looking for a legitimate temporal code in the brain

Searching for a general mental code may bee seen as a theoretical endeavor that simplifies a large amount of what it is known about specialized functions and regions of the human brain. The aim of this article is to scaffold a deterministic, non-mechanic, framework of computation, that due to four conditions: analogical processing, the emergence of language, the increase in cerebral tissue (particularly of neocortex) and ethics allow mind, and hence consciousness, to appear in its full totality in the human brain.

Several mistakes appear to have happened during the last centuries in the approaches to mind-brain relationship. Categorical and semantical mistakes are the most prominent and the irreducibility of the mental to the physical pays a lot of dues to the propositional nature of mental events.

Propositions are well-constructed sentences according to formal-linguistic rules that allow the formation of non-countable sentences with a finite set of rules and with a finite set of primitives. Building well-constructed sentences is an ability of linguistic-apt devices that has no relation with the empirical verification of their truth or falsehood. Propositions, being true or false, are defined according to non-empirical criteria of verification. One can verify the very nature of a proposition without being committed to its truth. One distinguishes two series regarding propositions: propositions and non-propositions. Among propositions one distinguishes true propositions and false propositions (that according to some theoreticians, due to the fact of representing formal well-constructed sentences, could be seen as talking of "possible worlds").

We consider first, that all mental events are conscious or are able to be summoned by consciousness. Consciousness would be the stage in a metaphorical theater where mental life goes on, having different roles that might be played in this stage (the mental functions like memory, perception, language, thinking, the will, emotions, etc) and different performances, at each moment, to each mental-function-role (contents of consciousness in their specific actualization – a performance on memory, a performance on will, etc.). There is a light that focuses each part of the scene – attention - the very counterpart of consciousness. Consciousness is the place where mental functions are actualized in mental contents and attention is the process by which each of them are focused at each moment.

Mental events are the result of a certain style of cerebral processing in a way that one can affirm that every mental event is a cerebral event, but not all cerebral events are mental events. Consciousness cannot be seen as an epiphenomenal description of what is going on inside our brains. It has to have a function in the cerebral economy, and we are going to do a pilgrimage over the hypothesis that consciousness is a "user friendly" device that allows the subject to control, reset, learn and train certain non-convergent, unstable, undecidable or risky events that are not able to be handled by the cerebral-automatic mode.

Dealing with cerebral-mental dilemmas is a very complicated issue regarding the variety of different fields and theoretical methods to approach the problem. We are going to choose three models of description that might allow our theoretical rerouting strategy to become more plausible.

Three modes of description to mental processing and their respective field can be presented. The mind can be seen as symbol process of manipulation that takes formal processes of connecting symbols building inferential necessary chains of reasoning. Cognitive Science pays a lot of tribute to this conceptions and Symbolic Artificial Intelligence (SAI) works with these assumptions (1). The mind, and hence consciousness in a certain aspect, can bee seen as representational apparatus that manipulates symbols according to formal rules. Symbols-propositions-representations are the very primitive core of the block over which mental-formal-logical rules will connect entities, in order to build reasoning and intelligent behavior.

A second way to describe consciousness and mental events is a semi-dynamical way that tries to mimic the way neurons perform computations. Connectionist Artificial Intelligence (CAI or neural networks) is one of the fields that acquired importance in a non-logical, non-translatable to rules, way of modeling mental processes (2). In spite of the fact that there is a debate about the antirepresentationalistic and associationistic nature of neural networks, it is still the same set of primitives - symbols – that are used to interpret input and output vectors that feed these nets. The status of dynamics in neural networks is not fully due, because there is a certain degree of dynamics during the training mode, but, as soon as convergence is reached, the net only matches examples with a local "representation" (a process of statistic partitioning of composed functions).

