave in an unpredictable way. Taking, for example, a system composed of three bodies interacting with each other according to gravitational force, for certain parameter values, in certain initial conditions, there will be unpredictability of the next state of the system, in spite of the system being totally governed by deterministic laws. The appearance of unpredictability for a certain class of parameters, the so called bifurcations parameters value, allied to the non-linear and to the sensibility of the initial conditions, gives rise to a multitude of states the system can assume for "the same bifurcation parameter value". Despite being deterministic from the formal point of view (that is to say the equation that describes the evolution of the system) it is unpredictable from the content point of view (that is to say the states the system assumes in state space or the contents of the system).

Very briefly, this is why there may be determinism, nomicity, from the formal point of view, and unpredictability from the content point of view (only for certain branches of the system).

Supposing that we now have to deal with a system that is very complex, as the CNS is supposed to be. We know from the Mathematics that it is very unusual for a complex system to be linear. They tend to be non-linear, because this is the "formal " source of the systems' richness. From the physical point of view there are many people that tend to relate complexity to thermodynamic dissipation, and then to relate dissipation and entropy from the physical point of view, with non-linearity and sensibility to the initial conditions, from the mathematical point of view, to explain the emergence of complex and new patterns in nature. We are not going to touch this complicated issue of dissipation and entropy. Whatever the underlying physical phenomenon that leads to non-linearities may be, it is certain that from the mathematical point of view that non-linear systems are richer than linear ones. It is legitimate to assume that the CNS besides being non-linear in its dynamics (this we know because the simple generation of the action potential in a neuron is a matter of non-linear dynamics) can have a certain class of non-linearities that are sensible to the initial conditions (Walter Freeeman showed that there is chaos in the olfative cortex [6]. Then we can speculate that:

a) awake normal states are non-linear systems far from the bifurcation points.

b) dream states are non-linear sub-systems that tend more promptly to reach bifurcations and even chaotic behavior

c) psychotic states are awake states that are closer to bifurcations values.

What is important to say is that when one is dealing with non-linear systems, the bifurcation point is very sensitive. A small correction can push the system back to ordinary values, i.e to the equilibrium. This is why we may suppose, from the formal point of view that these systems are good metaphors to inspire our understanding of these three mental phenomena: normality, psychoses and dreams. Moreover, it is a source of inspiration because the same system can behave differently according to parameters. If during wakefulness one has an intense contact with the environment through the senses, one must hypothesize that the predictable sequence of contents that characterize wakefulness is a matter of external correction of the bifurcations the system tends to present. Why do we know the system tends to present bifurcations and chaos? Because during dreams, mental sequences are bizarre, and because the sensoring control, i.e the feeding of environment is abolished. Psychoses in this reasoning would be bifurcations during wakefulness that are not corrected by the environment, due to a bad-functioning of the system.

From the formal point of view it is licit to presume that the source of variability, of richness a system has is due to non-linearities. But non-linearities, when sensible to the initial conditions can have intervals that unpredictability and a multitude of states occur, violating determinism in its predictable axis. (Determinism has two axis: one that is due to the strict determination of the state of the system and of the law that governs the systems behavior, other that is due to the forecasting of the next state, given the present one)

Given that it is not a bad speculation that rich behavior needs non-linearities to have multiple states, and that coherence seems to be something tied to determinism, i.e to nomicity and predictability. This is why we propose that normal wakefulness must be robust, i.e. it must be sensored by the environment preventing the system to reach bifurcations and even chaos. Dreams instead, when the system is not fed by environmental parameters present a big tendency to reach bifurcations and chaos. From the formal point of view this is what would explain that the same cognitive system, when awake is very coherent and when dreaming is very bizarre.

MENTAL CONTENTS

The problem of the contents or the individuals that will substitute state-variables in a dynamical system, for example that one that describes higher mental functions, is a matter of the blocks the culture and the ordinary language furnishes to the subject to name its internal states. Trying to find natural types in the contents of mental life is as difficult as trying to find natural types in culture and in language. Language, culture and mental contents are all the same blocks.

