Transduction Notes

							1)     Transduction Notes: I have taken a modular picture of the CNS, dominated by successive
     abstractive processes, and I've shown, I think, that some sort of analog coding is necessary
     throughout the CNS. In addition, I'm coming to a concept of "representation" which uses, first, a
     development of the idea of the "cell assembly" involving both top-down and bottom-up feedbacks
     and the relaxation time of the neural discharges, and second, the idea of internally-directed top-
     down stimulation of sensory and motor processes. In other words, a representation, in this sense,
     is a kind of inwardly directed generation of sensation and motor control of that sensation. Now
     the interesting question here is just how the system "sees" a representation like this "from inside."
     In the case of normal sensory stimuli, for example, this question does not really have to arise, as
     in, for example, the case of insects, which do not have such representations but do have
     integration of sensory/motor processes. I think that this type of representation is the ultimate
     basis of consciousness, although at the above level it is not yet conscious.
2)     Searle maintains that computers cannot think at all, and that even deduction must be interpreted
     with, in Searle's case, a biological entity. In the case of transduction, the claim is that thinking
     must proceed with a part of the system physically corresponding to the world interacted with.
     The problem is that there is always structural correspondence, diagrammiticity, in Peirce's terms,
     at some level between the "model" and the reality, which includes computer models. The other
     problem is this distinction between thinking and understanding. Searle more or less makes this
     distinction in the Chinese Room when the claim is that this room can produce Chinese sentences.
     Thinking is extended to formal syntactic manipulation, but the room does not "understand" the
     Chinese. But what does that mean? The Chinese room can converse, which means that it can
     employ words coherently in appropriate situations: it can pass any Turing test. Does then this
     criterion mean that the room is not conscious of meaning? Can a biological organism, an insect,
     perhaps, possess understanding without consciousness? Can unconscious human beings possess
     understanding? What of the unconscious production and understanding of words themselves; is
     this an illustration of the difference between consciousness and understanding? Yet how in this
     latter case would ―understanding‖ be different from what the Chinese room can allegedly do?
3)     Searle‘s argument about simulation is equally ambiguous. Certainly, a computer simulating the
     physical processes associated with thinking: the transport of ions across membranes, for example,
     is not carrying out those physical processes, and that simulation will not cause any real axon to
     fire. Even, in a sense, it will not cause the next line of the simulation to run: that is caused by the
     physical processes in the computer. However, a computer, as with a brain, is at root a physical
     entity, and does carry out symbolic operations for which physical operations are necessary and
     sufficient. If thought in a brain is not literally the movement of physical entities, then brains also,
     though at root physical entities, carry out operations that are not literally the movement of
     physical entities, yet for which physical operations are necessary and sufficient (more on this
     below).
4)     What then is the difference between structural correspondence at some abstract level, which (we
     assume) computer programs do possess, and correspondence at a low level, that of the physical
     instantiation of the entity? The answer would seem to be functional: does not the structure of the
     system have to match the structure of the reality with which it is interacting, so that the system is
     not just pure symbols, uninterpreted? The level at which they must match is the level at which the
     interaction between the system and physical reality occurs. If the entity is interacting with
     lightbulb sockets, then the structural correspondence must be low, at the level of that physical
     object. However, one might then argue that a computer then can output strings of meaningful
     symbols, because is it interacting with human beings, physically realized symbolic-manipulators.
     But if this argument is correct, and a computer, in its functional realm, is accomplishing
     transduction (in something like a human sense), then does transduction have anything to do with
     meaning, qualia, or symbolic anchoring?
5)     One answer to this might involve the integration of the transductive processes with the other
     processes in the entity. In the case of the computer, perhaps these are not particularly well
     integrated with the "thinking" processes of the computer. Thus, if there is no meaning ―within‖
   the computer, and if meaning for a computer must come from the transduction involving the
   person using the computer, i.e., from the totality of the human/computer system, then that person
   is essential for meaningfulness in relation to computers. If that is true, however, can we extend
   this argument and say that meaningfulness for a person intrinsically involves the environment?
   But if the person has within them the transductive process this problem does not arise, or at least
   is not so clear-cut.
6)    What then is the essence of transduction? One alternative is that this essence must be the active
   and continuous comparison between two or more structures, rather than the merely passive
   translation implied by the above discussions. In the case of human beings, two of the structures
   compared are that of the representations of sensory stimuli and the internal abstract symbolisms.
   In the case of computers, does the direct translation of one abstract symbolism into another (e.g.,
   machine language and C++) constitute a form of transduction? If that is correct, then a straight
   translation from one form of code into another cannot properly be termed transduction if we want
   transduction to be something human. But if that is the case, then the relatively simple
   analog/digital translation found in the human sensory system cannot be transduction. Only
   through corrective and modifying feedback can there be transduction. However, what then of
   feedback between abstract codes? If abstract codes can interact and modify each other (which they
   can in a machine), then is this not the same process? How then is a code recognized as one relating
   to or structurally matching some aspect of ―reality‖? Must "reality" impinge, i.e., must a
   description of the world structurally correspond to it? It would seem, if this line of speculation is
   correct, that the form of the code is irrelevant, only that one code be taken as a "source" and
   another as a "model" for the first, with reciprocal interaction between the two, with the "source"
   defined as primary in that function, i.e., harder to modify.
7)    In addition, if the above is true there must also be input-output comparisons. Feedback between
   sensory forms and effector forms must ultimately govern this system. Feedback must start with
   inherent biases to guide the feedback directions and strengths, and these must be refined. But this
   refinement may involve some or all of coding configuration refinements, coding specificity, and
   intermodal discrimination refinements. The anchoring (to ―reality‖) of the coding is not sufficient
   if it only proceeds in one direction, e.g., sensory to internal symbolism, because in that case the
   coding translations are entirely arbitrary, even if fixed. Only with effector feedback can the symbols
   ultimately be anchored, and only if this feedback is biased in particular directions or to attain
   particular goals; if that is true, the code cannot be entirely arbitrary. But if this is true,
   correspondence between codes, even analog codes which have structure corresponding to the
   sensation, is irrelevant for symbolic anchoring, because this correspondence is arbitrary as far as
   its extent and parameters go. The analog nature of coding does not relate to symbolic anchoring,
   only the input-output biases do. What the code parameters do relate to, however, is the nature of the
   experienced qualia.
8)    In addition, the concept of transduction involves the idea of a system with arbitrary symbols of
   some sort. But what does that mean, exactly? In a digital computer, what are arbitrary are the
   locations of particular values, the addresses, which can change, and the values themselves: the
   patterns of binary or "duple" notation which comprise the value of a particular variable at a
   particular address. The physical substrate is unaffected by the former, and changed by the latter.
   The same values at the same address can have multiple meanings depending on the program, i.e.,
   the ultimate interpretation of the symbols lies with the programmer. Also, even given a particular
   computer running a particular program, memory addresses are arbitrary and variable, depending
   on the computer's memory history. The same or different number at the same or different address
   can be interpreted in an infinite number of ways. Secondly, it is conceivable that there are two
   identical or virtually identical programs whose symbols are interpreted differently. Thus, the
   calculations for asteroids or billiard balls may be the same and involve exactly the same logic and
   program, but be interpreted differently by the programmer, and so forth. This is Searle's objection
   with the example of the Chinese language and the stock market. If this (argument of Searle‘s) is
   correct, the computer succeeds in separating semantics and syntax in a way that humans cannot.
      This is an ironic refutation of Searle's and the traditional position on human thinking, given that
      the Chinese Room argument is correct, as necessarily involving this separation.
9)       However, one problem with the above is that all real-world stimuli possess analog dimensions,
      usually that of intensity. So a correspondence within the nervous system to that analog structure
      is not unique to that code, and thus meaning, unique meaning, based on that correspondence,
      cannot be had. Thus the correspondence within the visual system of frequency of neural impulse
      to intensity of light is basically the same correspondence as that within the aural system to
      intensity of sound. These are both analog, both structurally similar to the real-world structure, yet
      both are coding the same thing, the abstracted characteristic of intensity, in the same way. Thus,
      given some interpreter looking at structural correspondence, that interpreter might be able to infer
      intensity, but that intensity could not be assigned a sensory modality; thus there is no symbolic
      anchoring.
10)      Another problem with transduction, however, is that there are multiple ways of representing
      simultaneous variables, which are arbitrary. Thus, with sound, the frequency of neural discharge
      could be proportionate (if logarithmically) to the amplitude, which it is, or the frequency of neural
      discharge could be proportionate to the frequency of the sound, which it is not, and the neural set
      could relate to the amplitude, instead of to the frequency, as it does now. In both cases, there
      would be a) analog codes, and b) direct relation between a physical characteristic of the stimulus
      and the code (incidentally, the neural set relating to the frequency is a digital code, even if the
      transduction mechanism is analog).
11)      How does symbolic anchoring arise at all, then? The only way would seem to be a) the linkage
      of a code to some physical characteristics of the transducer, and b) the encoding of the nature of
      that linkage. There has to be a code for the code. So this either must be hard-wired, through
      evolution, or it must be learned. But is there any constraint on the nature (analog/digital) of the
      code? It would seem that symbolic anchoring does not necessitate analog coding, merely this
      physical/code connection. For example, while the ear has an analog representation of amplitude
      in terms of firing frequency, its representation of frequency is in terms of sets of cells firing. But
      this must be digital; there is no direct translation of frequency, certainly at the highest levels, into
      firing frequency. Particular sets of hair cells at particular locations in the cochlea are activated
      according to frequency. This then is a digital code, by every definition; in fact, this code is similar
      to the absorption spectrum coding in the visual system. However, at some level, there has to be a
      physical instantiation of the coding which directly relates or connects to or determines the code's
      meaning. Thus although digital computers are physically instantiated, that instantiation has no
      ultimate relationship to the meaning of their codes.
12)      But what if these computer interpretations were fixed: i.e., built in to the physical circuits? What
      if particular inputs led to particular memory locations, particular logic circuits, and so forth?
      Would it not be the case then that the interpretation of particular calculations would have to be
      made in certain ways, so that meaning was inherent in the processes? Would not then digital
      computers have "meaning" or "understanding" in the sense of humans? However, even then the
      second objection must be answered. If a digital computer is a purely symbolic device, then even in
      this situation, the billiard/asteroid objection is still valid.
13)      The problem here is that one could argue that while a particular computer is running a
      particular program, this situation does in fact exist. That is, the configuration in the computer's
      RAM is assigned particular values and meanings according to the parameters of the program;
      even if these are initially or intrinsically variables, when the computer is running a particular
      problem, those variables have been assigned. That they could be assigned other values is not
      relevant, or is it? How? It is still true that the basic operations in the computer are general Boolean
      operations: AND-OR gates, truth tables, and so forth. While a truth table formulation is fixed, and
      even though that formulation is assigned values, nonetheless there are no values inherent in the
      operation of the truth table determinations. However, in the CNS there is no such general process.
14)      To approach this in another way, let us ask the following: what would a general-purpose,
      syntax-only device have to look like? Would it be a digital computer, a Turing machine? This device
   would have to use purely formal operations on general symbols for which an infinite number of sets of
   individuate meanings or values could be substituted. The operations on the symbols would have
   to consist of no more than a set of abstract logical processes, strictly well defined, and the
   meanings, when substituted, equally well defined. But this does describe Turing machines, and
   digital computers. Could this device be symbolically anchored? Even when it was running a
   particular program, the values at the addresses, duple numbers, have no interpretations. Is it
   possible to derive their interpretations from the logical processes into which they are substituted?
   This seems unlikely, since it assumes that every given logical process, however simple or complex,
   uniquely describes some aspect of reality or some model.
15) Then the opposite: what would a semantically based machine, a Peircean "theorematic" machine,
   look like? It would have to not employ a finite set of well-defined operations, nor could it use general
   symbols. All the symbols would have to not only have specific intrinsic values but specific intrinsic
   meanings; thus in a sense they would not be symbols at all, but entities. The operations on these
   entities would have to be specific to both the entities and the goals of the operations. But this
   sounds like an analog computing device: a slide rule or a brain.
16) But what is coding, what is computing, what are symbols, in this case? Consider a disturbance
   propagating inside a rock. This may be looked at as a totally passive system, with the pressure
   wave propagating through the rock, producing, say internal fractures as a purely meaningless
   physical process, or it may be looked at as something analogous to the reaction of a primitive
   worm, with a few neurons, to some stimulus. Why cannot the pressure wave be understood as a
   wave of transformation, of information, a signal, which is transduced into the action of fracturing?
   But here there is no transduction, since the pressure wave is exactly the action of fracturing.
17) Similarly, do the reactions of the neurons in a simple worm constitute a translation of an
   environmental stimulus into information, into meaningfulness, into symbols? Here the neural
   activation is not the initial response to the stimulus, which is a chemical reaction of a receptor, nor
   is it the physical action of the worm to the stimulus, which consists of muscle contractions. There
   is some sense in which the initial stimulus has been converted from one type of physical event,
   say a light flash, into another, a neural discharge and then to another, a muscle contraction (two
   types of propagating chemical events). Yet that conversion is as physically bounded and driven as
   would be a chemical change in the rock as a result of a pressure event: a shock. If the latter chemical
   change resulted in a change in the rock, would the pressure have been transduced in the same
   sense (i.e., perhaps into meaning, even symbols of some sort) as the light was transduced in the
   worm, or is the opposite true of the worm? (This line will be continued later: see 60).

18)  Now consider whether feedback is necessary for a system to have meaning. What is the nature
   of the feedback? If we assume that a thermostat has no viewpoint, no ―that which it is to be,‖ i.e., a
   thermostat does not have meaning, what does and why? The feedback of transduction, as we have
   seen, is only related to code maintenance, refinement and translation.
19) That is, a straight-through code translation, however fixed at the input end - anchored - is merely
   a symbolic substitution, whether digital/analog or not. In addition, the input end is always
   arbitrary, even when an analog code shares structure with reality; the particular structural parameters
   are arbitrary. Thus in this simple case there is no true symbolic anchoring. Only when the code is
   compared to an output is there the possibility of anchoring (or when the transduction is hard-wired,
   through evolution, say; see 273 below). However, a simple feedback loop, which is compared to
   an output, as with a thermostat, is again no more than a straight-through code substitution,
   maintained within some parameter range. The only possibility for true anchoring, then, is when
   the code can vary, either because the internal codes can vary or because the output relationship
   can vary (the next level of complexity beyond human, then, is when input parameters vary, and
   must be selected, as we select modalities). The internal coding can vary only if there is interaction
   between that coding and some other input or output source (learning). The output coding can
   vary if leaning is necessary and/or if there is interaction between various input sources and
   output effectors, altering responses. If there is no possibility of variance, then the concept of symbolic
      anchoring is empty: coding translation is straight through. Thus a thermostat or any similar device
      cannot "have meaning."
20)      Anchoring, then, must be the dynamic act of anchoring, not the state of being anchored at a
      particular set of input-output relationships. What is the act of anchoring? As outputs are refined,
      inputs are also refined, and so is intermediate coding. The same holds with variance. The
      refinement and variance have the possibility of code drift: when input and outputs do not match:
      an input with no feedback response, an input with destructive feedback response; internal codes
      that do not result in output, or result in destructive output. This seems to imply a system which learns
      (and note Peirce here), even to the extent of taking unforeseen inputs and responding with
      previous functional outputs (this also eliminates the Chinese room problem and Chalmers'
      thermostat claim).
21)      To put this another way, non-recursive functions do not describe symbolic anchoring (however,
      evolution can provide the temporal dimension also; see 273 again).
22)      Even if the computer were doing operations relating specifically to trees, for example, shaped as
      triangles, and the computer was doing this with pure symbolic transformations relating to the
      triangular nature of trees solely as triangles, and even if trees were the only things in the universe
      that were triangles, the computer would still not be "thinking of" or "understanding" trees, only at
      most certain abstract relationships involving triangles. In other words, performing only processes
      which relate to entities as some set of abstractions of their characteristics or of their relationships is
      not to deal with those entities but only with the abstractions. What it would be to deal with a tree
      as such would be to deal with the data relating to the tree on all the levels of abstraction available.
23)      What then is "abstraction"? The answer to this can only be functional and relates to the nature of the
      original sensory information, and thus does not really matter as to specifics. The reasoning is this: whatever
      the original sensory information about some entity which distinguishes that individual from others of the
      same class is just exactly the individuate, "concrete," information about that entity. Any information
      abstracted from this is concerned with classes, perhaps infinite classes, of entities. Abstract information
      cannot be unique to individuals in the functional sense that it cannot retain the original, "sensory,"
      information that distinguishes individuals; that is just what "abstract" means. If a computer deals only
      with abstract information, it can never have "meaning" in a human sense because it can never deal with
      individuals, only with classes.
24)      Thus, some of the original sensory information discriminating individuals must be retained in
      order to have "symbolic anchoring," if symbolic anchoring relates to having meaning in a human
      sense. Whatever type of code that information consists of is irrelevant as to symbolic anchoring,
      but not to the nature of qualia. This captures the difference between "syntax" and "semantics," and is the
      reason why lists of labels of characteristics, however long or complete, cannot capture individuation. The
      members of such a list are abstractions, and the argument is that the totality of these abstractions,
      given that one list is different in at least one of its members from all other lists, will characterize
      that individual. In some sense this is correct; the individual will be identified. However, since that
      identification is performed utilizing lists of abstractions, the implication must be that to proceed
      from that level of abstraction to the concrete, sensory level, one level down, is a function of a
      unique list of abstractions. However, given the argument above, this is incorrect; it must be that
      the retention of original differentiating information is the distinguishing factor. That is, it must be
      the case that the original differentiating information, when coded, contains at least some part of
      the code which is unique to that individual, in contrast to consisting of a unique list of terms. Even
      if that list contains one term not contained in any other individual's list of terms and that term is
      an abstraction, then by virtue of the nature of an abstraction, i.e., a class descriptor, it is in theory
      possible that some other individual could contain that term in its list. Given that, that description
      is not a "concrete" description. A "concrete," "sensory" description must have some individuating,
      non-class descriptor. That is, since that identification is performed utilizing lists of abstractions, it
      is argued that to proceed from that level of abstraction to the concrete, sensory level, one level
      down, is a function of a unique list of abstractions. However, given the characterization above,
      this is incorrect. Any list consisting solely of class information, by its very nature, must be a list of
      abstractions, and thus any of its members may always stand for any of a potentially infinite set of
      individuals. But this just reemphasizes that digital computers deal only with abstractions. It must
      be, then, that the retention of original differentiating information is the distinguishing factor
      between digital and analog systems.