A third way to describe the mind is as a decision-making device, dividing the mind in functional modules of semi-encapsuled-specialized information-processors (and the weak correlation between these functional mental-modules with cerebral structures – the very aim of Cognitive Neuroscience), that operates under risk, manipulating scenarios and controlling behavior in order to solve problems that are non-trivial and that can not be pre-wired as reflexes or automatisms in the central nervous system. Many ethical dilemmas, besides other different classes of problems, fall in this definition of mental processing. Cognitive Science, be it Symbolic (SAI) or Connectionist (CAI), try to mimic this ability of the mind. A problem that impresses all of us is the fact that decisions under risk are not fully constraint by normative decision-making theory but follows a specific, apparently non-logical, human style of decision (descriptive decision-making theory).(3) We do not satisfy utility functions during decision making, and usually we violate probabilistic rules during mental-conscious processing of decision.

These three ways to describe the emergence of mental-conscious powers in the human brain, defining, successively, the mental or the conscious as the place where symbols-representations-propositions are manipulated according to logical rules or the place where a dynamical architecture approximates the local solution (convergence) to a composed function or the way we manipulate risky decision making scenarios have something in common, despite the fact that each of three have their problems.

Modeling the mind as a symbol manipulator have two strong consequences: 1) it implies the dissociation between computational/algorithmic processes (the mental realm) and implementational strategies (the hardware-cerebral level); 2) it implies that the mind could be mimic by an Universal Turing Machine, but due to gödelian inconsistencies and incompleteness, it has to solve the dilemma that symbolic-algorithmic architectures show the Halting Problem and that consciousness does not show it (what has led many theoreticians to propose a quantum mechanics, non-algorithmic, version to mental-conscious processes). (4)

Modeling the mind as a neural network has three problems: 1) the entities that interpret the input and the output vectors are still considered as given primitives, focusing the very nature of the mental in the process by which one connects entities and not in the very nature of these entities (cats, triangles democracy, anxiety, free will, etc); 2) the dynamics of neural networks is still very poorly dynamical, starting from an oversimplified conception of neurons and synapses; 3) the quantity of different architectures that allow the same approximation of a composed-function problem suggests a kind of tabula rasa empiricism that is incompatible with a rationalistic, a priori, view of the central nervous system and the active processes by which we build our expectations of the world, being only corrected and not governed by the senses. (5)

Modeling the mind as a problem-solving or decision-making under risk device offers a lot of trouble because: 1) every decision-making device must have a certain degree of "conscious control" (as a type of control) and representation of the problem-space but not every decision-making device has to be conscious; 2) performing decisions has to be seen twofold; the process by each one decides and the process through which a subject subsumes decision, what implies a certain self-representation and a language with which one manipulates the decision (that is captured in part by protocol analysis in Cognitive Psychology); 3) not all conscious events imply directly or explicitly a decision- making strategy.

That’s to say that there is, in our opinion, a certain equivalence between the three series (SAI, CAI and decision-making), and there might be a strategy that surpasses the above listed troubles.

Consciousness has been seen as a property that might emerge, from the cerebral point of view, due to a large phenomenon of synchronization of various cerebral areas. The 40 Hz hypothesis to the processes by which the brain summons neuronal assemblies to allow consciousness is a very problematic concept.(6)

The real dynamics that mediates cerebral processes has many different levels of analysis. At the neuronal level, one can see oscillations in different frequency domains due to series of spikes (action potentials) that reflect the discharge of the neuron to a given supra-threshold stimulation. At the level of assemblies of neurons one can see oscillations in another frequency domain that are captured through different methods including local field potentials (LFP), multiunit activity (MUA), etc. Large amounts of cerebral tissue may show oscillations in other frequency domains, captured through different electrophysiological methods (EEG, EP, MEG), etc.

The concept that plays a very important role in Neurophysiology and in cerebral circuitry is a localizationistic one. The Barlow concept (7) of receptive field reflects the assumption that in different scales one can find in the space domain a certain degree of place-coding, being the frequency a measure of the intensity of the pair stimulus-prototipicity. The hebbian concept of pairing assemblies and neurons, largely applied in the neural networks models, assumes a constant that increases or decreases the communication of two different sources of information, be they two neurons or two assemblies of neurons.