Choosing the right formal way that describes a system is one thing. Choosing the class of individual that can be substituted in its bound-variables, or in its state-variables is another thing.

The most important thing that appears in this sort of reflection is that the system that allows the emergence of phenomenological mental experiences is not tied to awake states. It appears during dreams. But what is absent during dreams is exactly the voluntary control, the coherence and the "determination of the next states" that tend to be felt as "freedom". We suggest that these entities, volition, control, "freedom"and responsibility are the result of the formal liaison of the system to the environment, instead of being internal agencies. When the cognitive system is turned-off from the senses, it presents a full amount of bizarreness that can be, metaphorically and speculatively, described as being chaotic intervals in the dynamical system.

When we say that dreams are the real road to understanding cognitive phenomena we are saying that:

1) one has to consider mental events

2) one has to consider conscious events

3) conscious events appear during wakefulness, dreams and psychoses

4) conscious events are characterized by phenomenological aspects and by control, volition, coherence, responsibility

5) representation as intentionality appears in language related states, i.e. wafulness.

6) dreaming states lack intentionality, control, volition, etc

7) then phenomenological experiences need not to be intentional

8) control, volition, freedom, and the linguistic usage of intentionality are not exclusive and absolute predicates of the mental, because they are absent during REM sleep

9) Finally: control, volition and freedom, as coherence and pragmatical properness are marks of the environment adjusting the system all the time to prevent bifurcations and chaos.

10) To build a higher mental cognitive system would be to allow the system to have bifurcations and chaos when turned-off from the environment, and not to have them when turned-on

11) To build a model of psychoses is to search for a system that has a transient bad functioning of one or more of its parts even during the feeding sensoring of the environment.

CONCLUSION

There is a big debate about the function of dreams in a cognitive system. One model, based upon neural networks suggests that the function would be to erase "bad memories". Another based on more neurophysiological data proposes that it is a matter of fixation of memories.

I think whatever there is in the function, erasing or fixing, dreams are from the formal point of view the purest cognitive phenomena; they present the full range of states, some coherent some bizarre. It is like the battle of a natural system that resists tailoring according the rules of culture and of language. These rules, in spite of adaptive, are not natural, they are contingent and this could be why the system resists and is so unstable from the structural point of view, when dealing with these supposed to be "coherent contents". During normal wakefulness, however, the system must behave according to the environment, and this is why it so intensively monitored preventing bifurcations and chaos to happen.

Volition, "freedom", beliefs (as an intentional state), responsibility and control are a priori capabilities of the system, closely tied to its structure. But the way they shall appear, and even its triggering appearance is due to the external variables. Were we to be turned-off from the world we would not be free, purposeful, controlled and responsible. Intentionality, control and properness, maybe are due to the context and not to an internal agency of the system, but due to its a priori structure.

Eliminating intentional idioms from the dream states we can adopt neural-cognitive models that dreams to be fully reductive. Cognition projected to the neural substratum might be better analysed considering dreams as the purest form of cognition, be they the road to the unconcious, or only the remodeling of the conceptual framework that underlies the human mind.

REFERENCES

[1] Flanagan,Owen Jr. (1986) The Science of the Mind. MIT Press.

[2] Fodor,Jerry (1975) The Language of Thought. Harvard University Press

[3] Hooker, Clifford (1981) "Towards a General Theory of Reduction" in Dialogue vol.XX 1-3

[4] Del Nero,Henrique (1993) "Do Behaviorismo às Redes Neurais" in Coleção Documentos IEA-USP, série Ciência Cognitiva.

[5] Pribam, Karl (1991) Brain and Perception. Lawrence Erlabum Associates.

[6] Skarda,C. and Freeman,W. (1987) "How brains make chaos in order to make sense of the world" in Brain and Behavioral Science 10,161-195