25)     But what is the "original sensory information"? Given the arguments above, it cannot be simply
      the lowest level of analog processing; that is not sufficient to distinguish this information from the
      corresponding level of analog information from other sensory modalities, if only this is retained to
      the highest levels. Thus there must be retention of the patterns of abstraction of this lowest level, i.e.,
      there must be retention of some at least of the higher levels of abstractive processes throughout
      the system. This does conform with our experiences, i.e., we can, although imperfectly and
      modified, see the edges, colors, lines, etc., which are constructed at fairly low levels. In other
      words, we do not just see a house as a distinguished, named, meaningful whole (shape), we see
      the component lines, colors, edges, etc., that have gone into the internal construction of the house
      even as we see the whole gestalt.
26)     This is also what makes Searle's analysis {Searle 1990 #9}, as far as it goes, correct. Inasmuch as
      the digital computer deals with symbols that represent classes, it deals with syntax, not with
      semantics, and can never have meaning until it incorporates and retains information that
      individuates individuals in its input. This is also what makes Peirce's analysis {Ketner 1988 #16}
      correct. Inasmuch as humans do necessarily utilize this individuating information throughout
      their thinking, that thinking is different in kind from that of the computer. This is also what makes
      Lakoff and Johnson's work important. They have, in recognizing the semantic component of
      thought and its pervasiveness, begun an analysis of exactly what the nature of individuating
      information is.
27)     But this now becomes the central question. What is the nature of the sensory information that
      individuates individuals? Suppose we assume that sensory coding is analog coding. The problem
      here is that one might suppose a situation in which the neural outputs from the cochlea, for
      example, were routed into the post-retinal stream, and vice-versa. To put this another way, visual
      information is taken from two planar sensory surfaces, consisting of cells that represent
      information according to the spatial relationships of the cells stimulated. In addition, frequency
      sampling is performed on the retina in three areas over the electromagnetic octave: low, middle,
      and high. The summation of this sampling results in a net frequency assigned to a given area on
      the retinal projections superposed. Further up, this information is processed according to an
      opponent frequency system as well as an amplitude system. The combination of frequency
      sampling and opponent sampling and computations result in the qualia characteristics of the
      electromagnetic octave: colors. Now why could not sound be processed identically?
28)     Sound, however, is processed according to frequency and amplitude over a much greater
      frequency range, with virtually continuous sampling over that range, which is both frequency and
      position coded (position of the stimulated neurons on the basilar membrane) by the neurons.
      Further up, there is no doubt some sort of Fourier addition performed on the various frequencies,
      with the complication that while sampling at low frequencies has both frequency and position
      components, that of higher frequencies has only position information, since the frequencies are too
      high for the neurons to track (although perhaps they follow a lower harmonic and add position
      information to that frequency information).
29)     Can it be assumed that the various information transfer/processing neurons (neurons in the
      network past the sensory units) are identical for the various sensory systems, and if not, what are
      the consequences of the dissimilarities? That is, neural discharges do not have color or pitch, etc., so all
      that assuming this difference would do would be to push the problem back a level to asking how
      the neural coding differences between cell types resulted in different qualia - and why they
      should. Frequency coding for individual neurons, and phase relationships summing on cell bodies
      are the parameters in this system, roughly speaking, unless the various neurotransmitters have
      classes of effects specific to various sensory modalities and other functions. However, how would
      this affect the summing on the cell bodies, which is, after all, the base processing of the system?
30)   In addition, although in theory this seems arbitrary, in fact one could do the same with one‘s
   sensory apparatus and its codes. The physical nature of the eye, for example, leads to its particular
   coding, which is both different from that of the ear and related to the structure of the eye. Thus
   given the physical structure of the eye and its attendant neural processes, one could infer at least
   some of its coding structure. But in general this reasoning does not work at the input stage. That
   is, to take Searle‘s example, a stomach uses chemicals to deal with chemicals: the physical entities
   of which the stomach is made and with which it works are of the same type as those entities
   which the work on and, in a sense, represent. But in the nervous system, the nerve impulses are
   not of the type that they react to and with. That is, the sensing of pressure does not take place
   employing a pressure wave traveling up the neuron, but through transduction of the pressure on
   the receptor into nerve impulses. Thus Searle‘s analogy does not, strictly, hold, because in the case
   of the stomach there is no coding, strictly speaking, nor transduction, necessary; all that is
   necessary is that two physical systems match functionally. Whereas in the nervous system, it is
   necessary that a physical system be translated into code, so that the operations of the body,
   instead of being a continuation of the incoming physical processes, proceed in reaction to the code
   rather than in reaction to the physical processes originating that code, in contrast to the stomach,
   for example. And has been argued, the coding of those impulses, although it is analog and does
   structurally match some aspects of the pressure, is arbitrary, as arbitrary as are digital computer
   codes. That is, the frequency of the neural impulse coding for amplitude of pressure might as well
   code for the angle at which the pressure is applied, while coding for amplitude might be
   accomplished through activation of greater numbers of neurons, instead of the opposite (but see
   273).
31) However, given what we experience, the coding is not arbitrary. Consider Searle‘s argument
   based on the arbitrariness of the symbols, i.e., that one could look into the Chinese room at the
   symbols and their manipulation and not know whether Chinese was being processed or stock
   market prices. His counter-example is that of the stomach, where given some knowledge of
   chemistry, one can look at the stomach as a physical system and infer what it is about. This
   argument has force. However, with the CNS, if one knows something about visual
   phenomenology, one can look at visual vs. aural coding, for example, and infer which coding
   corresponds to which phenomenology based on the correspondence between the
   phenomenological structure of the sensory modality and the structure of the corresponding code
   for that modality, as described above. Thus in the case of the mind, the phenomenology, the
   experiences, correspond to the physical functioning of the bodily organs in Searle‘s argument. It is
   not the analog nature of the coding, not the structural correspondence with reality, but it is that
   the coding for a particular sensory modality is unique and reflected in the phenomenology of that
   modality.
32) However, it must be said that ―coding‖ in this sense does not refer to some simple patterning of
   any given stream of impulses, but must refer to large spatial/temporal patterns. That is, it is
   largely unknown whether the neural representation of color, on a per-neuron basis, could be
   distinguished from that of sound. However, it is clear that the patterns of impulses from the retina
   to the cortex, and within the cortex, in virtue of their interrelationships, must be different from
   those, say, originating in the ear. For example, the multiple parallel streams from the retina must
   interact according to much more complex excitation/inhibition rules because of the three-color
   opposition than sound streams, since there is no comparable opposition in sound.
33) I will assume for the sake of argument that these parameters, while relevant, are integrated into
   the common pool of dynamic presynaptic computations on the cell bodies. Given that this is true,
   it can only be the variety of processing above which determines qualia. If this is true, each sensory
   modality must be processed differently from each other. If this were not true, then when two were
   processed similarly, how would different qualia be experienced? At some point, if only because of
   the experienced unity of consciousness, there is a central processor of some sort. Thus this central
   unit would have to ―know" that the same types of information structure came from two different
   types of sensations. But then "different types" of sensations - different qualia - would have to be
      assigned by the central unit to codes that were the same in all but their originating location. This is
      perhaps possible, given the modality/space organization of the cortex. But where or how would
      the central unit originate these types? It would have to perform a) intrinsically , i.e., genetically
      determined, different types of processing on similar codes, or from b) experience, i.e., it would
      learn by trial and error to disambiguate, i.e., perform different types of processing, on the codes.
      Alternatively, the codes are different and lead naturally to different interpretations, i.e., different
      qualia. This latter is also supported, as has been argued, by the anatomy and physiology of the
      relevant sense organs, and by what seem to be the similarity, even identity, of sensory modality
      phenomena over people.
34)     That is, given that neural impulses have no qualia, and that qualia must be constructed in part at
      least from their patterns, why should there be such uniformity of experience (over individuals), if
      not through hard wiring? Skeptical questions aside, this does seem to be the case.
35)     But the above did not answer the question (26) about individuation. However, that would seem
      to be as follows: each individual element, cognitive or sensory (for example), is different from
      others because of whatever subtle differences in characteristics, manifested as differences in
      appearance, i.e., in phenomenal characteristics. Differences in observable characteristics may be
      coded as analog (with matching structure) or digital (arbitrarily structured) streams of
      information. A digital stream might be individuated on the basis of, say, time stamps, but
      nonetheless, each number in that stream is a separate entity in itself: it must be interpreted,
      assigned a quality, and in addition must be placed in some particular relation to the surrounding
      numbers: a pattern must be imposed on this stream.
36)     This last point is critical. In the CNS, the analog stream moves along axons and is then presented
      to a cell body. The cell body, as far as post-synaptic, pre-discharge potentials, integrates the
      streams over time. That is, the dynamic patterns of depolarization on the cell body, because of the
      physical structure of both the stream and of the neuron, change the temporal stream of impulses
      into a varying area of depolarization on the cell surface. In other words, the physical make-up of
      the neuron does for the analog stream what either abstraction or interpretation does for the digital
      computer: creates a pattern from the stream of information. Let us look at this in more detail. As
      impulses arrive to the end of the dendrite, they are transduced into packets of neurotransmitter,
      rapidly diffuse across the gap, and release the transmitter on the cell body surface (except for cells
      which are stimulated or inhibited directly by current). This results in (usually) subthreshold
      depolarization of the cell body surface, which is a dynamic analog realization of the incoming
      neural signal. That is, the depolarization spreads out from the area of initiation proportionally to
      the frequency and amount of stimulation at that initiation area, and the area of depolarization
      resulting increases to some maximum, with the degree of depolarization declining (according to
      some function) of the radius of that area. This area is temporally dynamic, so that it is in a state of
      dynamic stability after some length of time given a constant stimulation. When that stimulation
      increases or decreases, the area correspondingly grows (given a hyperpolarizing stimulus, of
      course) or shrinks. Now this just covers that one dendrite, and usually a cell in the cortex has
      several hundred or thousand such. Given that, these areas are constantly interacting, adding
      and/or subtracting the depolarizations (or hyperpolarizations) where they overlap, until the cell
      has "computed" the resultant, and discharges.
37)     This system provides for both temporal and spatial pattern synthesis on single cells (for multiple
      cells, see below). That is, there can be no internal "perception" of something as a pattern unless the
      pattern's components are integrated simultaneously into some resultant. Temporal synthesis can
      take place because of the maintenance of the multiple depolarized areas as they interact; spatial
      synthesis can take place given the same type of interaction from incoming signals originating in
      different functional locations.
38)     As far as each individual cell body goes, the patterns are probably similar across sensory
      modalities. However, multi-cellular processing is accomplished through the creation of
      foreground/background, i.e., contrast through excitation/inhibition differentiation as stimulus
      intensity varies. Thus, in the aural modality, at the beginning of the processing, pitches, let us say,
      are individuated through the excitation of cells firing at a given pitch and the inhibition of
      connections to those cells firing from different pitches. The primary pitch is the foreground,
      brought forth; the others become the background, receding. Thus neural pitch unit processing
      proceeds to structure our experiences. Analogous processes occur in the visual system; however,
      instead of frequency, spatial figure components are created: lines, angles, etc., and in addition,
      colors are foregrounded as blends and contrast pairs. It would seem that qualitative information is
      carried up, in effect, through the individuation of foregrounds, to differentiate both between and
      within modalities.
39)     Thus the analog nature of the neural impulse is crucial because of the construction of the cell
      body. It is in fact on the cell body, rather than along the axon, that it is important that the temporal
      pattern in the environment is reflected, since that is where it is interacted with other such patterns;
      the axon is merely a conduit.
40)     A single or small set of neurons responding to an angle is differentiated from those neurons
      from those responding to a pitch. How is this done? First, there is merely the location of the
      neurons and their output neurons; processing location translates, ultimately, into response
      location or differentiation. Second, it is possible that the actual coding is different in each set,
      either by virtue of the actual impulses on each cell body, the set of impulses in an ensemble
      relating to angle, for example, as a whole, or in the effect of different neurotransmitters for each
      set of cells that are modality differentiated. The first would seem more likely if it were not for the
      ultimate integration of sensory modalities; once these are integrated, processing location is
      irrelevant, and yet modality information is not lost.
41)     What of the spatial pattern? In the aural mode, this is not so important, although phase relations
      are incorporated. In the visual mode, it is interesting that spatial patterns, as mentioned above, are
      integrated onto single or small and well-connected sets of neurons. Those can be conceived of as
      single cell surfaces connected by short axons. But even so, when more than one neuron is
      involved, the necessity for hyper-polarization, i.e., neural firing, to produce integration must
      imply the loss of information and processing. The "binding problem" is not solved by merely
      noting time-congruencies over ensembles, but involves the integration on individual cells.
      Congruent processes, whether by time or function, must be automatically combined into patterns.
      How is this accomplished in the visual system, which has so many (two dozen, at least?) different
      functional areas? To create a singular visual pattern, all those areas must not merely be processing
      temporally congruently, for even if that were the case, what would "know" about that
      congruence? Why would that differ from, say, two persons‘ cortexes processing simultaneously?
      Even if some sort of processing result from those areas was sent to one particular area for
      "integration," if that merely consisted in various sets of neurons in that one area firing
      synchronously, without information sharing, that synchronization would be non-functional,
      unless the information were blended as is the temporal information on the surface of the cell body.
      How might this blending be accomplished?
42)     There are only two possible ways. One would consist, as mentioned below, of a set of neurons
      being the functional equivalent of a cell body surface, if they were tightly enough interconnected.
      That set then would function as a single neuron, in effect, through feedback loops. The feedback
      loops are absolutely essential for differentiating this processing from linear processing, where the
      unidirectional linking does not achieve unification of the informational streams. In other words, in
      a neural ensemble, feedback compensates for the absolute integration that occurs on an individual
      cell body: it compensates for the lack of time delay and for the reverse flows. In other words, on a
      cell body, there is the equivalent of a single dynamic interactive system: all the components are
      (possibly) recursively interacting. However, this recursion, in an ensemble, must be accomplished
      through feedback consisting of spike trains. This delay, however, is no more inconvenient, so to
      speak, for this system than is the speed of light delay in large physical dynamic systems; the
      system is still analog (see 71).
43)     The second method would be through the alteration of established information flows. These
      flows would have to be loops, since self-monitoring for the informational input would be
      necessary in order to compare present and past flows. The conceptual pitfall here is to conceive of
      the "streams" or "flows" of information down the axons as the essential and functional
      information, whereas in reality this is merely the transduced form of the processing which is
      taking place on the cell bodies, pre-discharge, the essence of the neural informational processing
      activity. Both of these cases, with this caveat, would seem to be identical. This would account for,
      perhaps, temporal and spatial integration. What of the creation of qualia?

44)     The argument: an entity interacting with the world must do two things, first, retain information about
      sensations as individuals, second, create class information. If only the former is done, then the interaction is
      on a par with any natural object: a rock which is deformed, for example, retains and manipulates
      information about that sensation on the level of the individual stimulus. However, if only the second is done,
      then the entity is a computer, dealing only with classes in the form of abstract numbers. This information
      can never have any intrinsic meaning or quality (see 16 and 59).
45)     However, the difficulty is that there are no ―colored‖ information streams, in effect; even the
      deformation of a rock, if it is transduced into electrical fields, for example, loses the ―deformation
      quality‖ in the process. Neural impulses similarly do not have the qualities we impute to the
      corresponding sensations. How then can qualities arise, individuate information be retained, and
      class information be generated, in the same system? The key to the first must be in the nature of
      the processing in the individual biological systems, at the lowest level. This relates, in addition, to
      the individuation of the sensory modalities themselves. That is, each sensory modality, in order
      not to be even potentially confused with any other sensory modality, must proceed through
      processes unique to that modality. Perhaps this must happen even at the membrane level; this
      may account for the multiplicity of neural transmitters. At any rate, the processing functions on
      the cell body must be unique to each sensory modality (although see 32). In addition, these
      processing functions can account for the formation of classes. That is, the temporal integration of
      impulses on the cell body literally crates a physical reality, a response on that cell body, which is a
      gestalt, a combination integrated over time, of the post-synaptic, pre-discharge potentials on the
      cell membrane. The cell‘s firing conveys to the next cells information about that integration. This is
      the creation of class information from the individual sensations.
46)     One question here is whether information about individuals has been lost, and whether that is
      relevant. On the one hand, if information about individual basic sensations is lost, one might
      argue that the system, analog or not, is now equivalent to a digital computer in that the
      information cannot be related to individuate meanings anymore, i.e., the information is, once
      integrated, necessarily about classes of stimuli. On the other hand, if the processing functions, that
      is, those processes of the cell membrane which have produced the class information, i.e., the
      coding created from the temporal gestalt, are still unique to each sensory modality, then one can
      argue that individuate information has in that sense been retained, since the resulting gestalt is at
      least unique to that modality. That is, the significance of the individuality of the information is not
      in the retention of the original sensations, but of the original sensory modality. In other words,
      even upon translating the membrane potentials into a number, i.e. the frequency of the neural
      impulses upon firing from that membrane, that number is physically associated with a unique set
      of processes: those on the membrane producing them, and those on the membrane receiving them,
      which are unique to that sensory modality. Thus it is important that at all levels of each modality
      processing unique to that modality be implemented and retained in order to retain meaning.
47)     But it must be remembered that however much the original sensory information has been
      abstracted, something of that original information is retained, because that abstracted information
      is still experienced as possessed by that sensory modality, and also because we seem to be able to,
      at least in special circumstances, experience at least some of the original information, i.e., we see
      that squares have corners and that those are two lines at an angle, and that latter is a relation
      between two lines.
