CONSCIOUS CONTROL:

FROM FREUD TO THE FRONTAL LOBE

Henrique Schützer Del Nero

Lucia Maria Argollo Maciel

Alfredo Portinari Maranca

José Roberto Castilho Piqueira

ABSTRACT

By the end of the last century, when Freud wrote his Project of a Scientific Psychology, the epistemic battle between reductionistic and emergent models of mind were beginning. In this abandoned text, that was published posthumously, the author tries to give a quantitative description, with neurological plausibility, regarding the knowledge of that time, of the processes that might underlie psychic function and dysfunction. Inserted in the flavor of his first topography, in which Freud states that there are three classes of states: conscious, pre-conscious and unconscious, the Project proposes a model based upon quantities of energy that flow through three main types of neurons. The interplay between perception, memory and consciousness, in spite of the highly and dated metaphorical aspect of the work, are still plausible, mainly if one looks to biochemical aspects of hebbian memories, to neuropsychologycal aspects of the relation between the hippocampus and the frontal lobes functions and to the tentative models that treat neurons as oscillators, extracting from dynamical systems analysis of their connections a rich source of lessons that must enlighten the process by which conscious control may appear as an adaptive interpreter of states in the state-space of a very complex non-linear system that uses bifurcations as a source of topological variability, linearizing automatic control and partitioning endeavors through non-linearities and maybe chaos. Artificial models of mind will be incomplete if they don't face the problem of voluntary-conscious control.

I. Freud, Neural Networks and Reduction

Our goal here is not to stress Freud's work, but only to use it as a source for an old dilemma that still pervades the cognitive sciences: the role of consciousness and of volition-control of behavior and mentation (manipulation of "mental representations").

Freud's project [1] is an essay written at the end of the century, trying to launch the metaphorical basis for a scientific psychology based upon neural processes. It postulates that there are three types of neurons in the Central Nervous System, connected or not by contact-barriers (an equivalent to what later would be qualified as the strength of the synapse). These neurons are: j-neurons tied to perception, receiving a quantity Q of energy or information from the environment; y-neurons that store memories and increase or decrease their connectivity through experience and w-neurons that are tied to consciousness. The scheme works as follows: Environmental information considered as quantity Q of energy reaches the system impressing j-neurons. They retransmit these bits of information to y-neurons. There, memories happen, through modification of the strength of connections and impressed by another quantity Q' that is information that comes from the body (proprioception and pulsions). Finally, this n-y informations plus Q' reach w-neurons that are tied to conscious experiences.

The second Freud, or his second topography moves in a more functionalistic-emergent vein, proposing three classes, ego, id and super-ego, as the loci of mental functions and its neurotic deviances (Freud worked very litlle with psychotic deviances, maybe reflecting a good intuition of the limits of psychoanalytical theory). According to the tradition of Frege and of the developments of logic in the XX century, mental functions came to be considered as emergent predicates of neural implementations. Mental blocks were to be considered as primitives over which rules apply, creating strings of sentences or thoughts. The models of traditional Artificial Intelligence after Turing, Newell, Simon, Marr among others, followed this prescription: there is not a perfect or complete translation from a mental idiom unto a neural one, hence the level of algorithm and of computation must be dissociated from the level of implementation. This is what is called functionalistic-emergent tradition, endorsing projects of traditional AI and capturing the very essence of the major part of Freud's work on mental functions.[2]

Neural Networks recently became amovement which has surrendered towards putting neurological data among the elements that must be considered if one expects to understand cognition. Initially inspired in a very restricted metaphor of connections, neural nets quickly evolved to a clear separation: a) pragamatically inspired networks, with refined mathematics beneath them, became a powerful tool to manipulate a certain class of theoretical and practical problems where data are known and rules are shadowed. These networks, in spite of the historical commitment to neural-inspiration, may have slim CNS (Central Nervous System) plausibility. b) neuronal networks that try to capture data that come from the neuronal research: these networks are more committed to giving a plausible explanation of what goes on in the CNS, but the problem of what level of the CNS is being considered still remains. If neural nets considered mental representations and interpret nodes and connections with mental categories, they are so functionalistic as traditional AI are, with the exception of the nature of the computations they perform. Briefly, there are two classes of problems involved when one deals with the dichotomy reduction x emergence regarding the relation between mind and brain:

1. There are semantical aspects of the projectability (in Goodman's terms [3]) of mental blocks unto physical blocks.

2. There are syntactical, i.e. computational-algorithmic aspects of projectability. Then, one must build the following statements:

* Traditional AI (logical rules of computation) + mental primitives = emergence of the mental regarding the physical (symbolic paradigm - Fodor, Pylyshin, Marr)

* Neural Networks (differential equations, tensors, etc) + mental primitives = emergence of the mental regarding the physical (sub-symbolic paradigm - Rumelhart, Smolensnky)

* Neural Networks + neural signals = reduction of the mental to the physical (anti-representationalism- Freeman, Pribam, Posner)

From the above said one must see that Neural Networks are not a sufficient condition to consider the mental to be reduceable, or translated, to the physical. Sometimes they may be a necessary condition, considering that the analogy is rich enough to allow some interesting results regarding the presumable way the CNS manipulates information.

What comes to the arena, recalling Freud's Project, is that there are really two main styles of computation: traditional AI or connectionist AI (neural nets). Freud pursues the second within the Project and the first in his second topography. But is the style of the Project really reductionist in the sense above said? Or it is only a kind of more intimate relation between mind and brain, in spite of the emergence of the semantical categories? One may postulate hysteria as primitive, or as the name for a basin of attraction. Both, however, are emergent regarding the semantical categories, in spite of being different regarding the style of computation.

Models can pursuit reduction from the syntactical point of view and from the semantical point of view. Syntactically speaking neural nets and Freud's Project are reduction-driven. Semantically speaking they're not, as far as taking advantage of categories that are assumed to exist-- i.e. mental categories (thoughts, feelings, goals, intentions, beliefs, etc). Neural networks represent a serious attempt to face the problem of grasping the so called brain-style of computation, but this is only half of the job. If one doesn't look at the astonishing difference between voluntary-conscious modes of operation of the CNS, one always deals with "automatic" aspects of control and of computation. The crucial aspect of the models must cope with the problem of consciousness and of volition, proposing metaphors for it. In this sense the Project touches an important point: in spite of being emergent, in our definition, from the semantical point of view, Freud sees the importance of putting consciousness in the arena, maybe as an inspiration that any model of mind that doesn't pay attention to it, even being reductionist and quantitative, will soon be doomed to failure.

II. Consciousness, volition and semantical reduction

The way to cope with the problem of models that are only reduction-driven in the syntactical vein, as Freud's Project and the major part of neural networks, is:

a) to adopt an antirepresentationalistic approach

b) to consider the dichotomy conscious x automatic process

c) to consider voluntary control as a mark of cognitive systems able to display consciousness

d) to consider conscious, hence voluntary, control as a problem of neural structure and function and not as a mark of "intentionality" in Brentano's sense, what should imply impossibility of reduction using laws and bridge-principles that were nomological, too.[4]

e) to propose mathematical tools that can enlighten the style of brain-computation that allows classes of different phenomena to exist: automatic and conscious-voluntary ones.

f) to understand that syntactical reduction is tied to the nature of the algorithm and semantical reduction is tied to the nature of the way one considers codification to hold in the CNS.

a. Antirepresentationalism

There are clearly two meanings in the term representation when one deals with cognitive architectures.

* Representation₁ can be consider as intentionality, i.e. the mode certains operators, like beliefs, fears, hopes, etc fulfill with real or imaginary objects, sentences like: "Paul fears that....". Intentionality is the very mark of the mental according to many authors and mainly those that use Traditional Artificial Intelligence to model the mind.

* Representation₂ can be considered as a kind of map that connects external stimuli and internal equivalents. This concept is more tied to the neural maps of function, closer to the way neuropsychology works and recalls the mathematical notion of domain and image of a function.