Synchronization among assemblies of neurons and over larger areas can be grasped through correlation methods. It has been seen that oscillatory/synchronized modes can underlie different functions in the brain: 1) consciousness; (8) 2) binding of different aspects of perception (bottom-up modes); 3) attention (top-down modes) (9); 4) time-windows (fusion and order thresholds for different modalities of discrimination of perception) (10); 5) memory; 6) gating in the hippocampus; 7) matching in the hippocampus; 8) voluntary control over behavior.(11)

The 40-Hz gamma-spindle, the mostly commented mechanism that might underlie these processes of rhytmicity in the brain, has to be seen with caution regarding the search for a legitimate temporal code. Real dynamics must be antilocalizationist in its very nature and the phrenology that still holds in several models that take use of the 40-Hz mechanism is incompatible with a real temporal code that uses oscillations and synchronism as a tool that enables time to be a supplementary vector code.

Localization of functions in the brain is a strong tradition in different fields of Neuroscience. The current image techniques (PETscan, fNMR) have tried to catch preferential loci of functions, despite the fact that, in our view, a legitimate temporal code would render impossible the pair function-place (phrenology) for mental functions, mainly for consciousness, the prime mental function. Genuine temporal codes render, in our view, the brain as an evanescent omniconnected lattice that recruits committees to deliberate and control unstable situations (new, risky, responsibility-laden or to be learned situations), the very core of mental activity, opposed to cerebral, phrenologically-ladden, automatic strategies of coding and control. (12, 13, 14)

If one assumes that objects are given primitives in a cerebral or mental domain and that mental processes rest in the process that mediates the computation over these objects, one has to surpass the problem of relating mental objects to cerebral objects. No one thinks that there is a way to translate single mental objects (cats, triangles, democracy, anxiety, free will, etc) to neuronal states. But in the same way the research rests largely in a concept that each module of the brain, be it a neuron, be it a larger assembly, represents something and the frequency of discharge – the substratum of temporal codes – only measures a certain degree of intensity of that object (in our view, a pseudo-temporal code concept, phrenologically-laden).

Imagine a neuron – suppose a sensory neuron in the primary cortex – and try to see that there might be an object or a predicate that neurons represent and that the frequency of spikes might give us a supplementary information besides fidedignity (the Barlow, localizationistic, rate-coding concept). The frequency domain, mainly in the interspike interval (7,13) might give us functional information of the mode of control – conscious or automatic – that needs to be recruited. Interspike intervals in the train of action potential carry a lot of information and a first order statistics that only grasps the mean frequency (rate- coding) doesn’t grasp the richness that might exist in the irregular interspike interval (7). Stochastic explanations assume that these intervals occur at random, but at the same time, chance can be seen as a reflex of two different phenomena: a) real chance that only states that the system is non-deterministic; b) presumed chance, that reflects a deterministic system discharging in a chaotic mode, what leads us seeing the temporal series as random, in spite of the order and determination that exist underneath.

If one considers the temporal domain as a source of random behavior it is very difficult to search laws that mediate the conversion from the signal domain to the semantical, rule-governed, linguistic domain that characterizes our conscious-mental production of propositions (be they explicit or be they translatable to propositions to express themselves - what we call implicit propositional characteristic of the mental). Neural objects are unknown and the only representation of these objects we have remain in certain primary sensorial areas that respond preferentially to certain features. Besides these primary sensorically localized objects, we have only indirect methods largely tied to Neuropsychology – lesion methods and image methods - that identify pairs of cerebral regions and functions/objects being processed.