48)     One implication of this might be that if processes become universal, then universal classes
      become possible: true abstract reasoning becomes possible at the point of integration of sensory
      modalities if their information is integrated simultaneously and outputted as classes incorporating
      that integration.
49)     But the ―retention of meaning‖ as defined above is not the experience of qualities. Nonetheless it
      must give rise to that experience, because, again, especially upon integration, the neural impulses
      themselves, and the membrane depolarization itself, does not per se have quality, especially the
      qualities we experience. Where then must those qualities come from? They must result from some
      sort of internal reaction to the individual processing functions, as an abstraction of the running
      processes. Without being able to describe the nature of this reaction, one must still ask what the
      function of this reaction is. One obvious answer is that if indeed the various sensory modalities
      are integrated so as to produce classes of information, then the retention of information about the
      individual modalities is still desirable. Thus the system must react to the individuality, must react
      uniquely, to the particular processes at some level of each sensory modality in order to maintain
      discrimination.
50)     That is, information at the highest level of abstraction, upon final integration of the various
      sensory modalities, cannot be allowed to conflict with information about sensations from the
      individual modalities, for fast reaction times, at least. But what is the difference between merely
      labeling each modality, as a digital computer might do, and actually sensing, i.e., internally
      reacting to the processes of, each modality? It may be that there is no functional difference, and
      that qualities are an accident of the necessities of the biological system. That is, in a digital
      computer, a number or label can be associated with, i.e., activated simultaneously with, another
      string of numbers in order to assign a ―quality‖ to it. However, this does not seem possible in the
      CNS. All processes are aspects of the integration of the sensory modalities: there are no arbitrary
      numbers, labels, existing in the CNS. The obvious seeming exception to this is the abstract
      modality. However, if the above reasoning is correct, the abstract, non-sensory labels, largely
      words, are in fact the result of, dependent on, the lower-level processes. This would explain a lot
      about the nature of language: why it requires sensations, complex sensations, yet is
      simultaneously independent of them.
51)     That is, why does abstract reasoning proceed through symbols, and why are those symbols
      similar to, and part of, particular sensory modalities? Given that abstract reasoning is the result of
      the integration of sensory modalities into inter-modality classes, just as sensory information has
      been integrated, then ideally, the internal reaction to those processes must be, in effect, a sensory
      modality which is not a sensory modality. We might make two assumptions here, then. First, that
      only sensory modality qualia are available to us as reaction experiences, and second, there is only
      one functional center which does the final abstraction and processing for multi-sensory
      integration. The first assumption might be a consequence of processing or even evolutionary
      limitations: the verbal/abstraction abilities may have evolved much later than the sensory
      modality processing individuation, and thus may have intrinsic limitations, and/or have had to
      borrow. However, if this assumption is not correct, then it would seem to follow that abstract
      reasoning must employ qualia not sensory derived; that is, qualia that originate from the reaction
      to the result of multi-modal integration just as the sensory qualia result from the reaction to uni-
      modal integration. Here the example of the emotions might be relevant.
52)     The second assumption has the same reasoning relating to parsimony, speed of reaction,
      minimizing processing resources (metabolic, at least), and unification of functions.
53)     In addition, a word, as a ―label,‖ is an integration of various processes itself, not at all a label in
      the sense of a number or string in a digital computer.
54)     That is, consider a neural event as a signal, to be integrated with other neural events as internal
      signals and to produce structural changes and thus response changes in the system. An animal
      vocalization is no different; a bird‘s call, for example, is basically a signal from one set of units to
      another, transduced to and from sound, to produce the same kinds of changes and responses as
      any internally originating signal. A vocalization then is functionally no different from any other
      neural integration, but it originates in another entity. In addition, it can activate many systems in
      parallel, because of the flexibility resulting from the external origin and resulting transduction.
      But it doesn‘t always. The cries of amphibians and insects are extremely behavior-specific. Again,
      the abstraction of a symbol results from the multiple systems integration into some relatively
      small arena.
55)     There seem to be two major threads here. One involves abstraction as it relates to the difference
      between computers and human beings or between logic devices and neural networks. The
      contention is that logic devices necessarily work with abstractions, i.e., symbolic expressions
      involving symbols whose interpretations must only be in terms of classes, and those classes must
      be expressed in symbols themselves abstractions; i.e., they are never "built up" from symbols
      denoting individuals. Since they are abstractions, the codes utilizing abstract symbols must be
      interpreted in terms of individuals to be meaningful, first, because of the multiple possible
      meanings, i.e., interpretations in terms of individuals, that any given abstract symbolism makes
      possible. If there are such multiple possible interpretations (e.g., Searle's example) and no way to
      choose between them internal to the logical device, then the symbols have no internal
      interpretation beyond manipulating the logical, abstract expressions, and those symbol strings
      and manipulations can never be meaningful internal to the logic devices. Second, given that
      symbols without individuate interpretations can only be interpreted in abstract terms, i.e., in
      terms of classes, and those classes, in turn, must be equally abstract symbols, this seems to create
      an infinite regress.
56)     Thus: an "empty" symbol has no content except relative to the internal operators of the system:
      its "meaning" consists of a description of how it is operated on by the system's operators. What are
      "external" operations? What else is relevant to a symbol? If a system has a finite number of
      operators, then any given symbol, no matter how derived and complex, and only be operated on
      identically with some class of symbols; nothing but a finite set of symbols in a finite-state machine
      can be operated upon uniquely. But one could reply that the number "3," for example, is in fact an
      individual, even though it is operated on identically to all other numbers. However, in
      interpreting a computer program, etc., one would have to ask, "3 what?" Even if the answer were
      that the symbol merely stood for the number, "3" is not an individual. And in fact if the answer
      were that "3" was only a number, then it would be a class term, and the operations would be
      considered syntactical. "Meaning" must then be something that is added after the symbolic
      operations have been determined; it is not relevant to the system's operators. Is it possible then to
      speak of "semantic" operations on symbols? It would seem that semantic operations must be
      confined to "meanings"; but what then are meanings? If there are semantic operations, are not the
      "meanings" determined by these operations as empty as "syntactic" content? One merely has two
      possible ways of manipulating the same symbols. Meaning's realm, then, must be confined to
      what Searle has termed the "background." But that does not answer the question as to what
      "meaning" refers to, if not to some aspect of the manipulation of symbols.
57)     The problem here seems, at first glance, to be the symbol-referent distinction. But this is not
      actually the case. Because the nervous system works through transduction, and the nervous
      impulses do not possess the qualities of the physical events that evoke them, the nervous system is
      always dealing, in this sense, with symbols. The question is not how quality flows directly into the
      nervous system; it does not. The question is what the difference is between the nervous evocation,
      the impulse, as a kind of symbol, and the representations that we abstract from these evocations,
      which we experience as symbols, and which we have constructed machines to produce and
      manipulate. And just so, the "symbols" that computers manipulate are nothing of the kind for the
      computer; that device merely assigns, reassigns and reads magnetic field domains.
58)     Yet the question of quality is also ambiguous. If the structure of some aspect of neural coding
      corresponds to the structure of the world, its quality, to some extent, has been captured in the
      neural coding. Thus, as argued above, one can compare coding structure and real-world event
      structure, and in turn relate that directly to the phenomenology of that sensory modality. But even
      this is an illusion. The nervous system imposes that coding structure on the world; there are no
      colors nor pitches, and perhaps Kant is correct in an even more radical dismissal of
      correspondence. If this is true, then it is completely false to speak of structural correspondences
      between the world and qualia, as has indeed been argued above. But then the only
      correspondence left is that between the coding structures and the phenomenology: we create
      qualities, quite literally. But if this is true and there is no such correspondence, then what is the
      difference between our neural coding and the logical coding of computers? One can cite the
      obvious: our coding is analog in the sense of being continuous: the frequency aspect of the
      representations. However, first, this analog nature does not account for all aspects of the coding;
      neural spatial position relative to other neurons is relevant and is a digital coding dimension
      (certainly at least in the sense of not being a continuous variable, given a finite number of neurons,
      as is frequency (but for a counter-argument on position as analog, see 58)), as has been argued
      above. Second, one can claim that digital codes can be as explicit as needed.
59)     That is, the argument is that digital codes, if the strings are long enough, are as accurate as
      analog codes, given the presence of noise in the system. However, this assumes that accuracy to be
      essential. In fact, given that the analog system is characterized by the correspondence of some
      real-world aspect of that system with some relevant aspect of the coding, the noise in the system is
      just an aspect of the coding itself and is in fact something that a digital symbolism is intrinsically
      unable to include. The further question of whether this is the actual and relevant difference
      between digital and analog systems, given some of the arguments above, is an interesting one. An
      analog code, in this view, necessarily contains a temporally varying "noise" component which a)
      renders a temporally fixed digital code intrinsically approximate and b) is, whether the noise be
      considered random or not, nonetheless renders a precise digital code (and all digital codes must
      be precise because they are finite-length strings) approximate.
60)     This difference, then, would seem to come down to two things. First, the individuate/class
      distinction mentioned above. Computers seem in some sense different from the CNS in that they
      deal exclusively with classes rather than with individuals in their coding. Secondly, the interaction
      between the system and the environment. Consider the latter point. As long as an entity's "mind"
      consists of symbols interacting only with other symbols, the entity "floats," in effect ({Harnad 1990
      #73}), and the symbols' interpretations must be in terms of other symbols, or come to an infinite
      regress. But for an interaction with the world to differ in kind, given the indirect nature of the
      sensations realized as neural codes, from an internal interaction, necessitates that this interaction
      can somehow alter the symbols, their operations, or some aspect of that symbolic system in ways
      that any of the internal symbolic systems cannot. The "environment" must be surprising, variable,
      not encompassed by the limitations of the symbolism. Otherwise, that environment could itself be
      another symbolic system interacting with the others. The limitations of such interactions must be
      transcended by the environment/symbol interaction for it to be meaningful. Second, there should
      be a difference between the environment, as coding, and the internal codes. This may be, at least
      in part, the individuate/class distinction mentioned above. But neither of these is decisive. What
      are the "limitations" of the internal codes, and how do other codes "transcend" them? Gödelian
      limitations come to mind, but is it sure that sensory variability as translated into coding variability
      transcends these? The latter point, then, is not decisive. What of the former, the individuate/class
      distinction?
61)     The assumptions here are that there are, first, distinguishing characteristics of each non-identical
      individuate sensory event which differentiate it from every other sensory event, and second, that
      these are reflected in the coding of that sensory event. If this is not the case, then there will be,
      potentially at least, occasions in which two sensory events which are not identical are confused, or
      at least treated as instances of the same event, and the coding will be a class coding. But the
      problem is generalized with computers, in that first, any given symbol used by a computer can
      stand for an infinite number of particular events, and second, since computers are finite-state
      devices, there can be at most a finite number of symbols listing the characteristics of an event.
      Thus, as has been said, for computers, any finite list of class-symbols can only stand for another
      class.
62)     The other problem implied by this approach is that any individual must be symbolized by a list
      of characteristics; but the mere concatenation of terms comprising a list, even if it's elements are
   combined according to complex formulas, may be insufficient to represent a set of characteristics,
   and at least not what the CNS does. But when this is compared to the coding in the CNS, the
   obvious difference, that of the frequency dimension of the CNS coding, which produces an analog
   component to the codes and thus a potentially infinite possible number of symbols, seems only to
   apply to some aspects of the codes; other aspects are seemingly, at least, digital: finite-state: i.e.,
   arrays of neurons whose location (in the array) codes properties. However, even these are analog
   in the sense that there is never a single neuron firing, no matter how specific a property; there is
   always a finite probability, on a changing analog scale, of other neurons firing. Is the presence of these
   analog components sufficient to differentiate the codes?
63) Let us approach this from another direction. What is the difference between symbols and
   "things"? A symbol as neural code is much more like a "thing," i.e., an object in the world, than it is
   like a symbolism… another Kantian implication. The reason is that we deal with, react to, neural
   processes in the way we react to "things": we cannot assign meanings to them (as we can to
   words), we have to learn (rather than assign to them) their characteristics. That is, the difference,
   really, between a "thing" and a "symbol" seems to be the same as that between a "neural process"
   and a "symbol," because neural processes, "codes," as I have been using the term, are not arbitrary.
   We have no choice in how we react to neural codes, in part because we are those codes, in part
   because (and this is the same thing) the codes are part of the physical structure of the nervous
   system. The transduction that occurs between physical event and neural code is not one of
   symbolism or representation, but of true physical conversion, i.e., of one dimension of a physical
   dynamic into a corresponding dimension of (physical) neural dynamic. What then is, really, the
   difference between our CNS and a rock, for example (see 16, above)? For one thing, transduction
   does take place in the CNS and not usually in a rock (although piezoelectricity is a
   counterexample). Is it then no more than complexity? One cannot have it both ways; if Searle is
   correct, then there is no difference beyond complexity between our CNS and a rock. If Searle is
   wrong, then computers can (potentially at least) understand. But there are two differences
   between neural codes and physical conversions: first, the codes are different substances; second,
   the codes are variable, i.e., we learn.
64) It is the first distinction that seems to get to fundamentals. Neural codes are all identical in
   physical substance: they are ionic movements across membranes; and in addition, they are
   identical in type: they are frequency/spatial impulse patterns, no matter the sensory modality. But
   it is exactly that modality which, in a rock, is preserved, in effect, even if the vibrations are
   transduced into electrical fields they are perfect physical analogs of the original. Neural codes are
   not merely different physical entities, they are structured, at least at the base level, i.e., the level of
   individual impulses, differently than the physical events. That is, light is EM frequencies,
   transduced at the retina into molecular absorption spectra. Neurally transduced it is the frequency
   of bursts of ion movement through the membrane of axons, or temporal patterns of depolarization
   on the cell bodies resulting from the axon firing. There is no correspondence here between the
   physical events: either the EM waves or the EM absorption spectral distribution; except at a higher
   level: between patterns of neural firing where frequency and location corresponds to color
   absorption intensities and the EM absorption spectra (although there is correspondence between
   intensity and firing frequency). But that frequency/location coding could have been coding for,
   say, sound spectra, given different sensory organs. In that case, 1) the sound would, I believe, be
   seen, quite literally like we now see light; 2) nonetheless there is an indication here of the
   arbitrariness of the coding. We might see the sound, but it would still be sound, with
   characteristics different from light, at least in the way it interacts with the world, which we would
   perceive.
65) That is, the sense of "could have" in the above is a fairly exact one. The neural impulses for
   sound and light, below the level of patterns of impulses, are exactly the same. This comes in a
   sense down to the question of interpretation. A mechanical transduction is not an interpretation.
   But what does that mean, exactly? There is a sense in which the transformation of mechanical to
   electrical energy, deterministic, similarly structured, or not, is an interpretation of that mechanical
   energy. In the neural system, however, the process of abstraction and gestalt formation is
   different. That is, interpretation in the CNS consists of the universal process of the formation of
   higher-order abstractions that occurs throughout the CNS. The first stage, then, is pure mechanical
   transformation: that of, say, the light into molecular energy, then into neural impulses on the
   retina. However, at the point at which those impulses, through, say, center-surround focusing,
   produce neurons firing at edges, that focusing has created the edges quite literally by abstraction,
   arbitrarily, from what is probably an infinite spectrum of choices for basics.
66) Now one interesting thing here is that despite the creation of an edge and thus the forcing, in
   effect, of the system to see edges, there is, from only the firing of the "edge receptor" which is next
   in the chain, no way to tell, from outside, that this is an edge receptor. That cell, taken by itself,
   could be specific to any kind of stimulus, or not. First, how then does the system, the entity, know
   that this is an edge receptor? Second, what is an edge? Why is it seen as it is? Clearly there must
   be some sort of pattern integration here. Let us look into this: an edge is a boundary between dark
   and light areas. Now, dark and light, at the retinal level (beyond the EM stimulus) can only refer
   to mass firing or lack of it. So there must be an integration here between the firing at the
   transition, and some recognition of what the transition is of: dark and light, resulting from some
   other neurons integrating light levels. We could (and do), for example, also have edges between
   colors, with identical light levels. Color edges and intensity edges then must be differently
   constructed, and that construction (or some result of it) is essential to the experiencing of those
   edges.
67) But we can keep pushing this back. Suppose there is a cell that measures intensity, and it fires
   simultaneously with the edge receptor, so we "know" what kind of edge it is. But how do we
   know why the intensity measurer is firing, what it represents? The only way around this, it seems,
   is to monitor the environment and correlate that with the firings. But of course that is impossible,
   or leads to an infinite recursion. Suppose however that the environment was just exactly the
   surface of the retina. What that would imply is that data from the earliest firing of sensory
   neurons must be accessible, even if in some sort of attenuated form, throughout the system. Or
   that we are only aware at the very last stages of the processing, and experiences are arbitrarily
   assigned at that point. But this latter seems impossible, given the close ties of experiences with our
   sensory inputs, even given the lack of awareness up to, say, area IV or so. That original data could
   be retained through the preliminary areas and carried forward.
68) But the individuate/class distinction seems unclear, even spurious. That is, it is clearly the case
   that any sensory system, even an analog one, has limits to its resolution. Thus there will be
   instances and areas in which sensory events that might be distinguished by a more sensitive
   system or one attuned to some specific characteristics are not. Thus all sensory events are in this
   sense class events, and all sensory codes class codes. But is this the actual distinction between
   symbolisms? For analog sensory receptors, there is always an attempt, at least, to distinguish or
   name separate events: each distinguishable sensory event has initially, as it is sensed, a proper
   name, in effect. Whereas for digital devices, initially there are no proper names, except as lists of
   class names. Even if the proper names given by analog devices are inaccurate or identical,
   although they are not supposed to be, they are regarded as proper names, i.e., unique identifiers
   of particular sensory individuals, whereas the digital elements can only be class names. But given
   that the system consists solely of neurons and neural codes, what is it to "regard" a symbol,
   whatever that may be in some specific case, as a proper name, i.e. as unique? It must be that this
   "regard" is in fact the reaction to the particular code, the input, to some neuron or connected
   group of neurons (see above). If that is true, then each reaction must be different from every other;
   this seems possible, but is there some theoretical reason for believing that this actually happens?