Representation 1 is not compatible with the project of a semantical reduction, contrary to representation 2 that is neutral, not having any peculiarity that precludes this kind of reduction to hold. Anti-representationalism must be considered against type 1. This is a very common misunderstanding in the texts and articles, maybe due to the fact that representation 1 is the unique meaning for representation.

b. Conscious x automatic process

Consciousness has been considered a taboo among a lot of serious researchers : either it means a pseudo-concept or it means a distant attribute of the mental that has to wait till more data is known. In our view, this is a misjudgement of the importance of this entity, or class of processes, or states: if one doesn't face the problem of conscious control, one doesn't grasp the most crucial point of cognitive architectures, and of the semantical relevance of the brain structure to the mental function.

Consciousness has to be divided in two different aspects:

a) conscious phenomenal experience: this is a mark of a very inner experience, objective regarding the subject, but always subjective regarding reports. It is impossible to treat this inner-experience as an object of science, at least of hard-formal science [5]

b) conscious control: this a mark of a kind of attitudes that can be overtly considered as a different way of processing information, opposed to automatic- unconscious control. Conscious control can be regarded as a scientific object , being able to occur in third-person statements with public verifiability.

Some authors state that consciousness is the phenomenal, first-person's experience (e.g. Searle in [5]). We don't deny that these kinds of experiences exist, but we have serious doubts, as Searle, if there is a way to build a science of these inner-private experiences. Science is a way to write declarative and normative sentences that can be publicly verified or refuted. Inner-private experiences can become public throughout the meanings of language. If this happens to work regarding everyday problems with a degree of accuracy compatible with common-sense reasoning and knowledge, this doesn't imply that this kind of reasoning and knowledge can, or must, be transferred to the scientific style of stating things. One of the mistakes of current programs is the effort to capture the process by which common-sensical reasoning works, as if it were a well-succeeded strategy. It is well succeeded in terms of predictions of a very restricted amplitude. The more sophisticated knowledge gets, the more formal tools are recruited, the more counter-intuitive statements are set, the more resistance or ignorance common people show regarding the issue. If one accepts a theory of motion based upon the Aristotelian impetus one is prepared to do research on a phenomenal common-sensical approach of consciousness. If, otherwise, one needs something less impregnated of anthropomorphism, then consciousness, despite the presumed depth and certainty of your experience now while reading this page, should be considered as one possible mode cognitive architecture deals with inputs and outputs taking advantage of two ways of processing internal information: an automatic one and a conscious one. Consider, for example, driving a car: the first steps of learning are very hard, slow and conscious. The well-learned task, except for risky moments, becomes more and more automatic. In this sense, automatic (hence not-conscious, well learned, pre-wired or well-represented) means a kind of optimal solution for a problem that has, perhaps, two sets of variables that show bijective and inversible functions among their elements. The richer the task the more partitions have to be built in terms of getting optimal relations. For simple, trivial and old domains, it is better to have a way of processing information quickly and in an optimal way. Creativity and judgment are not problems of optimality. They don't have to be solved in seconds, but they have to be well solved, or at least in a way apt to be justified.

Conscious control, contrary to the shadows that preclude phenomenal experience to be an object of a scientific inquiry of cognitive systems, means a way of manipulating information in a highly complicated way, semantically loaded and slowly computed. The interplay between syntactical automatic memories in the hippocampus and semantically-conscious frontal lobes must enlighten these differences. This difference must be in the core of a renewed Project, where neuropsychologycal aspects play a role, as old inpirations did when Freud wrote and shelved his suggestions.

In a recent work Moscovitch and Umilta revisited some concepts and proposed a model of a highly modular system that processes low-level information (in our terms, syntactically loaded) in the hippocampus through working-memories and that recruits frontal lobes as a source of interpretation of these shallow-outputs to create a goal-purposeful device. Information in the frontal lobes would be more semantically loaded, interpreted, and more tied to consciousness [6].

In another classical series of experiments, Libet showed a slow negative potential that precedes "voluntary action before a subject mentally decides that he intends to make a movement."[7]. These experiments, showed, in our opinion, that there is a preliminary event of conscious control and then a phenomenal experience. In spite of looking counter-sensical, what defines conscious control is not ipso factu the subject's report of conscious awareness (since we put in doubt the publicly objective status of this statement) but the style, function and structures involved in computing as such, and not in the automatic mode.