If one assumes that objects exist in a certain local, the temporal domain that mediates the connection between these objects exerts only the role of a channel that transmits information between two traditional, symbolic sources. The 40-Hz hypothesis, and all other hypothesis regarding temporal codes, can bee seen in two different aspects: two symbolic sources use a channel – a carrying wave – to transmit information in a way that resembles a modem that connects two serial, digital, traditional, von Neumannn computers, where symbols are manipulated through logical rules (the core concept of SAI). There is nothing new using a conception of synchronizations if the carrying wave transmits symbolic entities. The question remains about the very substratum of these symbols in the brain. (Apart from the fact that 40 Hz as a carrier frequency is very poor, from the information point of view, what would impinge a large scale of parallelism to get some relevance in terms of the ratio of information that can be transmitted in this frequency).

A second consideration regarding temporal codes, e.g. the 40-Hz hypothesis, can be that neurons and assemblies would not gain anything if they represent objects in the frequency domain, because the localizationistic approach would be the same. Suppose that a neuron, instead of representing a cat, represents mammals. In the frequency domain, one might try to find other frequencies that represent types of mammals, including a certain frequency that would represent cats. It is absurd because it is still a kind of localization, an assembly representation that, instead of representing objects, represent classes, being the temporal code a kind of intraclasses identification process. It is still something, be it an object or a class, that is represented in that cerebral place that is discharging in a certain frequency (or in a multitude of frequencies). Barlow and Hebb mechanisms still hold in respect to the representation of classes (an assembly represents a class, and a certain constant of increase or decrease relating assemblies, represents a relation intra or interclasses) and the temporal code is false because frequency is only a carrier wave between digital-simbolic structures. That is to say that 40-Hz and other oscillatory/synchronized events in the brain are not signs of a temporal code, if we assume localization in terms of symbols that use oscillatory waves to communicate two different sources, or if we assume that frequencies codify intraclasses operations (being the class still localized in a certain region of the brain).

Objects as functions: the temporal code as source of controllability for consciousness (12,13,14)

If there is a certain degree of localization of objects in the brain – be this localization static or transient – the real temporal-antilocalizationist domain must have another function. Objects might not recruit the legitimate temporal code to be represented (what does not preclude a certain frequency measure of fidedignity regarding primary pair stimulus-prototipicity), but their function in terms of the mode of control that must be recruited to manipulate them –- conscious-voluntary or cerebral-automatic -- might use the temporal domain (second-order statistics of the interspike interval and of mixed frequencies that relate assemblies) as a code to partionate classes of objects in terms of their controllability (modes of control that must be recruited). Each object would have two labels in each processing stage – the characteristic of being an object, what is able to be described in localizationistic ways, and the characteristic of recruiting different modes of control, what involves legitimate temporal-antilocalizationistic codification and is the very core of a general mental code, that uses analogical processing as a tool to discriminate and partionate modes of controllability.

Let us suppose for example a certain degree of representation for a certain complex object – e.g. "driving a car". This "object" recruits a large amount of processing in the brain, the subject that manipulates it being able to quickly adapt to different types of cars. Driving a car requires learning and, after learning, one is able to handle it in a quasi-automatic mode (differently from the first stages of the process in which the mode of control is fully conscious-voluntary).

Objects that pop out into consciousness are pre-consciously processed in order to allow the system to route them to "consciousness" (and hence to a certain degree of voluntary control over them) or to the automatic mode. There might be some label in the way the system represents, through oscillations and synchronizations, an object, for example "driving a car"or a "cat", that enables, without consciousness, the system to know if the processing-address is the frontal area (more tied to voluntary-conscious manipulation) or the cerebellum (more tied to automatic modes of manipulation).

The question about the automatic "cat" (non-conscious) and the voluntary-conscious "cat" has two different approaches: either there is a label in the object (voluntary-conscious or automatic), being the control mechanism of voluntary modes and of automatic modes the same; or the mechanism of control is diverse regarding some specificity in the way the object is represented in pre-conscious stages of manipulation.