   In fact there is; first, given that any cell membrane is a) constantly in flux, b) altered by below-
   threshold depolarizations on its surface, so that it does not, after these depolarizations, return to
   exactly the same resting state, c) altered by super-threshold depolarizations, i.e., discharge, then
   no cell membrane, even if presented with two identical inputs after sufficient resting to return to
   "normal," will respond identically to the first and second input. It is, then, the analog nature of the
      cell membrane response (depolarization over time and surface area) which enables all codes to be
      proper names, in effect. However, does even this assign quality, meaning? The unique response to
      each code event may insure that the system is not finite-state, either in symbols or operations, but
      this does not seem to be something that moves beyond syntax, which now varies in analog
      fashion, to semantics. Thus here the individuate/class distinction seems spurious.

69)   The second thread involves coding differentiation as it relates to specific sensory modalities, and
   the necessity for fixed, i.e., organically constructed, processing and interpretation of the codes.
   This fixed processing would seem to be the only route, given the above, to insuring, first, the
   separation of the various modalities, and second, the interpretation of the codes into qualia with
   structure corresponding to the coding structures. A criterion might be as follows: given that the
   logic followed by a symbol manipulation system would be equally applicable to either following
   the stock market or to language processing, a la Searle, what would it take, as far as neural coding
   is concerned, to differentiate those two applications of the symbols? We can go further and take
   the same logical or mathematical symbols and apply them to two different situations; analytic
   geometry, for example, may be applied either to maps or to architectural drawings. Let us assume
   that the same figures and operations are used. While the computer does not differentiate because
   it does not need to for the purpose of this example, what in the neural code does? A simpler
   example: suppose the analytic geometry was of either shaped color patches or of uncolored stick
   figures. This would seem to come to a combination of the fixed processing, and the
   individuate/abstract differentiation above.
70) Now there is an interesting implication in this for Searle's "background," which is that since
   digital devices work with symbolisms whose referents are classes and not individuals, as argued
   above, the background is an appropriate concept here, when differentiating two (of the potentially
   infinite number of) applications of such symbolisms. The background cannot be, by definition,
   included in the operations of the symbolic system, nor, similarly, can the differentiating factors
   (e.g., colors vs. lines) between applications of analytic geometry. If these factors are included,
   there has merely been created another symbolic application, and further differentiating factors
   within that application, analogous to the above, which alter none of the operations, can be found.
   But Searle's "background" operates in precisely this manner. It is thus suitable as a description and
   criticism of both the digital and the analytic approaches to symbolisms, devices, and logic; and
   appropriate as a criticism of Searle's own system, as he is (perhaps) aware. However, given a
   system which employs analog neural nets (not necessarily confined to biological systems), in
   which individuating information is retained throughout processing, the "background" could no
   longer exist per se. All possible differentiations result in different symbolic processes, since all
   operations on classes are tied with operations on individuals. To demonstrate this would in fact
   show a decisive difference between computers and analog neural nets, since the computer then
   would be in the situation of a class symbolism processor which in order to successively
   differentiate individuating characteristics, could not operate with approximations, yet which had
   no choice, because of those processes, but to use them. The question is then how, in analog neural
   nets, abstractions, i.e., class symbols, can be formed which retain such individuating information.
   The question then recurs: just what is individuating information? That is, in effect, just what is an
   individual? The only possible answer is that every possible input, whatever an input is, must be at
   least potentially distinct. This is clearly not true in the digital system; every letter on the keyboard,
   for example, is equivalent in each successive processing of that letter; restating the name of a
   variable does not alter it in any way. The only way a system can know whether information is in
   fact individuate is if current and historical responses are compared in some fashion. If there is no
   match between these, then the system must always extrapolate, approximate, or interpolate in
   some fashion. That comparison is implicit in analog neural nets, since each past event is saved, in
   effect, in the state of the system, and in an analog system each past saved state can be unique. But
   it does not seem necessary, except for the purposes of analysis, that the system actually "knows"
   explicitly that there are individuate symbols as well as class symbols.
71)  Suppose that this information did not merely specify unique responses and results of processes,
   but that those responses were structured according to the particular sensory modality and in
   addition the processes were constrained to those peculiar to that modality. We have seen that this
   is sufficient for at least some of the sensory structure being contained in qualia.
72) To return to qualia: no neural impulses have greenness. This and all qualia, then, must be the
   result of some sort of internal response to the coding for green in the visual system; a response to
   an information or coding structure. "Greenness" here is therefore taken to name not merely the
   quality, but some at least of the retained individuating information of a given sensation or sensory
   input. "Greenness" is a class name, standing however not for a general quality, which cannot exist,
   but for the name of the generalization, the abstraction, over individual instances of similar qualia,
   in this case, greenness. They are similar because the system is reacting to that pattern similarly.
   Why does the system know that it is reacting similarly? We cannot as yet answer that question,
   but if it were not the case, then the abstraction, the class formation, would be impossible, and it is
   evident that it takes place. But the quality of greenness is not an abstraction, and thus there is no
   such general quality, but only the green similar to every other individual instance of green (pace
   Dennett). Sensory qualities such as greenness, then, would seem to be indications of and instances
   of individuating information.

73)  Now let us look at another aspect of this picture. Where in all this is consciousness? One might
   look at all these sensory processes as mere processing, refinement of the picture, so to speak; the
   addition and construction of components: the angles, lines, colors. Similarly, if we want to bring
   consciousness in with the "body image," we have the same situation: what goes on in the
   cerebellum, the motor cortex, and so forth except the same kinds of refinement of kinesthetic and
   related information, with in addition effectors for processing? The integration of vision and motor
   control, for example, can, in this picture, result in no more than the mechanical synthesis of two
   forms or aspects of information about the world, and concomitant responses (thus LaBerge is
   refuted). The RAS serves to activate this processing, but again, there does not seem to be any need
   nor indeed any way that consciousness and/or meaning in anything like Searle's sense (however
   clear that is) can be generated here.
74) The only place that the actual data can come together is in the prefrontal cortex, which also has
   input/output from the thalamus (but see Taylor on the parietal cortex: another strong possibility.
   Note the injury data: one can have consciousness with great injury to the prefrontal lobes, whereas
   injury to the parietal lobes is very critical; similarly with injury to the ERTAS, in fact. But I will
   refer to prefrontal as the source for consciousness below as a shorthand.). Thus the seat of
   consciousness may be there. The activation role of the RAS is just that. But what exactly is going
   on in the prefrontal cortex that is not in the cortex? First, there is multimodal/emotional/bodily
   integration. This is where the modalities stop refining and abstracting individually and continue
   that process multi-modally, in combination. Thus functions can be generated which are
   combinations of different processing patterns. Second, this integration must be controlled by
   feedback. That is, its extent, amount, and focus or spread over the modes must be controlled.
   Third, multi-modal integration for the purpose of integrated processing, i.e., processing taking
   into account all types of "information," would seem to imply, at least at the highest levels, one
   "center" of integration. That is, even if modes were successively joined, either that succession
   would have to eventually encompass all, or that succession would be a series of parallel
   integrations leading to one end result. There is no other way to insure lack of contradiction,
   integration, of response.
75) The feedback loops must be the key here. There are two types of refinement, then. One consists
   of the successive processes of refinement and synthesis described above, where components are
   generated, integrated and abstracted. This can be activated or inhibited, mechanically (i.e., in
   theory, it seems, without consciousness), by the RAS. The other must consist of refinement,
   modification, and control by internal and external feedback. The first operates employing the
   balance of activation and inhibition described above to create abstractions and boundaries, but
   this is a linear series, with bottom-up processes. The second must use top-down processes (and
   sideways, e.g., from parallel cortical columns) in order to modify and stabilize the results of the
   former. Stability of structures and components implies stability at all levels of abstraction; and
   conversely, adaptability implies modifiability at all levels of abstraction. The feedback then must
   consist of a balance of negative and positive feedback, just as the abstractive processes consisted of
   a balance of excitatory and inhibitory. Whereas the first generates static areas of response and
   processing stability, i.e., the components of constructs, the latter must generate dynamic
   components and areas of stability, which must include both high-order structures and the
   dynamics of their interactions: patterns of response but also patterns of construction and
   association.
76) Now the feedback can come from within; i.e., the maintenance of stability can be internal to the
   construction, i.e., from the response processing pattern which is the equivalent of a particular
   component, or from without, as above, from processes generating higher-order abstractions. The
   former certainly seems possible, but unfruitful in the generation of meaning or consciousness. But
   the former processes, i.e., internal stabilizing feedback, do not need external models; the processes
   themselves are their own reference points. External stabilizing feedback, however, must refer to
   models or reference points outside of the process being stabilized. This implies that in some sense
   the processes are duplicated or retained in the higher-order processes, in order to have reference
   models. But this cannot be the case in detail throughout the system, or the whole cortex would
   have to be duplicated. That implies that the low level processes, at least, must be, in effect, hard-
   wired, which is supported by developmental data, and not only for visual processes, but for all
   sensory and even verbal processes (see also Regier). Yet given the plasticity of the cortex and the
   considerations above about the retention of individuating information, something of the earlier
   processes must be retained at a higher level, to be employed in the stabilizing feedback function.
   As the degree of abstraction increases, however, so does the plasticity of the structures, and
   indeed it is likely that their size in terms of the size of the cell assemblies, i.e., the cell-body
   equivalents (see above) involved also increases, particularly if sensory mode integration is
   involved. But the plasticity must exist, then, at the edge of stability, and must therefore have
   stabilizing feedback, from both within and without (after writing the following, I read Grossman's
   1980 article).
77) How would the feedback stabilization work? Let us take the case of the perception of a square;
   this necessitates the perception of corners and of equal sides, where the corners are at right angles.
   Now this latter necessitates, given normal cortical structure, some set of cells optimally firing
   when a right angle is presented. Presumably, nearby cells fire at slightly different angles (in other
   cortical columns: 2-dimensional arrays made 3-dimensional by processing) and inhibit the cells
   with right angles, and are inhibited in turn by those cells, depending on the presentation (former
   for non-90, latter for 90). Now what if the 90-degree firing cells (9dc) drift off slightly? Then they
   will fire when some other nearby set fires and each will inhibit the other; one will win. But in this
   case, a 90-degree angle is either seen inaccurately, not differentiated from another angle, or both
   are seen as 90 degrees. In either case, there is an error in the system; how is this error corrected
   for? Internal feedback has failed, and parallel feedback also; we assume that bottom-up processes
   cannot correct this error because they are the source of the error, or at least irrelevant to it. The
   only thing left is top-down correction. How can this work? What is above is the synthesis of the
   square as a whole, let us say. But that synthesis has already taken place, and if we assume
   selective top-down stimulation, rather than parallel inhibition, we are in a situation in which this
   stimulation feeds down (because of redundancy and multiple activations through other
   characteristics, e.g., equal line lengths) to the neuron that should fire, and perhaps inhibition to the
   neurons that should not fire, only when the square is presented, correcting the erroneous neuron.
   So this kind of correction can only come from a cell or set of cells which have synthesized
   information from below, because of the necessity for redundancy. This is, interestingly enough,
   also the equivalent of top-down construction of components. That is, given that the higher level is
      set for some gestalt or synthesis, and the lower unbiased, the above processes can bias the lower,
      thus in effect constructing the stimulus components from its abstractions.
78)      The top-down corrective processes, then, are stabilizing and constructive processes, while the
      bottom-up are constructive, but potentially unstable. We might speculate (see ―locking-in,‖ below)
      that consciousness stabilizes attention, and is self-referring because there are no higher levels, and
      so the system at the highest level must be self-reactive ("higher" states of consciousness as
      attempts to create higher-level stabilizing processes?) and self-stabilizing.
79)      Given this overall picture, the idea of an abstract symbolism seems difficult to attain, even far-
      fetched conceptually. That is, if this neurological picture is correct, then "abstract" symbolism
      must be based on this successive abstraction and synthesis of the various input modalities (which
      must of course include internal inputs); the use of inhibition and stimulation to create successively
      abstract components, and the use of inhibitory and stimulatory feedback to stabilize and give
      flexibility to such components. What are the assumptions of this neurological picture? The first is
      that both inhibition and excitation are required to create abstractions through the creation of
      boundaries and thus of bounded areas. The second assumption is that these processes, elaborated
      from bottom-up linear processes to top-down non-linear processes, are universal and recursive
      within the nervous system. The transduction issue, as it relates to the meaningfulness of abstract
      symbolisms, must then be reconsidered, and even reversed. The problem of symbolism now seems to be the
      question of how symbolisms can possibly become abstract.
80)      And indeed the history of language in human cultures seems to indicate that true abstraction
      and further, the idea of true abstraction, is extremely difficult for human beings. That is, words
      have not originally been seen as arbitrary, but as inevitable; not as symbols, but as parts of what
      they symbolize (e.g., Frazer); and the same, originally, with numbers. The Searlian questions then
      seem irrelevant, given the neurological system realizing the symbolisms; the Searlian
      "background" a construct resulting from the misapplication of the analytic idea of abstract
      symbolisms and operations. The above picture seems if anything to approach the Peircean concept
      of mind and thought, and be still beyond that of, for example, Andy Clark's attempt at integration.
81)      We now ask the next easiest question leading to consciousness, the Searlian question: what is the
      essential difference between computers and an analog neural net? At this point what seems to be
      one essential difference is the hard coding and inalterability of the symbols in the digital
      computer. That is, in any given program, 1s and 0s, and all higher-level symbols, are either
      unchangeable, or change according to rather simple and definitely well defined rules. In contrast,
      in an analog neural net (ANN), the symbols are constructed from the ground up. Alternatively, if
      the base neural codes are considered to be symbols in some sense, then the meanings of those
      symbols, i.e., the parameters of their interaction with other such symbols, are constantly changing
      according to context, with only general boundaries or parameters of application and action. That
      is, any train of neural impulses impinging on a cell body (indirectly when mediated by a
      particular neurotransmitter), when repeated, has roughly the same local effect, i.e., depolarizes the
      membrane at a particular synapse the same amount for the same length of time. That effect,
      however, the "meaning" of that "symbol," cannot be the same unless the whole context of that cell
      body is the same, which may be the case. But it does not have to be the case, whereas in a digital
      computer it must be the same. There are no symbols with fixed meanings in the ANN, only
      symbols with meanings that vary within certain parameters. And we may assume further that
      those parameters may in some cases be the equivalent to (or identical to) attractor basins, so that
      the "meanings" may widely vary (as the basin changes) given proper circumstances. For example,
      if it is not a single cell but a set of cells, as described above, which is the equivalent of a single cell
      membrane, then much more variability in response over that set to a particular train of impulses is
      possible, even if the train is to the same neuron in the set, than with a single cell.
82)      The effect of the neural codes, then, is always at least potentially individuate. There are at this
      level no class symbols, as there are with digital computers, since with the latter a variety of instances
      of the same symbol must produce (rather than by chance produce) the same response. This really
      seems to be the bottom line; there is no way around this for a system to remain a digital computer;
      and this does seem to capture something of the individuate/class distinction above.

83)  Let us now consider further the problem of qualities. As I pointed out above, there are no red
   neural impulses. Responding to Chalmers et al, there is also no red information; for that matter,
   there is no red in the world at all. There are neural impulses "coding" for frequency (actually, for
   absorption spectra of the retinal receptor molecules), and similarly, perhaps, for "information," but
   the problem of qualities remains. That is, how does one go from quantity, in effect, i.e., patterns of
   basically identical stimulations (however complex they are) which are the equivalent of numeric
   patterns, to qualities? If indeed that is the extent of the situation, there is no alternative than to
   have the system which interprets: responds differently to different patterns: a particular function
   and the response to that function is a particular quality. The questions then are 1) what is a
   function, and 2) what is a response?
1)   Is relatively easy to answer, given the analysis above. Successive functions are the successive
   syntheses of responses, building abstractions. However, to become single functions, the actual
   synthesis is necessary, i.e., single cells or closely linked cell assembly (CA) must be responsible for
   simultaneous integrative processing of the multiple inputs to be synthesized. To be a "cell
   assembly," a group of cells must perform this integration analogous to the way in which single
   cells perform it at lower levels, i.e., spatial and temporal integration of the incoming signals must
   occur. In order to have this integration over a group of cells, they must receive the impulses nearly
   simultaneously, must perform their individual integrations, and the results of those integrations
   must be transmitted to others in the group while those others are performing their integrations:
   the group must be linked through multiple feedbacks within a short processing period. Then the
   outputs of that CA must be treated by the next higher stage as single entities, i.e., integrated by a
   higher level single complex in the same manner. That "single entity" is a single function,
   analogous to, say, the integration of joined lines so that preferential angles are isolated. The
   preferential response of the neuron to a given angle is one thing, the integration of inputs on the
   cell body surface resulting in that response is another.
2)   What then is a response? How can there be a response, for example, to the integration of inputs,
   as above, i.e., to the function: the spatial/temporal pattern on a particular cell body or the
   extended cell body of an assembly? Surely there can only be a response to a response: to the result
   of that integration, the neural firing? A) There can only be a response to the final firing; B) there
   can be a response to the pattern on the individual cell body surface. If (A) is true, then in order to
   synthesize a function, the whole group of neurons integrating the possible angle responses must
   be simultaneously monitored, so that each angle is not merely coded by the response of a
   particular set of neurons, but that set is placed, so to speak, in the context of the responses of the
   whole set of possibly responding neurons. That is, a single cell assembly must respond to the set,
   previous to its integration on the single cell body, as a particular function. If this is not the case,
   then the functional entity: the function, per se, is lost. Simultaneous monitoring of the whole
   group is necessary, and a response to the resulting function must occur, not just to the neurons
   actually firing. B) The pattern is monitored, perhaps through dendritic back-propagation.