Voluntary control, consciousness and semantics are to be considered in a solid block of relations, regarding the type of structures that are involved in some complex tasks, to launch good metaphors that inspire what we call semantical reduction.

c. Voluntary control

What may characterize conscious control is the recruitment of large areas of the neocortex mainly located in the frontal lobes and to have a highly interpreted, hence semantically loaded, way of handling information. The underlying mechanisms that must be in the core of these processes are biochemical changes in the rate of quantal liberation of vesicles of neurotransmitter at the synaptic clefts, as proposed by Eccles [8], a capacity that would be evolutionary tied to the new areas of the brain, and that occur in a non-causal domain. Measures of information in this case would require an apparatus that is closer to the quantum formalisms, being the measure of probability a Gabor function and not a Shannon one.[9] Classical ways of seeing probabilities at the neuronal level as they proposed by Freeman through the tools of the Theory of Dynamical Systems, or Quantum Fields, as proposed by Pribam, using dipole oscillations at the dendritic branches, are two ways of searching for the low-level mechanisms that subsume the voluntary-conscious mode of computation. In both approaches the basic formalism that lies in beneath is the synchronism of oscillations: in Pribam's works there is an oscillatory field of dipole molecules, in Freeman's work there is a classical transformation of pulse at the neuronal level to frequencies at the assemblies level, letting non-linearities to produce bifurcations and chaos as a source of novelty, learning and categorization. Each bifurcation would constitute a branch in a strange Rossler's attractor and that would be the core mechanism of high-level processing. [10]

Volitions in this functional view could be the process that occurs in new structures of the brain, in a complex way that is semantically loaded, since it manipulates shallow outputs from the hippocampus and from other areas. Of course, in this sense, voluntary control is a concept firmly tied to consciousness. Freud's work induces a misjudgment of the term volition, because unconscious voluntary wishes are assumed to exist. In the phenomenal aspect of consciousness this may be true, but in the computational-control inspired aspect it is not possible.

To be conscious it has to be potentially voluntary-driven. To be voluntary- driven means to be highly interpreted (semantically loaded). To be semantically loaded means to recruit large areas of neocortex and to be computationally slow and sparse, but creative in terms of rearranging representations (in the sense 2 above defined). Maybe the process by which some aversive memories allow avoidance subsumes the best example of conscious computation without consciousness awareness. But to accept that the psychoanalytical work through memories, recollection and interpretation can allow relief of neurotic symptoms is a tacit acceptance that one thing is the phenomenal-conscious experience and other is the highly interpreted and sparsed conscious encoding. In this sense, Freud could be correct had he stated the clear separation, but of course the nature of volition in the control-computational view (maybe as non-causal in the "quantum interpretation", or not-predicatable in the dynamical systems interpretation) is far from the deterministic aspect of the will of Freud's unconscious. Freud could be redeemed regarding the misjudgment of the very status of consciousness in a scientific project, but the status of volitions as purposeful must be rechecked in order to adapt recent and old metaphors about the psychic system.

III. Outline of a semantical reduction throughout formalisms and brain mechanisms.

Genuine reduction from the semantical point of view would be a kind of constraint the brain level imposes over the cognitive level. Neural networks are the first step regarding the brain-style of computation. But are they the last model? Moreover, are there models that can have the virtues of being syntactically (level of algorithm) and semantically (level of codification) reduction-driven?

Neurons are oscillators in the Hodking-Huxley models: be it in the classical transformations that allow to consider assembles of neurons governed by van der Pol equations, be it in the "quantum field" approach, there are some fundamental facts:

a) frequencies and time are essential to understanding dynamical computation and codification in the Central Nervous System

b) learning, conscioussness and the will are evolutionary gains that allow the system to program itself in different environments and in on-line situations. These evolutionary traits are essential to cope with complex situations, but pay the cost of lacking robustness and globality of solutions.