Consciousness in our view is a function that deambiguizes non-convergent (from the point of view of a neural network description of the problem-space), unstable (from the dynamical point of view, mathematical structures that describe a certain object and that presence of a bifurcation value in one parameter, hence leading to unstability from the structural point of view and to topologic modification in the phase-portrait of the system), undecidable (for symbolic formal system that exhibit Halting Problems or that are not able to verify gödelian sentences), and risky decisions (from the point of view of a decision-making description of the system, recruiting risk and the necessity of justification to one’s acts – one of the very basis of ethics and moral behavior).

Consciousness is in this way a kind of metarepresentational pattern that is summoned when: a) there is not enough representation to the problem to be handled automatically; b) when there is a need to create a version of the decision that can be summoned as a justification to one’s act. The problem-object that is under cerebral manipulation is presented in a user friendly, linguistic-propositionally-laden form to consciousness through synchronizations of differences sources – one that represents the problem, and others that link the description to linguistic and memory representations that render the contents of consciousness similar to the mental-linguistic environment we use to live with and that is largely introjected through the submersion in a semiotic culture (besides the a priori aptitude to manipulate series of linguistic events – the propositional a priori capacity of the human brain). The qualia of consciousness is the only emergent property that is reached whenever large amounts of neurons synchronize to allow decisions under risk over unstable mathematical structures that describe the problem-space being handled.

The function of consciousness might be controlling plans of behavior, and interpretations of perceptions that are generated non-consciously and that show unstable trajectories in a mathematical deterministic non-linear system that show structural unstability (hence bifurcation parameters value). The function of consciousness would be to represent the problem in a user friendly linguistic-propositional way of description (that is possible through the synchronization with representational patterns of linguistically-linked categories and memories), allowing the subject to operate in a certain time window (it is supposed to be 100 to 150 msec) vetoing or allowing the plan or the interpretation to go on, and may be resetting parameters in order to allow the system to manipulate this problem automatically in the future.(11,15)

Volition and free-will are compatible with the interpretation of a system that generates hypothesis and plans in a preconscious pattern but, in certain situations, in our opinion, describable through unstable trajectories from the structural point of view), recruits a metalevel, linguistically-embedded, to solve the problem.

The metalevel might accommodate different degrees of consciousness, being three the main types: control without awareness (priming and other events of consciousness – i. e. a particular mode of control over action and a certain way to semantically describe the object of manipulation), control with awareness (the typical state) and, in the supraindividual domain, a certain mental mechanism of control over behavior (the "collective consciousness" that is present in the Word 3 of Karl Popper) (16) without explicit awareness. Priming and social-moral rules might be symmetrical regarding a conscious style of control without awareness, and consciousness, in the individual domains (control plus awareness), would stay in the middle of these symmetrical types of semantized control without awareness.

The temporal code would be, in our opinion, something that segregates modes of control, or routes to different controlling mechanisms to each object, in the brain. Each representation of an object, be it simple or complex, like "driving a car", has a certain degree of localization (captured in a Barlow-like operation of what is going on in the frequency of spikes domain and linking different assemblies with a constant hebbian rule of connection). But the real temporal code might use information in the interspike interval and in the non-linear domain of frequencies that describe the object in order to route the present object to conscious control or to the automatic mode.

The real temporal code would be this way a label that is grasped whenever an object is presented to the brain, and that generates an error signal in different levels, whenever a certain loop doesn’t have enough representation to lock the problem. (13,14) This way, if one considers successive loops that try to lock a certain oscillation that represents a problem-object, the system either locks or sends an error signal to another loop. This is done in the frequency domain and to interpret this in the functional way that divides objects primarily in voluntary-conscious objects and in automatic ones, one has to approach what is going on in the neuronal level, synaptical level, assembly level, with a real dynamical approach, seeing the interespike interval as a potential source of temporal codes that allow the segregation of different modes of control (apart from the characterization of the object itself). The connection among neurons recruits, to this operation, frequency-synapses (17), a new way to see different functions of synaptic links, according to partitions in the frequency-correlation domain.