84) The creation of the cell assemblies (CAs) necessary for the formation of high-level abstractions
   must result in groups of cells which are the equivalent of the single cells which respond to the
   complex stimuli as described above. The sense in which they must be equivalent is, of course, a
   functional one, but there are also necessary structural parallels. That is, on the surface of an
   individual cell body are in many cases literally hundreds of inputs, continually interacting. Now
   in the case of an assembly of cells, the interactions between inputs on different cells cannot take
   place through the same mechanisms; the cells must communicate through spike discharge
   frequency. The critical component of this interaction in order to consider a group of cells the
   equivalent to one large cell is the timing relating to the characteristics of single-cell pre-firing
   depolarization. That is, on the surface of the cell body, before the spike discharge, the multiple
   inputs are interacting through subthreshold potentials spreading on the cell surface. The critical
   component of these impulses for the purpose of this discussion is the feedback between a given
   area of depolarization and the source of some other area; given this (hypothetical) feedback, the
   cell body surface becomes a complex of dynamic charges capable of chaotic behavior (and recently
   I found out that metabotropic receptors have effects that last, not milliseconds, like ionotropic
   receptors, but seconds to minutes). Now to duplicate this functionally in a CA, the spike
   discharges which communicate between the cells in the assembly must be close enough that
   feedback from the receiving cell (cell B) is capable of being itself received at the initiating cell (cell
   A), while cell A is in the process of communicating with cell B. There must be, in other words, the
   capability of a simultaneous two-way communication between any two cells in the CA in order
   for it to be functionally the same as a single cell.
85) Thus this may in part be defined by temporal boundaries; i.e., in order for the necessary internal
   feedback to take place within a reasonable time frame, one in which effects of previous neural
   stimulation have not vanished (the neural relaxation time), the time intervals from mutual inter-
   stimulation must fall within some limit, although sub-areas of the assembly can remain constant.
   The latter case implies, in effect, that large assemblies are made up of arrays of smaller assemblies,
   in which stimulation interval boundaries fall within the cell-body (note the distinction between
   "cell body" and "cell-body") decay range. In that case, one can ask whether the temporally
   extended sub-assembly stimulation intervals are in effect geometrically assisted, i.e., the physical
   extension of the cell-assembly creates a "ripple effect," in which stimuli move outward (or across)
   to interact with another assembly at their boundaries, so that the interval from the center or
   opposite edge, interacting with another opposite edge, may exceed the individual cell-body decay
   time, which is equivalent to the maximum propagation time within a cell-assembly. This case
   seems doubtful, however, if the cell-assemblies are attempting, in effect, to form a super-assembly,
   since this super-assembly must have some differentiating characteristic of this sort from any
   group of functionally related discharges. This places, however, an upper limit on the size of any
   cell-assembly, given by neural cell body decay time. But again this latter time interval is not clear;
   is it defined by the interval from the onset of stimulus to the elimination of all effect from that
   stimulus on the cell body, or to the elimination of all effect from all its interactions with the
   members of the cell assembly? If the latter, then the assembly boundary may be quite large. The
   functionally defining characteristic, it seems to me, must be the above-mentioned ability to have
   the equivalent of the feedbacks on the individual cell bodies, i.e., feedback which can alter the
   input as it impinges on the cell body. Without this, the system, it seems, is merely linear, i.e., it
   does not interact internally similarly to the interactions possible on a single cell body.
86) But more generally, this kind of response can be conceptualized, at this initial stage, as an
   extension of the top-down stabilization process. Given that process, the necessity for a response to
   an overall pattern, a function, follows from the variability of that function's neural realization.
   That is, if different populations of neurons within one person (because of cell death, growth,
   general variability of responding necessities) must carry out the same function (e.g., visual
   processing to arrive at the identification of similar objects similarly over time) then to achieve
   within-system stability, that function — that population processing dynamic — must be
   monitored and stabilized. The point is that the previous example of stabilization, where single
   cells were monitored, is not sufficient, because of the above variability and the necessity for large
   cell assemblies to carry out higher-level functions. Thus the function, as a whole, as process and its
   result, must be monitored and stabilized, as realized on a CA. This is, I claim, the first step toward
   the creation of qualia (a "functional stability" (FS) theory).
87) To recapitulate: the top-down feedback stabilizes functions. In the simple case of visual
   processing, in order to stabilize a square, for example, the feedback must monitor the square's
   components and correct drift in their responses. But the function which accomplishes this is, in
   effect, the function describing the square, because not only will the drift occur, but compensation
   for that drift must occur not only through this simple kind of stabilization, but through the
   recruitment of new neurons, and/or new connections on the same neurons, because of the
      inevitable changes resulting from death, metabolism, related memories laid down. Thus inherent
      or implied by the top-down correction is the realization of the functions that are being stabilized.
88)     But what then is a function, in neural terms? It cannot consist of some sort of monitoring of the
      actual neurons firing; there is no way that a neuron is ―aware‖ of this; a neuron is only aware of
      the inputs impinging on its surface. Thus the function must be the functional identity of a set of
      inputs on the surface of the later neuron; this can compensate for cell death, movement, etc., of the
      neurons at the previous stage: as long as the pattern of activation on the later neuron remains the
      same, no matter how it is achieved, that function is constant. Extending this to CAs, the function is
      the functional identity of a set of cells. What is "functional identity"? This can only be defined in
      terms of the set of cells that the higher set stabilizes. That is, there must be feedback to stabilize,
      and whatever it stabilizes defines that feedback pattern.
89)     But this opens up several questions. Why is any particular pattern stabilized? What stabilizes the
      stabilizing set? And, in addition, what changes the stabilizing set, so that it can 1) resist, for
      example, cell death, and 2) incorporate functional and adaptive changes? The latter would include
      attentional focusing. Answers to the first questions can be constructed reasonably easily, in terms
      of forward and backward feedbacks, for example, even in local terms. Local answers seem
      however at least incomplete, given the complexity of the system and its necessities; we are, after
      all, ultimately trying to take into account multi-modal integration. But reinforcement from other
      modalities does not seem an insurmountable problem in terms of stabilizing functions. Indeed,
      this is necessary to form inter-modal conceptions.
90)     That is, integrating visual and tactile information, for example, so necessary to deal with the
      world, merely in order to have top-down stabilization of the components, would seem to
      necessitate cell assemblies, in the above sense (81, 82). That is, how would single "grandfather"
      neurons manage the feedback to stabilize the hundreds or even thousands of cells presenting the
      various aspects of an inter-modal concept, especially if they are scattered over the cortex and even
      lower? A cell-assembly would seem absolutely necessary for this. The meanings of words, the
      integration of multi-modal information to the verbal, etc., symbols, would seem also to require
      this. In fact, the structure of languages might have some relation to the necessities of the multi-
      modal structural constraints given our normal sensory (etc.) modalities.
91)     That is, if we are putting together tactile and visual information as evoked by a word, the
      integration of those two modalities must operate according to some constraints, which may be
      reflected in the word's functional structure. That is, the processes of abstraction within any given
      sensory mode must reflect the structure as objects are built up in that mode: visual objects, for
      example, have certain structures based on the allowed types of combinations of elements, and so
      do tactile objects. What happens when these are combined? A cube, for example, has corners,
      which are sharp, not wobbly, usually. And so this is probably automatically integrated; whereas
      the sharp cannot be integrated with the sides. What determines and stabilizes this normal kind of
      relationship is fairly easy to guess at, but what of abnormal ones: it is extremely difficult to think
      of a cube with wobbly corners without some other cue to indicate that, like a cloth texture,
      because of the host of associations between smooth texture and hardness for a cube; but it is
      possible to encounter a soft plastic cube, for example, which looks sharp but is not, and to think of
      that cube.
92)     One interesting thing, here, however, is that it is very difficult for "normal" people to change
      these standard interactions without conscious intervention. Dali, however, and the liquid watches,
      for example, probably did not consciously combine these in this way, merely one day had that
      thought. Is this abnormal multi-modal functional patterning something necessitating
      consciousness? Why?
93)     Let us look at the necessities for the formation of abnormal patterns, especially multi-modal
      ones. We assume the existence of the above stabilizing functions, and in addition of some sort of
      associative functioning, so that remembered patterns, reinforced patterns, operative patterns, are
      easily evoked and then once reinforced in some manner are subject to forward/backward
      feedback stabilization. The system, then, holds itself stable, and so it should, since the
      environment is, mainly, stable.

94)  Let us consider more deeply the basic question of why we have qualia. All the above leads to the
   conclusion that we should have some sort of representations of the environment; but none of this
   answers why we are not zombies: why we experience these representations. Let us briefly
   speculate on the nature of feedback, assuming for the moment that consciousness has to do with
   feedback loops. Starting with the loops that stabilize the system, as described above, what can we
   say about any necessities implied by feedback? In order to generally analyze feedback, we must
   attempt to differentiate between types of feedback in terms of structural considerations. That is, if
   we take a simple negative feedback loop, as in a thermostat, I will assume, first that this cannot be
   conscious in any sense, and second, that nonetheless this leads toward consciousness. I have
   described recursive feedback loops in the "Ideas" document, in detail. But again there is only
   speculation there that this must be the basis of experience; there is no necessity. In the process of
   analyzing types of feedback, we look at linear feedback sequences and must conclude that these
   cannot give rise to consciousness; they are just thermostats in series, in effect. The same argument
   goes for parallel feedbacks; these are merely loops working simultaneously. I then concluded that
   we must have recursive loops to instantiate consciousness. However, what exactly is the
   functional difference here?
95) In serial and parallel loops, there can be no transformation of the information: no change except
   quantitative, no qualitative change or abstraction. Yet we have seen that the essence of the
   processes in the CNS is just this process of successive refinement and abstraction. If we have, then,
   a feedback process in which the result of the loop forms a component of a more abstract entity,
   then we have transformation analogous to that in the CNS neural processes. What we saw there
   were processes integrating neural responses to inputs which incorporated differential receptor
   fields. The synthesis of the different responses, of a single neuron, to a stimulus impinging on the
   differently responding parts of its receptor fields resulted in various focusing and abstracting
   effects on the part of that neuron. Suppose that we contemplate a feedback loop consisting of
   multiple loops itself, a recursive loop, so that those sub-loops are, in effect, the receptive field of
   that higher-order loop.

96)  Let us take a step back. We have stabilization processes on neural abstractive processes, top
   down, that fix the function which is the result of the abstraction process. That is, this process is the
   equivalent of creating a function on the input to the neural group which results in a single or set of
   cells that respond according to that function, embodying it, so to speak. Now this process
   continues through the cortical columns, forming functions, which are successively stabilized.
   When the column comes to an end, what happens? The result of the abstractive/focusing process
   must be then sent to another module. That is, we must combine shapes, then color and shape, for
   example, after each has been processed singly. This combination must be processed analogously
   to the processes leading to it, i.e., there must be a set of cells responding to particular
   combinations of firings in order to create multi-modal functions. Actually here we need to talk
   about "intra"-modal functions, i.e., those within a sensory mode but involving functions which
   arise from different cortical processes, like shapes, or colors, which belong to the visual mode, yet
   are, within that mode, different sub-modes, in effect. Thus intra-modal combinations must come
   first, to integrate the modality, then we must have inter-modal functions.
97) Thus there must be areas of the cortex specialized in intra-modal integrations, then in inter-
   modal ones. It seems likely that we do not do this as much as the initial intra-modal processing,
   before the integrations, on the whole, just because of the lack of cortical volume at this point. Now
   what is the qualitative difference, if any, between this inter- and intra-modal integration and the
   initial intra-modal processing? There does not seem to be any need for a difference here. That is,
   the above speculations about the origins of consciousness being in this kind of integration seem
   wrong. However, the origins of language, or at least of the symbolic functions which must be
    dependent on the inter-modal functions we create, would still seem to be this integration. What is
    there to go on, except extensions of the function forming and stabilizing processes described
    above for intra-modal processing? And if this is true, then the complex feedbacks, which I
    speculate are necessary for consciousness, will not arise from this. From what, then?
98) We might think about the parallel to serial conversion necessary for dealing with the world.
    Without considering "information capacity," which seems a red herring, we can still see the
    necessity for overall focusing simply because we must make choices and cannot behave in two
    different ways simultaneously, nor select contradictory things simultaneously. That is, the
    "limitations" of consciousness in terms of linearity, singling out something, are probably due to no
    more than behavioral necessities; and this consideration fits with the embodied viewpoint. But
    whatever the reasons, it is the case that many parallel streams must be switched between and
    some chosen for further processing, some for responding. Here we have a continual dynamic of
    sensory filtering and selecting, based on present and past biases, in which the different functions,
    described above, must be selectively aided and inhibited. This can only be done through feedback
    control. This then is a level above the stabilization level described above. That is a fairly strictly
    negative feedback level necessary to maintain local functional integrity. But the level at which
    choices must be made as to which functions to emphasize and which to inhibit must be a
    combination of positive and negative feedbacks. In addition, internal functional control must be
    exercised, that is, the elaboration of functions must be aided or inhibited.
99) This latter is an interesting sidelight. What does the "elaboration" of a function consisting of
    focal, say, processing, consist of? More cell (sets) recruited to refine the field resolution? This must
    be a parallel recruitment, of course, since columns are limited in height. Here is where Edelman,
    or perhaps Baar's global workspace comes in, since we might consider that there is "competition"
    for these neurons (see ―locking-in‖ below-225). However, this seems a misleading conception. If
    what is going on is a local rather strict negative feedback stabilization of neural functions, then the
    global feedback, the function of the RAS, needs to counter this. This supports the view of the RAS
    as a stimulator to a system that would otherwise shut down. In fact, given this picture, any kind of
    creativity and flexibility would seem problematic. The only contradiction to this might be the
    animal studies showing the drive to seek novel stimuli. This could also be seen, however, as a
    counter to the prevalent negative feedback.
100) The problem with Baar's and similar systems is their passivity, the bottom-up character of the
    system, even given lip service to top-down considerations. That is, given the speculations above, a
    bottom-up analysis would indeed present us with a system that necessitated competition between
    resources, i.e., neurons available for processing sensory inputs. But the above conclusions would
    seem to indicate that this kind of system would in fact not have too many problems, actually,
    because of the stabilization mechanisms tending to segregate processors. And indeed the
    development of cortical columns, which has been shown to happen spontaneously, would seem to
    support this (question: in models in which this happens, does the same stabilization function, and
    is this the cause?). And what seems to indicate that the passivity is misdirected and indeed that
    the competitive metaphor is also, is that without the positive feedback from something external,
    i.e., the RAS, the system tends to shut down or to remain constant. Now one can conceptualize the
    necessity for choice, given inputs requiring different and contradictory responses, as a kind of
    competition, but that necessity is not present at those early stages, merely the sensory processing
    in parallel. Given, for example, perceptrons with multiple simultaneous inputs, there is no
    competition; the system merely processes what is presented to it. Competition implies some sort
    of evaluation which puts boundaries around particular processes to create processing sets, in
    effect; but where in the system, at these stages, does this occur, and why is it necessary?
101) What if there are two images, etc., which require the same resources? It would seem that either
    what would have to happen is continued parallel processing or referral to a higher-level decision
    maker of some sort: a competition. Given stability considerations, the former would seem more
    likely. Given active contradiction, the latter. Resource allocation, per se, however, seems
    irrelevant. There should be nothing monitoring the number of active neurons for a given sensory
    processing function; how could the brain do that? That is, the question becomes, what precisely
    triggers or necessitates competition for processors, or alternatively, what triggers or necessitates a
    choice between sensations being processed? Two obvious answers: 1) contradictory responses for
    each, so that only one can occur, 2) the importance of the sensations: which necessitates one
    paying attention to it? Both of these, however, are not local competitions; they require higher level
    evaluation.
102) But at the same time we must recognize the opposing processes: that of conjoining or synthesis.
    Here we have a situation in which processes, even intra-modal processes, must be conjoined to
    produce functions that are meanings, concepts, qualia, and other higher-order thoughts, even
    propositional types of thinking. We may say that against this are the competitive processes, but
    the boundaries are unclear, especially within a tight or hot sensory surround. What is to be
    conjoined and what deleted? What I am trying to make clear is that both "competition" and
    "synthesis" are not clear-cut in any sense. However, we can still summarize and say that the
    processes which create the "basic elements" are, we are speculating, conservatory: stabilizing
    through top-down (local) negative feedback.
103) Let us assume, then, that the "higher" the organism, the more (nonlocal) higher positive
    feedback exists. That in lower-order organisms, what we have is a series of stabilized functions
    generating relatively few standard (hard-wired) processes, qualia, that are hard to change. The
    local aspect of the system dominates. In higher organisms, the positive feedback and multi-modal
    synthesis gets more prevalent. What does this mean, exactly? What should happen is that this
    feedback should destabilize the system to cause both flexibility and focusing. But what does that
    mean, exactly? Surely we cannot cause cortical columns to reform on the fly, so to speak. So this
    processing, as far as focusing goes, must consist of inhibiting the processing in parts of the cortex
    and exciting it in others. But exciting it cannot merely mean activating processing which is
    occurring for different input, but there must be a tie-in with other current input. This can only
    happen from top-down feedback, since it is the bottom-up which is acted on. But what kind of
    activation can be type-specific? There must be activation on whole cortical areas analogous to the
    local top-down feedback that stabilizes a process. That is, some very high-level process is causing itself
    to be recreated, in effect, while inhibiting other processes, exactly analogous to the local feedback
    stabilization process. This must be able to occur both for remembered processes, from memory, and
    for current input, due to immediate feedback.