Then: one must suggest that more primitive systems and automatic computations are closer to linear operations and linear codes. This guarantees invariance of solutions and quick responses to environmental stimuli. But, novelty, learning and creativity are all handled by systems that escape in certain structures of the well-behaved style of linear systems and codes. The neocortex structures, both in the macro as in the micro function, may display the complexity of behavior a conscious-voluntary controller needs. There the space is partitioned according to the underlying physical-chemical reactions (e.g. the rate of change of quantal liberation of vesicles in the synapse, the control of memories through long term potentials and its pre-synaptic control through CO and NO, the slow oscillation that pervades the cortex, etc) described by complicated mathematical structures. Non-linearities with central manifolds that indicate to non-linearizable systems (or quantum described systems with uncertainty aggregated, or dissipative systems far from the equilibrium point), are possible candidates to explain a real dynamical system, that behaves linearly where representations 2 are well formed, allowing speed and certainty to be an adaptive trait and behaves non-linearly when representations must be sought, rearranging the elements in new domains and with new relations. Bifurcations in the classical realm must be the mathematical expression of the most powerful structures of CNS, the Nature carefully selected allowing consciousness to appear: the possibility that quantitative variation of parameters allows dramatic change on the qualitative domain. Topologically, bifurcations are the source of qualitative variations in the space of states. One must not confuse the level of iteration in an equation and its phase portrait, power spectrum, flow, etc. The former deals with an interval of numbers that feed an equation allowing other numbers to appear as results. The latter are qualitative partitions in the topological space where things may look alike or not, but where the concept that governs similarity and difference is that of homeomorphism. Given this brief concepts of the Theory of Dynamical Systems, one may suggest that:

1. Linear systems don't require an excessive amount of semantics.

2. Non-linear systems are the source of conscious control.

3. From the Neuropsychological data, one may risk saying that information is iterated in a syntanctical way in the hippocampus and other functionally equivalent places. Whenever a bifurcation occurs, changing abruptly the topological behavior of the space of states, another system of neurons tries to grasp the variation from the topological point of view, and not the process by which this variation was created. This implies creating an interpretation that names each new topological branch of the space, handling it with simpler structures from the computational and encoding point of view. This may be the source of the progressive appearance of conscious control, and of the apparent distance, henceforth, "emergence", of the level of symbol manipulation and the level of implementation.

Shortly, a genuine model of semantical and syntactical reduction would have to:

4. have a neural networks style of computation but with genuine dynamics, maybe as coupled oscillators, or through the formalisms of solitons, etc

5. have a neural codification that grasps the richness of non-linearities or of other mathematical tools of description (non-linear codes, and maybe codes closer to Gabor table of frequencies versus time)

6. have a way to compare the supposed primitives of logically inspired architectures of cognition, intelligence and emotion and the multiple spaces build through the labor of non-linearities, dissipation, uncertainty, etc, that might be the essence of voluntary-control: the mark of cognition, of Culture, of mankind and of a legitimate science of the mind.

REFERENCES

[1] Freud,S. (1953-74) The Standard Edition of the Complete Psychologycal Works of Sigmund Freud. Ed. J.Strachey London: Hogarth

[2] Flanagan, O. (1984) The Science of the Mind. MIT Press

[3] Goodman,N. (1951) The Structure of Appearance. The Bobbs-Merril Company,Inc.

[4] Fodor,J. (1975) The Language of Thought. Harvard University Press

[5] Searle,J. (1992) The Rediscovery of the Mind. MIT Press.

[6] Moscovitch,M. and Umilta,C. (1991) "Conscious and Nonconscious Aspects of Memory: a Neuropsychological Framework of Modules and Central Systems" in Lister,R. and Weigartner,H. (ed) Perspectives on Cognitive Neuroscience. Oxford University Press.

[7] Libet.B. et al. (1983) "Time of conscious intention to act in reaction to onset of cerebral activity(readiness-potential).The unconscious iniatation of a freely voluntary act"in Brain 106

[8] Eccles,J. (1993) "Evolution of Complexity of the Brain with the Emergence of Consciousness" in Pribam, K. (ed) Rethinking Neural Networks: Quantum Fields and Biological Data. INNS Press. Lawrence Erlbaum Asssociates Publishers

[9] Gabor, D. (1946) "Theory of communication". Journal Of The Institution of Electrical Engineers, III, 93: 429-457

[10] Freeman, W. (1992) "Tutorial on Neurobiology: From single neurons to brain chaos" in International Journal of Bifurcation and Chaos, vol.2, No.3