A dynamical strategy to surpass phrenological conceptions of the brain and the very emergence of the mind and conscious might use the frequency as a mode to segregate types of objects and its destiny regarding modes of control. The conscious mode is a type-stage of manipulation that might be recruited whenever the stability is not present in one level and another level has to recruited (with a progressive redescription of the problem-space with linguistic tolls, what involves the synchronization with different brain areas that store linguistic-devices (proposition-capacity, linguistic events and memories). There might be a certain deterministic non-mechanic (non-algorithmic) way to interpret this phenomenon, and the description of it in real experiments is difficult regarding the quantity and complexity of the central nervous system and the estimate of the dimensions and form of the system that might be representing the actual problem.

From the theoretical point of view, we think that a legitimate temporal code would not have a commitment with the representation of objects, but with a representation of the function (mode of control) that should intervene to manipulate this object.

1. Pylyshyn Z 1986 Computation and Cognition. MIT Press, CA

2. Rumelhart D, McClelland J (ed) 1986 Parallel Distributed Processing: Explorations in the Microstructure of Cognition vol 1: Foundations. MIT Press, CA.

3. Kahneman D, Slovic P, Tversky A 1982 (ed) Judgment under uncertainty: heuristic and biases. Cambridge University Press, CA

4. Penrose R 1994 Shadows of the Mind: a search for the missing science of consciousness. Oxford University Press. Oxford

5. Llinás R, Paré D 1996 The Brain as a Closed System Modulated by the Senses. Churchland P, Llinás R (ed) The Mind-Brain Continuum. MIT Press, CA.

6. Crick F 1994 The Astonishing Hypothesis: the scientific search of the soul. Scribners, NY

7. Fujii H et. al. 1996 Dynamical Cell Assembly Hypothesis - Theoretical Posssibility of Spatio-temporal Coding in the Cortex. Neural Networks vol 9 N.8:1303-1350

8. Llinás R, Urs R 1992 Rostrocaudal Scan in Human Brain: a global chracteristic of the 40-Hz response during sensory input. Basar E, Bullock T (ed) Induced Rhytms in the Brain. Birkhäuser, MA

9. Eckhorn R et. al. 1992 Stimulus-Specific Synchronizations in Cat Visual Cortex: multiple microelectrode and correlation studies from several cortical areas. Basar E, Bullock T (ed) Induced Rhytms in the Brain. Birkhäuser, MA

10. Pöppel E, Schill K 1995 Time Perception: problems of representation and processing. Arbib M (ed) The Handbook of Brain Theory and Neural Networks. MIT Press, CA.

11. Lopes da Silva F 1992 The Rhytmic Slow Activity of the Limbic Cortex: an oscillation in search of a function. Basar E, Bullock T (ed) Induced Rhytms in the Brain. Birkhäuser, MA

12. Del Nero H 1997 O Sítio da Mente: pensamento, emoção e vontade no cérebro humano. Collegium Cognitio, SP

13. Del Nero, H 1997 Computação Topológica e Controle Voluntário em Arquiteturas Naturais e Artificiais. Tese de Doutorado, Escola Politécnica da Universidade de São Paulo

14. Del Nero H, Maranca A, Piqueira, J 1997 Topological Computation and Voluntary Control: a possible role of the neocortex and the very aim of cognitive science. Coleção Documentos IEA-USP, Série Ciência Cognitiva, 29 (technical repport)

15. Libet B 1993 Neurophysiology of Consciousness: Selected Papers and New Essays by Benjamin Libet. Birkhäuser, MA

16. Popper K, Eccles J 1981 The Self and its Brain: an argument for interactionism. Springer International, NY

17. O’Donovan M, Rinzel J. 1997 Synaptic depression: a dynamic regulator of synaptic communication with varied functional roles. Trends in Neuroscience, 20, No 10, 431-433