104) But if this is happening analogous to the local stabilization processes, then the additional
    intensity, processing, and depth which seems to accompany "focusing" would be missing;
    focusing would merely be in contrast to the inhibition of other processes. But this does not seem to
    be the case. So extra processing must be recruited from somewhere. If new cortical columns are not
    created, then old ones must be used, for different purposes. This actually makes sense, given that
    objects are built up from the same (sets of basic) elements. This does not cause the Gestaltist's
    position to fare well, in the sense that there must be a certain constancy of elements, realized by
    the processes carried out by the volumes of the cortex. There seems to be no other way to allocate
    the same "resources" employed for one process to enhance a second while being withheld from
    the first. But this conclusion supports Baars and the "resource" people; and in addition it supports
    the analytic position, where there are elements, even if somewhat variable since cortical columns
    can change slowly over time (also see Anderson and ACT theory?). What we have, then, instead of
    a "broadcast," a la Baars, is a recruitment through top-down feedback stabilizing processes that are
    conserving, in effect, the higher-level process.
105) In order, however, that higher-level processes be flexible, and in line with that observed
    flexibility (which indeed may be illusory), it must be that this conservation is less stable the more
    complex and inter-modal, i.e., requiring more integration over larger and more separated cortical
    areas, than is the more local integration. Now this large-scale integration, however, if it is
    analogous to the integrative processes on the lower (uni-modal) levels, must result in a relatively
    small set of supervening processes, which are able to be less stable, since they occur across cortical
    areas and thus have more potential for flexible blending. But also, if we look at these in terms of
    the stabilization process, we note that at the highest levels, there is both convergence and
    flexibility, yet there must be a limit to the abstractive processes. But as these processes become
    higher-level, building on themselves and abstracting from multi-modal processes, they can
    become smaller, in terms of neural sets. Note also the difficulty in stabilization at the higher levels.
    In the prefrontal cortex, there must be a limit reached also for the degree of abstraction and top-
    down stabilization the system is capable of, merely because one runs out of cortex.
106) But then how are the highest multi-modal abstractive processes stabilized? Perhaps this is where
    the ILN and whole ERTAS comes in? After all, despite their relatively low capacity, at this point
    that may not be a problem, since the limitations of the prefrontal capacity for abstractive levels
    have been reached. The highest levels of the prefrontal cortex, then, may be stabilized through
    processes reaching down through the thalamus and even RAS to reach up through the ILN and be
    "gated," in effect, because of the limited capacity of that system to stabilize the higher-level
    processes. This might also promote destabilizing the higher levels, sometimes a desirable trait. We
    see then that the limitations of consciousness may not be due to some sort of processing or
    capacity limitations in the sense of a capacity of "consciousness," but of the limitations of the
    system able to 1) generate higher-level abstractive processes, and 2) stabilize a limited set of those
    processes due to the smaller size of the centers involved: the subcortical aspects of the ERTAS and
    hypothalamus.
107) Here the stabilizing effects of memory can enter to add capacity, in effect, to the system. If there
    are memorized patterns, these can be perhaps used to stabilize particular prefrontal processes.
    Memory thus acts as stabilizing feedback on the higher levels. The RAS, then, can be seen as a set
    of processes acting against this stabilizing influence, yet it must also be involved in the choice of
    what to stabilize, given the limitations of the prefrontal capacity, as the processing is used up in
    creating the higher-level processes rather than in stabilizing them. Somewhere in here is where the
    "mirroring" aspect of consciousness must be a necessary part of the system.
108) Another point: in what sense can we say that a "function" or "pattern" actually exists, as an
    ontological entity? I think the answer has to do with relaxation time, in effect. That is, as long as
    there is an interacting system of neurons, or anything else; and one acts on another, that other on
    yet another, and the last acts on the first; then if that first element has not finished acting on or
    being affected by its acting on that second element, then that whole is actually a unitary system:
    an entity, in effect. The relaxation time of the elements is slower than the feedback time or
    interactive time throughout the system. When that boundary has passed (an interesting issue in
    itself), then we must say that the first element, if no longer affected by the initial impulse, is, when
    acted upon by the last element, independent of that latter element (and all others that have acted
    after its "affectedness" has ended). Another interesting point here is the question of whether there
    is a single system if the elements have different relaxation times, so that the third element, say, has
    finished interacting with the fourth even though the first is still interacting with the second. Does
    the interaction of the last element with the first still unify the elements into one system? I think in
    these circumstances one would have to say that there were two systems; but what if there were no
    reciprocal interactions between the first and third elements? Then we would merely have two
    chains, returning to the first element, and not two systems (although the second might be one,
    depending on its internal structure), nor one system, at all. Thus both reciprocity (i.e., recursion)
    and relaxation enter into the creation of a system, notably, I would say, of a cell assembly.
109) Given the interconnectivity of the CNS, it is certainly possible that even a 3-d pattern could have
    its various parts, or most of them, "seen" or "sensed" in some way; but that response to a pattern
    would be a pattern itself, and why should a repeat of a pattern give rise to anything else? One
    could say that this response is not a repeat, but an abstraction in the same sense as the abstractive
    processes throughout the cortex, but again we seem to be substituting one cell assembly for
    another with no abstractive processing benefit, as we had before.
110) What are the implications of the above for qualia? That is, to "see" a system as function
    embodied as pattern, what would be involved? It would not seem enough to have a "grandfather
    cell" which responds only to this system, because that loses the uniqueness of the function itself;
    that is, why would one of these cells have a particular quality, if "seen," and another some
    different quality, if any? In that latter case, one could argue that any quality could be "encoded" by
    any single cell, which seems absurd, if remotely possible: it might, for example, be some function
    of the patterns on the cell surface, but then the problem is the same: how are those patterns seen,
    as pattern?
111) But in neural terms, what must the response to a pattern be? It must be the time summary of the
    pattern on a cell-assembly (CA: see above) surface. That is, all or part of the pattern would have to
    fall on this surface simultaneously; the result would be the interaction of the various separate
    parts falling on the surface through the normal cell-body time response to stimuli (see above).
    Each pattern, then, over some time interval, would be both spatially spread out over the surface
    and temporally summed on that surface. That CA surface, then, will respond differently to the
    various sensory qualities because of their different processing patterns.
112) But what is the point of this response? That is, given such a pattern, let us use the visual analogy
    of a TV picture, moving around on the surface of the CA… ok, now what? Suppose there were a
    parallel TV screen facing that one and interacting with it… what would that add? Something else
    is necessary here. Now however we must think about the function of these patterns, and the
    function of qualia themselves. What we are doing, actually, with this system is responding to the
    environment, after all. But in order to do that we must integrate the sensorium with the body. But
    that is not straightforward: the body, after all, consists mostly, inasmuch as it is response, of
    kinesthetics integrated with touch. How is the visual sensorium, say, to be integrated with this? It
    seems to me that some translation must happen. But we are not involved with linguistics here; this
    is not a question of translating between languages; we must think in terms of the above CA
    patterns. The ―translation,‖ whatever that is, must happen between bodily CA patterns and
    sensory CA patterns. What is this translation, what mediates it, then, and how does it do this?
113) To put it more generally, we have seen the progressive conservative synthesis of sensory
    abstractions above leading to final sensory integration (and have not paid enough attention to that
    latter). But where does that lead? What is its point? To put it another way, how could this sensory
    abstraction lead to behavior, since sensations do not in any way resemble bodily behaviors? That
    is, there cannot be straight-through processes from the sensorium to behavior, even processes as
    relatively straight through as have been postulated above for the construction and conservation of
    sensory "information," because there is no resemblance at all between these classes of information.
    What we must say, then, is that there has to be translation from the sensorium to behavior in some
    terms. The sensorium must be seen as pattern implying response.
114) Yet the reverse could equally be true, and the whole analysis above could have been in terms of
    the body, either input or output, as successive conservative abstractions, which then must needs
    meet the sensorium: behavior to sensorium. Thus it would seem that this meeting in the middle is
    what is important. But this too is ambiguous, because of the various animals that cannot be
    conscious which have the same structural/functional set up. That is, if one contemplates insects
    (and assumes that they are not conscious), one must say either that there is a difference in the way
    their sensoriums are integrated, or a difference in the way in which their sensoriums interact with
    behavior, or that the above analysis is simply insufficient. Functionally, there seems to be very
    little difference between us and them except for complexity; i.e., they must integrate abstractive
    processes in their sensory modalities, integrate those modalities, then further integrate that with
    behavior. What is different between that and the above, which is supposed to culminate in an
    analysis of consciousness? Not only that, but if we site, for example, some distinguishing mark of
    humanity, i.e., the possession of symbolism: language, we then draw, it might seem, too strong a
    line between the conscious and the non-conscious organisms: why are not dogs, for example, or
    elephants, conscious? Yet that latter must play some part; note, for example, the macaque analog
    of Broca's area playing a part in empathy.
115) Returning to insects, however, we must note that sensory integration there could easily take
    place just as, for example, intra-modal behavioral integration takes place: that is, each leg, for
    example, has in effect its own processor, ganglion, which loosely interacts with the other
    processors. This could easily be the case for inter-modal integration and behavioral/sensory
    integration as well. The modular structure here would seem to be the structural basis, rather than
    the integrative structure in the mammals, for example.

116) But the idea of the integration of sensory and bodily information has another aspect, that of
    representation. Suppose the processes that were to control and manipulate the body and the external world
    were redirected internally, toward the results of those above sensory abstractive processes, rather
    than externally toward the body. Then those processes would in fact be the bases for internally
    manipulable representations. We have here, I speculate, a reason for the internal direction of
    consciousness, that is, to create and control representations. In addition, this gives us a criterion
    for consciousness, or at least for proto-consciousness. That is, it is not an organism that does not
    have representations in some sense (which can include the abstractive processes) which is not
    conscious, but an organism that cannot manipulate those representations as if they were objects,
    which cannot be conscious.
117) This internal turn could take yet a further turn to become the basis for self-consciousness. That
    is, turning the above around again, so that the sensorium processes the results of the body
    processing the sensorium, should produce qualia. One more turn should produce self-
    consciousness. Note that the different sensory modes do not process each other, merely combine.
    Thus, we do not "auditorially" manipulate visual data, etc., while we do kinesthetically
    manipulate visual data.
118) The problem now becomes analyzing how the bodily processes must "see" visual data in order
    to manipulate it. But first one must ask why, exactly, it is done this way. That is, in computers, for
    example, the processes which "rotate" images, for example, operate totally internally, on the codes
    for the images, varying them to vary the angles, etc. Why does it not happen this way in the brain?
    It would not seem impossible for this to be the case. However, there are three problems with this.
    First, how do we conceive of the (say) rotation before it happens? If this conception is done
    internally, then a rotation would seem to have to anticipate itself, in some way. This problem does
    not arise in computers since they are externally programmed. Second, the grounds of parsimony
    would seem to hold here; that is, we can already do rotations kinesthetically. Applying them to
    visual representations would seem to be just an internalization of processes applying externally
    anyway. Third, the idea of representations as internalizations of externally oriented processes fits
    this picture: the movement of the externally seen object is performed through kinesthetics, and the
    same inward turning of the visual can be applied to the kinesthetic to perform comparable
    operations on the representations. Anticipations of rotations then would be part of the "stream" of
    bodily movements, etc., already present. In other words, multimodal integration would be
    facilitated by a kinesthetic origin of non-kinesthetic processing of representations. Also, there is
    the supporting work of Lakoff and Johnson.
119) Another interesting implication here has to do with the problem of primary vs. secondary
    qualities. Why should there be the primary qualities there are? From the above standpoint, we see
    that those are just the qualities that translate best from the bodily processes. That is, to move a
    visual image we can evoke the kinesthetics of moving a corresponding visual entity, whereas
    there are no comparable kinesthetics for processing color, for example. But yet another problem
    then arises, as to why we experience the secondary qualities at all, if qualities are derived from the
    above interactions.
120) Now let us look at this in more detail. In order to have a representation, the neural processes
    must function as an object, i.e., as something like a unit: a function or process that can only be
    altered as a whole. If this is not true, then the representation dissolves into its parts, especially
    when manipulated. The key to conceiving of a whole in neural terms, I think, is the idea of the cell
    assembly (CA) developed above. That is, the CA formed by the interaction of cells interacting
    below relaxation time is in fact a unit, since processes within it are still able to affect all the cells in
    the group, including those still initiating the processes. Now some external process, i.e., a process
    originating from some other CA, then must interact with the first by first, initiating it, which
    would consist of activating enough of the cells within the CA to start the CA maintaining itself.
    Second, the interactions would then input "information" into the CA enough so that a) it would
    output something different from its normal outputs in response, and b) it would not dissolve into
    subcomponents. Here we have the most abstract description of the interaction of two CAs. The
    questions then are: what are the differences between the interactions of CAs from within and
    without normal modalities of interaction? Second, how does a CA initiating and responding to
    bodily responses interact with a CA that represents those responses, and how is this interaction
    different from that with the bodily output CAs?
121) To put this another way, it is clear that within modalities, there are CAs interacting all the time,
    producing abstractions and maintaining them. But the interaction of sensory and bodily CAs has
    been hypothesized to underlie the generation and shaping of qualia, through the internalization of
    bodily responses onto representations of sensory objects. Why should this underlie qualia and not
    the normal abstractive processes? On the path to this, it must be noted that representations, say in
    this case visual representations, must be constructed or generated from the same CAs that are
    involved with the original sensory impressions, for two reasons. One is simply simplicity; i.e.,
    why have two sets of similar processes in separate places? Another is utility; i.e., if the bodily
    processes work on the sensory processes when the body moves objects, then in order to move the
    representations in corresponding manner, the processes upon which it acts should be similar to
    conserve processes in the bodily realm. In other words, dealing with the sensorium, whether real
    or represented, should be accomplished through the same set of bodily CAs, again for simplicity's
    sake. And contrariwise, the bodily processes that manipulate the representations should be the
    same as those employed to manipulate the external world.
122) There is evidence backing up both of these. First, experiments have found that visual
    representation uses the same brain areas as visual stimuli, and those stimuli interfere with the
    representational process. Second, it was found recently that the cerebellum is used in some
    higher-order reasoning. Thus the interface between the visual CAs and bodily CAs is the same in
    both cases. If this is true, it implies that processes from the bodily CAs must be automatically
    internalized; in other words, there must be neural circuits from the cerebellum and motor cortex
    to the other areas already in place: the inward turning above must be hard-wired, and thus so
    must be the potential for representation. But these areas are very old; it must be, then, that animals
    also have representation, and this is also borne out in, for example, finding that rats have brain
    areas that seem to map mazes. This is an argument for the potential, at least, of animal
    consciousness, if the above arguments are correct.
123) As an aside to which we will return later, it may be that language is the result of integrating
    those bodily and sensory interactions, so that one level higher, both may be manipulated
    abstractly and simultaneously, as single CAs.
124) Note that much of the normal feedback for the bodily/sensorium interactions must be missing
    from the representational interactions of 118-9, simply because there are no peripheral
    sensor/effectors to provide it.
125) Now, in order that the manipulations of the representations be the same or similar, despite these
    limitations, as those of the sensorium from which they are derived, the bodily processes must
    "see" the same abstractions as they "see" when operating on the actual sensorium. This is another
    argument for using the same processes and area of the CNS for the representations as for the
    sensorium. It would be too easy for these to drift apart, without enormous feedback for correction
    at very low levels, a price that seems ridiculous (although what are we doing with memory?).
    Also, what the body processes operate on must be meaningful, in the sense that the sensorium
    operated on must correspond to real-world sensations and operations, or, at the level of
    explanation we are currently operating on, the neural processes must result from and correspond
    to those resulting from real-world operations. Thus the abstractive processes must be constrained
    by real-world constraints, and also, the interactive processes in the representations must be
    constrained by the same real-world constraints.
126) Now, returning to 117, when the bodily processes manipulate the sensorium, they must do so in
    constrained ways. For example, we cannot reach out and manipulate color; we must manipulate
    corners, edges, surfaces, volumes. In other words, the intersection of the sensory and bodily
    processes in representations must be in those aspects of the sensorium that the body can
    manipulate. What of the reverse? What does the bodily process "look like" from the standpoint of
    the sensory representation? First, it is interesting that this is a difficult question. This indicates that
    even at this level, there is viewpoint, in a manner of speaking, and that the viewpoint is that of the
    manipulator. The fact of viewpoint even here indicates, perhaps, a basis for a sense of self. The
    sensory representation is the passive, for the most part receiver of the manipulative process, and
    reacts to it by changing; is there any corresponding reaction and/or change on the part of the
    manipulating process?
127) What seems to be the case here is that the manipulating process is, in effect, a tool, in, I believe, a
    Heideggerian sense. That is, the hand must adapt itself, shape itself, to the object it grasps, and so
    must an abstract representation of the hand, even if there is no actual representation of the hand
    per se. The manipulating process must still fit itself to the object, both the object as shape, and the
    object as function, reflecting both the kind of manipulation to be (and being) performed, and the
    goal of that manipulation. Thus from the viewpoint of the sensory representation, the bodily
    representation must be adapting itself to change it; and indeed must then be affected by the
    subsequent change.
128) However, we are dealing here, so far, with mere neural impulses, even if they might be treated
    as single extended entities over the CAs, we must still be careful, at this stage, not to speak of
    qualia. But we have seen that the bodily and sensory representations do nonetheless correspond
    to, reflect, and change as do the actual sensoria and bodily movements. There are too many
    arguments, above, resulting in this conclusion. What is missing in the
    representation/representation interactions, however, is actual real-world feedback, as these
    processes are initiated and continue. This is, I think, the next crucial point. This feedback must be
    manufactured (or remembered), on the spot, internally, and that implies that there are monitoring
    processes that are tracking the interaction as a whole.
129) So the assumption that qualia must come out of the interactions of the bodily processes with the
    sensory processes, made in order to break the linearity and similarity of interpretation that
    confining process interpretation to the sensorium would result in, is supported also by the work
    of Lakoff and Tucker et al. But that implies that insofar as those interactions go, the interpretation
    of sensory processes by the bodily processes must be in terms of the bodily processes. But what
    does that mean? What is being interpreted, exactly? Suppose we take a visual impression, or
    stimulus, which has physical extent. (Again, I am, with difficulty, trying not to use terms
    necessarily implying qualia yet, and this must be taken as given even when the terminology seems
    to imply them, until I am explicit about this issue. I am assuming that all that can be understood in
    terms of insects is without qualia.) Now the corresponding bodily processes that have extent are
    either temporal ones, where one moves a body part through some distance, so that there is a
    continuity of kinesthetic feedback through a changing body position, or atemporal, where there is
    kinesthetic sensation (or touch, etc.) over an extent of the body simultaneously. Those processes,
    then, must be employed (it's all pre-qualia, remember) to interpret visual extent. We must take a
    visual impression of some sort and overlay it with that aspect of bodily impressions, or vice versa.
    Clearly, we cannot only use the bodily impressions or we would have no visual modality or
    qualia; we must use visual information. Thus the visual information must be compatible with the
    bodily information for that type of interaction, yet distinct from it. But all this could happen in an
    insect; to go beyond this, it must happen in the representation of the sensorium.
130) The representation is a dynamic entity; it must include the sensory/bodily interactions, and
    responses to those interactions, all "simulated," i.e., with none of those necessarily driven by any
    external inputs. Now one abstract reason that we must have in these representations, for example,
    visual color information, is that we react to it, and to the individual colors. That is, specifically,
    even different colors are functional in terms of our interactions with the environment, and thus in
    representations; if behavior is, say, partially color-driven, then we must connect it to bodily
    responses. Thus one cannot separate or isolate even such sensations, which seem intrinsically so
    disconnected from the body, with that bodily dynamic. This is at least an abstract reason for
    finding such an interaction. If there were no representations, then we might not have to have
    sensations, at least such sensations as colors, that are so far removed from the kinesthetics of the
    body, for example. But since there are these complete internalizations of the environment and of
    our interactions with it, then anything relevant must be included.
131) Now again, all the above could apply to insects, which, I am assuming, have no representations
    and no qualia. Nonetheless even such relatively simple entities perceive "colors" and employ them
    to discriminate the world and to behave differently in different situations; thus this visual
    information must be, if not integrated with kinesthetic information, at least employed to modify it
    in some manner for them also.
132) Now let us consider qualia. The above argument gives us, I believe, a handle on types of qualia.
    Since, first, all possible interactions of the body and environment must be included, then all
    aspects of our sensorium must be included in a representation. Now, there are two classes of
    interactions between the bodily representations and the sensory representations. First, there are
    those which have matching characteristics. Thus, for example, spatial extent is a common
    characteristic of the bodily sensations and manipulations, and also of the visual sensations. All of
    these have spatial extent, although they manifest it in different ways. However, when kinesthetics
    and visual extent are integrated, as they must be for a representation to be formed, then the
    matching characteristics of these different modalities must be juxtaposed or overlaid. Kinesthetic
    spatial extent involves, atemporally, different muscle, skin, and weight sensations which are
    spatially adjacent, i.e., which are not superposed, and so must be separated in some manner.
    Similarly, many visual impressions are not superposed, and so must be separated. When there is
    some sort of intermodal confluence, i.e., when we touch something and see and feel that touch, the
    separation of the impressions must be maintained, yet the superposition of the corresponding
    visual and kinesthetic impressions must also be created. This is a simple argument for the
    integration and its nature of these two modalities in this particular function; obvious extensions of
    this can be applied to other congruent characteristics.
133) An interesting aspect of these multi-modal characteristics is that they do indeed seem to be
    "primary," as I mentioned above. Thus one could argue that we can have colorless visual
    representations, but not that these cannot have extent; and this seems to be, in extreme cases, true.
    Reciprocally, one should be able to argue that one cannot have ("colorless," i.e., without, say,
    granularity) kinesthetic sensations without extent also; and this also seems to be true. My
    speculation here is that this is due, not to some fundamental characteristic of the "objective" world
    associated with "extent," but instead to the process of multi-modal integration or synthesis being
    fundamental to our representation, and thus (as I will argue below) perception of the world. Thus
    any process associated with multi-modal integration of matching multi-modal sensory characteristics
    must be "primary." One might object by asking how, if visual sensations and kinesthetic ones also
    were not spatially organized, could they be (intra-modally) discriminated? However, given the
    multiple possible state spaces employed in physics, for example, it would seem to be a possibility
    to have such discrimination based on other dimensions.
134) We might ask, for example, why such types of information as spectral frequencies (which are
    not, at this point, colors) are integrated into this scheme; why, indeed, they are necessary at all in a
    representation. And of course they are not always. Yet, in order that a representation be faithful to
    one's experience, for a variety of functional reasons (all too obvious to be worth going into here),
    all characteristics of all sensory/bodily modalities must be at least possible to integrate into
    representations (with the caveat that they may not, for reasons of, say, parsimony, always be so
    integrated). See also 128 above.
135) So we must again address the question of what a representation is. Above, I stated that it has
    something to do with the "inverting" or "reversing" of outputs and inputs, so that the sensorium is
    created internally, in effect, with the body and the other senses integrated. But there are two
    "representations" that must be contrasted with this in order to approach the problem of qualia;
    one is that of insects, and the other is that of computers. I have said that insects do not have
    representations, and this seems correct, but in what sense? It does seem true that functionally,
    insects cannot plan or remember in anything like a human sense, i.e., by manipulating analogs of
    the world; and this sense is, I think, what I mean by representation. But this leaves the question of
    why computers do not have representations; surely they operate with analogs of the world? There
    are at least two differences here, though. First, the "analogs" of the world that computers operate
    with are the computer's world, and not analogs, i.e., not models from the computer's standpoint.
    That is, symbolic or not, abstract or not, a computer has no world but this internal one, and so it's
    processes are directly analogous to that of an insect's, merely operating in its internal reality rather
    than in an external one. Second, the first part of this paper has extensively analyzed the
    differences between computers and the brain. That is, the computer's representations are always
    class symbolisms, and thus do not, in essence, represent anything specific.
136) A human being, and perhaps most animals, then, has representations that are distinguished
    from the world. Thus they must be 1) an addition to the perceived sensorium, 2) distinguishable
    from that sensorium, 3) actively compared to that sensorium, and 4) must contain individuate as
    well as class information about the world.
137) However, there is another problem with the above analysis. While the rationale for primary vs.
    secondary sensations seems reasonable, the interpretation of qualia as the results of the processing
    of one modality by another seems wrong, because, first, why should any one be primary; and
    second, the interpretation by a given modality of another modality can only be in terms of that
    first modality. Thus, while a kinesthetic process can interpret a visual edge, it cannot do so in
    visual terms, only in kinesthetic terms, i.e., by presenting that edge as sharp, say, or as an abrupt
    spatial movement or transition. But this does not change the visual information nor produce
    visual qualia, only at best a quality that is a combination of the two. This is of course necessary,
    but not the generator of qualia.
138) The point of a representation, then, is not to reproduce the world or to recreate the sensorium,
    but to take the results of that processing, that generation, and manipulate them in various ways,
    not merely kinesthetic. However, a representation in this system must be a whole, i.e., all the
    sensorium, bodily, emotional and so forth processes need to be integrated into representations, to
    extents relevant, perhaps to the relevance of those processes for particular manipulations. On the
    other hand, if we consider, for example, the visualization of a painting, it is rare that we hear any
    associated sound. This total sensory integration, then, is not clearly essential to the nature of
    representation.
139) What is, however, are the points in 134 above. Let us take, for example, the mental rotation of an
    image of a cup. Present in this image are not only its visual shapes, but its rigidity, its hardness
    and momentum. Not present, usually, are the sounds make when the cup is dropped or hit, but
    the glassy feel of the cup is present. Now for these, and perhaps its color also to be present, when
    that cup image is rotated, say, this cannot be done as some sort of manipulation of an abstract
    symbol, even a fairly abstract image of a cup. All the above information must be present, and this
    is usually also the image of a particular, if not a known, cup - it is at least similar to cups we have
    seen or moved, usually not of some grotesque cup or one very dissimilar to those we have seen.
    Thus there are particulars present in this image as well as abstractions. But this echoes a point
    made above, that individuate information must be present also. Thus it cannot be that the neural
    processes are operating merely on the top level of abstraction, the topmost cell assembly, of the
    processes leading to the creation of the idea of cups. But this is a very interesting point. If this is
    not the case, then what is? How much, simultaneously, can a motor process (whatever that is)
    operate on? And what does that operation consist of?
140) Let us say that this operation consists, in part at least, of calling up, generating, or evoking
    neural processes corresponding to the various positions in space of the cup as it rotates. Thus
    there must be both constancy and change in the processes evoked. The cup as cup and as
    individual "image" (no qualia yet, remember) must remain constant: its "color, weight, shape" - yet
    it must rotate in some "abstract space": we must "image" the various aspects of the cup as it turns,
    and in addition the image of its "active motion": its turning or rotation as it remains the same
    entity. Now in doing this, we are not merely evoking these complexes of processes, but we are
    evoking those that the sensorium evokes, for reasons given above. Thus the neural evocation is
    some sort of downward induction of processes: we start with "cup" and fill that in, in effect. This
    would permit the inclusion of individuate processes if it went low enough, and would also
    explain the lack of much of these in images. We can bring in the obvious things here: there must
    be feedbacks to check that the rotation is proceeding correctly, there must be feedbacks to memory
    to check accuracy of representations, etc., etc.
141) The next inference is to the general argument. Let us recapitulate. We assume (which seems very
    supported) the existence of "representations" in the above sense. That is, these are internally
    generated processes relating to the sensorium, reproducing to a great extent the processes
    resulting from sensory input, but internally generated. These processes interact with similarly
    internally generated motor processes, which are internally directed: toward the sensory processes,
    so that they interact with the sensory processes in ways functionally similar to the ways that
    motor processes interact with the sensorium: to produce analogs, say, of the above rotations.
    Notice that in this definition of representation there are no qualia. This is not a circular definition, as
    so many have been. Now, the next step.
142) If we are attempting to proceed from neural processes to qualia, we must argue that these
    processes, described above, are the first step in that direction. What we have, starting here, is a
    situation in which neural processes are reacting to and with other neural processes in ways, on
    some level of abstraction, similar to the ways in which they react to the world, since that is the
    purpose of representation. Now given that these are reacting in this way (the way they react to the
    world), they cannot solely be reacting to them as to neural processes, i.e., directly on the values,
    discharge rates, particular neurons, etc., that would result in the analogs of the above rotations.
    The inter-reacting neural processes involved in representation do not "see" each other as neural
    processes, as we, the external observer, would. Even if one maintains that this is possible (see next
    paragraph), I must maintain that at some point neural reactions must be on a different level, so to
    speak, because if it were not, the processes would not be functioning, ultimately, mentally, as a
    representation. Thus the inter-reacting processes that constitute representation must be reacting to
    each other, ultimately, as if there were qualia. It is still not possible to say that there are qualia, but
    here we have the first indication that functionally, at least, something like qualia must be starting
    to be present, in the functional dynamics of this interaction. More on this later.
143) One might counter with the claim, however, that the neural processes do operate on each other
    in those terms, i.e., as neural processes (how else could they interact – they are no more or less
    than neural processes, after all), and thus that the motor processes are operating, analogously to
    computer processes, directly on the values, discharge rates, particular neurons, etc., that would
    result in the analogs of the above rotations. But in this case, it is hard to see how these could be the
    same processes as the ones we employ externally, for three reasons, first, we are not aware of any
    of this neural information; second, the kinds of operations and feedback, while similar to the
    sensory analogs (internal representations), cannot be identical with them when dealing with the
    external world; third, the same neurons are not involved: those for the representations may
    overlap those for the external ones but cannot be identical (because of the top-down factors
    mentioned above). Thus it would seem that the system must be using some overall viewpoint,
    codes, abstractions, or something of that sort to use different but overlapping and functionally
    identical sets of neurons to deal with both the real and internalizations of the sensorium, with
    both real and internalized motor processes. It seems too much to assume that the same detailed
    neural codes would work both for representation and for reality; and this implies some sort of
    necessary creation of generalities, both sensory and motor.
144) The decisive argument along that line, I think, is the following: any entity employing
    representation must, absolutely, be able to discriminate between representation and reality, i.e.,
    between the internalization of the sensorium and motor processes and the sensorium as induced
    by the world and motor processes as operating on the world; because, quite simply, if it could not
    make this discrimination, it would confuse representation and reality. This of course would be a
    disaster. So the actual mechanism is irrelevant; the point is that there must be such discrimination,
    and it must happen before conscious choice, or what would that choice be based on?
145) Now if this is true and if it is also true that representation employs, at least to some extent, the
    same neural processes that sensation employs, then we must also say that there cannot be identity
    between these classes of processes, because of the necessity of their being discriminable. But if that
    is true, where is the difference? We see that it cannot be at the lowest levels, at least completely,
    because those are too sensory-driven, too ubiquitous and complex, and too dependent on
    stimulation for their values. That is, it must be remembered that in order to use different
    processes, the neural circuits, the CAs, must be different; that is just what ―different processes‖
    means. But if the originating sensory processes were different for representation, a duplicate of
    the neural sensorium at the lowest level would be implied; and that is just too much to duplicate.
    Our neural sensorium might only be partially used, and this, given the unreliability of
    representation, seems likely, but especially, given the existence of eidetic imagery, the CAs used
    must be those of the actual sensorium. It must be, then, at the highest levels at which the
    duplication, the lack of identical CAs in order to have representation discriminable, must take
    place. But if this is the case, then when we call upon or induce a representation, we must do it, not
    by inducing the sensory processes directly, but through downward feedback from higher-level
    abstractive processes: summaries of the sensorium. But this is beginning to sound like qualia: to
    converge to the same argument as #140 above.
146) The next consideration is this: that there must be not only an ongoing interaction between the
    sensorium and representation, but that interaction must generate some sort of results:
    comparisons between the two. We might ask what the sensorium ―looks like‖ from the standpoint
    of the representations, and vice versa: what the representations look like from the standpoint of
    the sensorium. That is one possibility; another is that there is a third part, which does the
    comparison, which has, in effect, a viewpoint on the interaction of the two. What do the
    differences and similarities look like from this point of view? Have we actually succeeded in
    deducing the necessity of a viewpoint? If we take the generation of abstractive processes, as
    above, to be fundamental, then is this part of that generation; that is, are these interactions
    abstractions in the same sense that the sensory, etc., abstractions are?
147) Let us return to the consideration of the function of representations. These cannot, in humans at
    least, just be echoes of the sensorium; they let us construct a goal state, then work from that to the
    real-world configuration which can realize that state. However, this goal state must be constructed
    from the top down. If that were not the case, this construction would be done either randomly,
    like animals in Skinner boxes, or only through logic, like computers. That is, a goal state
    constructed of specifics would be just a model of that state as it would look in the sensorium. Yet
    how is this model arrived at in the context of problem solving, if not through the same random
    groping that the actual goal state would be arrived at, without guidance from representation?
    Only through logical (etc.) operations on the same specifics. This implies that the guidance must
    be in some other form than such a model. Top-down goal specification allows us to have
    intuitions, to construct not only algorithms, but heuristics, in effect. But the ―top‖ of that process
    then must be some sort of generalization or abstraction of that goal state, if it is not put in terms of
    specifics of the sensorium. This is the first clear difference between representation and sensorium,
    and it is essential to our higher cognitive functioning. Yet animals that we do assume are
    conscious do not have this level of functioning: rats, cats, dogs, for example, do not, as far as I am
    aware, exhibit intuitive problem solving; but apes do. If this is true, then rats may anticipate
    through representation by manipulation of specifics, but without the higher representation of
    abstractions of states of the sensorium they cannot intuit, in effect. But if that is true, then these
    abstract states are not essential for consciousness, although they may be for self-consciousness or
    consciousness in a full human (and infra-human) sense: note the narrow range of animals that
    have the mirror effect, for example (also note recent results with elephants).
148) If this is true, then the system with representation, sensorium, and their interactions as relatively
    low-level models should suffice to generate simple qualia. Thus, I am at this point making a
    specific claim: that representation in the sense in which I have developed that term is necessary
    for qualia, and thus ultimately for consciousness. This is in explicit contrast and disagreement with
    those who claim that “proto-consciousness,” “pan-psychism,” and so forth hold; I am claiming that a very
    specific set of structures and functions, as described above (and continued below), must be present for
    mentality (i.e., qualia), then consciousness to occur, for reasons given above (and below). Thus, there are
    “thresholds” of such structure/function below which there is, first, no mentality, second, no consciousness.
149) Suppose we had a set of neurons (R) that acted as if they were the color red. They would have to
    do this within a model of the sensorium, of course, since they could not do this as part of the
    sensorium. Then suppose we had another set (V) that acted as if the first were the color red.
    Again, these would have to act on the model as if it were the sensorium, and the set R as if it were
    red, and the set V would have to either be in the sensorium (as argued above) or a model of it.
    Now what would ―acting as if they were the color red‖ mean? First, this would have to happen at
    a very high level of abstraction, past the initial stimulation, through when the color was
    determined through the interactions of the fundamental colors, and even probably including the
    various interactions of red with objects, feelings, etc. This is after all what being red means and
    implies. This would result, as argued above, in a CA that abstracted these various processes into a
    set of neurons. Notice that this, since it requires another set to interact with it, needs to be in the
    model of the sensorium, the ―representation,‖ as argued above. Thus it can in fact only consist of
    higher-level abstractions, as argued above.
150) So we have the representation of red, the sensorium‘s red, and we must recognize that the
    representation is a) a representation, and b) is not a representation of green. Now none of this, yet,
    would seem to necessitate qualia; all this might conceivably be done entirely through the
    interaction of CAs in which the neural sets merely responded to other CAs as neural sets. But if I
    am correct in my conclusions about the representation being exclusively higher-level, then it could
    not contain any low-level information about red. How then could it be identified as red? The
    above implications of red might not be enough; another color could (and here we see where the
    color-reversal arguments come in) have the same evocations. Then something else must be used to
    identify the representation of red as red. Now the other consideration here is that the point of
    having a representation is that it acts as an internalization of the world, so that one may run
    simulations, in effect, anticipate, have simultaneous memories and creative patterns, and so forth.
    But to do this, as I said above, the representation must be treated as ―the world.‖ The point of a
    representation is not that it is internal, but that it is an internalization of the external; and thus it
    must be processed as the external world is processed.

151) The question to answer here, at least the preliminary question, is: how is it possible to even
    conceive of qualia in neural terms? But the answer, surely, must involve the CNS‘s recognition of
    pattern. That is, if the CNS does not treat the neural impulses as a digital computer must, then it
    must be able to treat a set of impulses as a single pattern. But what does this mean, really? Above,
    we have seen how the abstractive process can work; that is a way that a pattern, or at least a
    synthesis of patterns into one or a small set of cells firing, can be brought about. But this cannot be
    the answer, because pattern here is reduced, from pattern to again single-cell firings. How can
    pattern be perceived as pattern yet not be reduced, within a neural net context?
152) There are two ways that this can happen. One is temporal, and one atemporal, and presumably
    the CNS uses both and in fact (see below) combines them. Atemporally, we can look at the above
    description of cell-assemblies. That is, one might view the CA, when presented with a spatial
    pattern of neural firings, responding with a spatial pattern of its own, one unique to the input. The
    interesting point about this is that due to a conception of the CA in which below-relaxation time
    impulse feedback unites the group of neurons, the temporal aspect is in fact included. Thus, a
    spatial pattern presented to a CA can only be recognized as a pattern and responded to with a
    spatial pattern of firings (or depolarizations) within the CA if the pattern presented happens
    within that relaxation interval. The temporal pattern is more interesting.
153) Let us suppose there is a feedback loop (A) somewhere in the CNS, which is self-sustaining IFF
    another feedback loop (B) interacts with it in such a way that A is ―driven,‖ in effect, by B, such
    that if B is not present sufficiently regularly, A dies out (this may also be an indication of how a
    neural net might instantiate a "rule"). Clearly, this all-or-none response on the part of A is a very
    crude form of recognition, which might be refined by adding outputs from A depending on the
    temporal input from B. Then we can say that the A loop, a temporal pattern, recognizes the B loop
    as a temporal pattern of firings in the overlapping set. If we combine the two, spatial and temporal
    pattern reception, so that a CA fires over a spatial area and temporal interval only if sustained
    over the area and interval by a second CA, which in turn might be sustained by the first, or by
    another, then we must maintain, I think, that there can be true non-reductive pattern recognition
    possible in the CNS (see 194 below for more detail).
154) Now the connection between representation and the non-reductive pattern recognition is fairly
    clear; in the interactive boundary between the sensorium and representation a non-reductive
    recognition is essential, since the abstractive processes there must be sustained from both sides, in
    effect (note the essential role, described above, in the temporal maintenance of the pattern
    recognition CAs of the driver CA sets). A possible objection to the above is the anatomical
    placement of the RAS, assuming that is involved in consciousness. In other words, what of a
    putative interface, as the above, between the cognitive faculties (which the above seems oriented
    to) and the emotions? However, if the arguments above are correct, what is necessary for an
    approach to consciousness is the ―inward turning‖ of an outward-directed or outward-oriented
    set of (sensorium-related) processes to create a representation of those processes. If this is true,
    then that must happen to emotion (or to those processes on the ―other‖ side of the cortex from the
    RAS) also for the above kind of interface to form. Perhaps it does, and we can represent emotion
    as we can represent the sensorium, i.e., internally create and manipulate emotions. This does seem
    to be possible to some degree (what of Jung‘s example in the four personality types?), and to that
    degree, there might be such an interface. If this is true, clearly it does not invalidate the above
    arguments relating to the sensorium, merely adds another set of non-reductive recognitional
    processes. Now one might then argue that this necessitates two (or more, given more such
    primary/representational boundaries) centers of consciousness, assuming this general argument
    goes to its obvious conclusion. However, it is not clear to me that the boundary processes between
    a class of representation and its generating processes, given that there are more than one of these
    boundary processes, cannot further be integrated, either in a superset of boundary processes
    abstracted from the primary ones, or merely (and I think more likely) in sets of interconnected
    boundary processes. Neuroanatomy must answer this question. In addition, when a
    representation interacts with its representee, it must do so in functional wholes. By "interact" I am
    referring to the comparison of the representation and representee for whatever purpose. Thus if
    for the representee a table is a whole, it must interact with its representation as a similar whole,
    and not, as a table, for example, with some part of the table.
155) We should recapitulate. At this point, there is an indication that qualia might be derivable from
    the interaction between representation and what is represented (the "representee": e.g., the
    sensorium). However, there still does not seem to be any necessary force behind this conclusion.
    Yet is such necessity necessary? After all, what I am trying to do is merely to show how qualia
    might originate in a neural net, not that they must so originate. It might indeed be the case that
    neural nets can have representation and yet not have qualia; but that will not invalidate an
    argument that shows how, given representation, a neural net can have entities that behave, at
    least, like qualia. Is this sufficient?
156) Let us consider something else, the interaction between "information" and the mental. A position
    like Chalmer's seems incorrect; either giving up on this question and saying that mind must
    permeate, in some "as-yet" unknown fashion, everything: a retreat to a very old position,
    alternatively, at worst, however dressed up in verbiage, a circular argument. However, there is an
    interesting comment on a position maintaining that there is a mental aspect to "information." On
    the face of it this is absurd; the initial definitions and applications of information are in
    communication theory and in classical physics, both far from necessitating the mental, and other
    meanings of the term seem merely circular to me, in which "information" incorporates qualities.
    Similarly to what I said above about neural impulses, there is no green information. However, one
    of the problems in the attempt to find a bridge between mind and the physical can, perhaps, be
    generalized to: a physical entity must, in order to become a mental entity, "change" in some way.
    Now usually this change has been taken to be from "physical" to "mental." But no one, really,
    knows what that might mean, i.e., what a "mental" entity really is, inasmuch as that knowledge
    enables it to relate to and be generated from physical entities; the usual response to this is to give
    up and call physical and mental radically different. Can this be bridged, however, through some sort of
    transformation of a physical entity into another physical entity that must be interacted with as an
    abstraction, not as whatever physical entities underpin (i.e., generate, give rise to) that abstraction? Some
    actual emergent entity? Why would it be mental? How could we know if it was mental? It is clear
    that we will never externally observe a green mental object as green; but how could we observe it
    (or its neural realization) and know that it was mental and that it was internally observed as
    green? To put this another way, if some abstraction of neural impulses were the basis of the
    mental, such as (to give a simplistic example) the temporal envelope of discharges of a CA, then
    although this is a physical event (or series of events), we could not observe it directly. What of
    even more abstract derivations from the physical? Why, since we do not know what the mental is,
    as it relates to the physical, could these not be mental?
157) Similarly, if we contemplate observation, we see that bouncing photons off neural circuits could
    not give us a picture of the mental aspects of those circuits, nor could any other purely physical
    conventional forms of observation, at least directly, unless Chalmers and other's with similar
    positions are right and "mentality" is all-pervasive. (This does in fact seem odd given that position;
    one would think, if "pan-psychism" of some sort is correct, that we could indeed, as external
    observers, witness the mental aspects of, say, neural circuits, since any observational medium
    would be, since all (according to their argument) is in part mental, itself in part mental.) What
    we're looking for, then, is some kind of emergent properties on the physical which correlate with
    the mental properties, and which interact with some other emergent properties in such as way as
    to make that correlation direct; that is, in such a way that the response of the "internal observing"
    emergent properties is a gestalt response, in the two senses of 1) being a response to the incoming
    pattern as a whole, and 2) in itself being a whole pattern. Whole here implies a) functional whole,
    so that functions (may) change as the pattern changes, and b) perceived whole, so that the
    response of other patterns to it changes as the whole changes. A change "as the whole changes"
    implies that either a change in any of the parts of an initiating pattern can produce a global change
    (a change in many parts) in a responding pattern, or that such changes might produce no change
    at all in a responding pattern (two members of the same class of entities or a single member with
    an insignificant variation).
158) In other words, the "gestalt" nature of the above patterns is, I believe, necessary in order to
    separate an emergent entity from merely a collective entity. If the entity cannot be treated as a
    single unit, which becomes a different kind of entity if broken up, then it is not truly emergent. For
    example, the neural cell body membrane is an emergent entity, because of its temporal/spatial
    signal processing (as described below). If it is separated into small enough units, they no longer
    have these capabilities, and if broken into molecules, cannot even be termed a "membrane" any
    longer. One must have a description not only of its component molecules, but their relative
    arrangements and the medium in which they are suspended in order to predict their properties to
    any extent. These properties, then, are emergent, and the membrane is an emergent entity. The
    problem with the mental, then, given all the above, is not with generating such emergent entities;
    it has become clear that they can indeed exist. The problem is that of observation, or point of view.
    Clearly, no matter how abstract, an external observer could observe the entities described above
    (e.g., 150-155), but not as mental entities. The "mental vs. physical" (scare quotes) problem has
    become, it seems, that of describing the point of view of an emergent dynamic pattern.
159) How can something without, we assume, consciousness, have a point of view? This question,
    however, does not seem clear. What is "consciousness" except a point of view? To put it another
    way, no one believes anymore that consciousness is an entity, whatever the status of the "mental"
    is. If we assume physicalism, which I am clearly doing throughout, then mind must involve, at
    least, the apprehension of, or reaction to, some part of the brain to some other part (that latter part
    being at the very least involved with sensation). But that implies that the "point of view" of which
    I am speaking is indeed an internal one, in which some part of the brain, in reacting to another
    part, by virtue of that reaction embodies and generates a point of view. In addition, for a pattern
    to react as a whole entity to another pattern (i.e., for the first to have a point of view) implies,
    among other things, that the same reaction (e.g., seeing the other as a particular quality) might
    with different patterns. But this implies that a reaction is not functionally analyzable into its
    component impulses, since those must be different in reaction to different patterns. This is
    certainly the outline of an argument; the question, clearly, is how a pattern of the above type, even
    if an emergent entity, can have a point of view: what that statement can possibly mean which is
    not circular.
160) Let us ask this in another way: what, in general, would be required for a pattern of the above
    sort to have a point of view? Surely we must say that first, that pattern needs some degree of
    stability: to have a point of view implies a standpoint from which this viewpoint is "had" and thus
    some sort of constancy from which to react to other, changing patterns. Second, we conclude,
    then, that this pattern cannot be purely sensory or representational, since those latter patterns are
    highly variable entities. Third, the pattern with the point of view must be a unitary entity of the
    above sort, i.e., a pattern which is the dynamic of a CA and united through below-relaxation time
    feedback into a single emergent entity, as described above. That is, it seems, given the arguments
    above, that virtually the only way to get dynamic stability in the CNS is through such a pattern. In
    addition, since we are aware, not of neural circuits, etc., but of "qualia," the kind of patterns
    necessary must be at least conceivably capable of "having" or being in states which correspond to
    qualia. Where do we find such patterns? Given the arguments above, they are first found, at their
    simplest, at the interfaces between sensoria and their representations. This is not to say that the
    latter patterns are complex enough to have qualia; we need to investigate this further.
161) But the reasons for the intuition that representation is necessary for qualia are clearer: first, only
    if two functions or the contents of two separate but related processes are similar, are there reasons
    to compare them (this does not imply, of course, no comparisons between other types of
    processes, only that the above similarity and interdependence implies comparison). That
    comparison would seem be have to be made in terms of the qualities of what those two processes
    realize. That is, the similarities between representations and their originating processes (e.g., the
    sensorium) correspond precisely to similarities in quality, and similarities in underlying neural
    processes are not implied or necessary (though they have in part at least been experimentally
    verified, i.e., the interference between symbolism in a particular modality and sensations in that
    modality - however, they have in part also been experimentally disproved: i.e., the separation of
    symbolic and sensory areas). But if the latter is true, then the comparison in step one must be in
    terms of qualities, realized in neural terms.
162) However it seems to be true that one can have sensory experiences without representational
    experiences, and representational without the corresponding sensory experiences. It is also true
    that one can have both simultaneously, as when one anticipates or remembers as one perceives.
    Yet Husserl points out the ubiquity of time-experience, and if this is true, then while one may
    have representational without sensory experiences, the reverse is not true, because one always has
    a simultaneous presence of the past and the anticipated future in one's sensorium; and these,
    especially the latter, must be representational. One may ask whether this can be true of entities
    without representation, and here one must refer to simple neural nets, in which responses can
    change, etc., without any kind of simultaneous representation of other sensoria.
163) In addition, the top-down stabilization processes described above cannot function, as described,
    to inform the CNS that a CA has fired during representation, since they (the processes) are
    downward directed. But it must be known whether the correct single cell or CA has fired/been
    employed, and checked against memory. Also, upwards abstraction can only go so far, so that
    process (class) cannot be used to check this. But when we are conscious of recognition, we are
    conscious of it as content, i.e., not merely as sensory patches, etc. That is, even in brain-damaged
    people, the recognition of a comb as an entity with particular functions and implications, even if
    wordless or incomplete, is more than the recognition of similar sensations, but includes
    recognition of content. Thus we cannot be conscious that something is a comb without being
    conscious of what a comb is. The implication is that we cannot explicitly manipulate the content
    that we are not conscious of in representation, although it may be present in representation. But
    suppose that Lakoff and Johnson, etc., are correct and there are unconscious processes that are
    representational, in the sense I have used? Then we can still say that, first, these processes must
    have been and probably still are at least potentially accessible to consciousness, and thus, second,
    these unconscious processes cannot proceed if the ability to be conscious of the corresponding
    representation has been irrevocably lost. Thus although in order to have consciousness one must
    have representation, the reverse is not implied, that in order to have representation one must have
    consciousness, at least all the time. What then is the difference between representation run with
    and without consciousness, if any?

164) But we have gone far afield from the neural problem; let us regress somewhat. We must address
    the "mind" problem in a slightly different way. It seems clear that, given that there is no green
    information, nor green nervous impulses, and given that we are materialists, that we can never
    directly observe greenness in others, but that it is, in effect, there, as internal "observation," i.e., the
    reaction of circuit sets against each other. Then what criteria shall we use? Given the arguments in
    155-156 above, it would seem that emergent dynamic circuits, as entities, must be employed, but
    for what? We must be able to say, looking at a circuit, neural or other, that this circuit is conscious,
    and/or that it is "seeing," i.e., realizing, green. On the face of it, this would seem difficult, but in
    fact the green part, without consciousness, is within reach today. Surely one can trace the
    impulses from the retina, upon the stimulation of a green light, through the cortex, etc., and
    differentiate these responses from those which occur upon stimulation with a red light. For any
    given person, this would seem quite possible, in theory. In addition, given the Mach and Hering
    principles, borne out, more or less, in the structures of the visual cortex, one could trace the
    stimulus mixing, i.e., the balances between the various circuits for opponent colors, etc., to find
    those that correspond to the pattern for green, in contrast to others, even generalized over people,
    since the general functions are known (e.g., Steven Palmer). Thus the neural correlates for
    greenness are not theoretically unrealizable. But of course one could object that these are merely
    the neural correlates of a response to green, not of greenness.
165) We must then couple the above with the other reasoning in this essay. The neural correlates of
    green must, at some level, be manifest as CAs, as described above. Those CAs, as aspects of the
    sensorium, must be functionally identical with those that are representational. There must be
    some mediating processes which, matching and checking those CAs for green in memory,
    sensorium, and representation, must deal with them as we deal with qualities, i.e., as unitary
    entities. At this point, we are, then, functionally, at least, dealing with greenness, and given the
    reasoning in 164, that is the best result we can have so far, as far as knowing the exact "feeling" of
    that experience. And this is only inadequate if one wants to actually "see" the greenness as green
    inside another's head/mind; every other condition (e.g., uniqueness, reproducibility,
    predictability), given the neural tracing ability above, is met, with these exceptions: 1) that we do
    not know, merely by inspecting the CAs, if a person's experience of green is qualitatively the same
    as ours, i.e., is not the same as our experience of red (although, given Palmer's analysis of the
    implications of the relationships in the color wheel, we do know from such inspection that the
    latter is not the same as, for example, our experience of purple - and analogous reasoning holds, of
    course, for other sensory modalities). This question is probably not answerable, as I said above. In
    addition, 2) we still do not know why we have experiences, and 3) we do not know how to inspect CAs
    to tell whether an organism has experiences at all. These may be, at least in part, answerable; we
    must postpone considering these latter questions further for now.
166) Do the mediating processes (between representation and the sensorium) have a "viewpoint"?
    This is still an open question; however, given that they are emergent entities in the above sense,
    their reactions to the CAs realizing green are likely to be wholistic, and must be different from
    their reactions to the CAs realizing red. However, what does not seem inevitable, at this level, is
    the presence of some constancy, which provides a relatively stable "standpoint" from which to
    evaluate the various reactions. That is, it would seem that a "viewpoint" does not merely require
    the above mediating processes, but some sort of constancy in those (or other) processes which
    notes but does not essentially change as the responses to the sensory (or representations of
    sensory) CAs change. Let us say that this is not the case. If that were true, then we would be, at
    best, in the position described by James and Husserl, in which there was, in effect, an eternal
    present, with no memory nor consciousness of change or time passing. The presence of
    consciousness in this kind of entity is at least open to question. At worst, there would be no
    consciousness at all, merely the changing "qualities" (the above CA-generated emergent entities),
    monitored (and created) by mediating processes. We must consider the nature of this constancy,
    then, and how it is generated.
167) Now I can argue that the mediating processes are not merely necessary, as described above (e.g.,
    161), given the interactions between representation and sensorium, but that in addition they are a
    further step in the universal abstractive processes I have considered present throughout the CNS.
    Surely processes which generalize the sensorium to categories/qualities are necessary in order to
    compare functional equals which are probably not neural equals: the greenness of the sensorium
    vs. the greenness of its representation, for example. That this is simply a process of abstraction,
    however, cannot be the complete picture; at the least, the control aspect of the mediation, if it lies
    on this level, goes beyond that. But if we can suppose that such abstraction is indeed the basis for
    at least the initial generation of the mediating processes, we can justify taking this abstractive
    process another step. Suppose, then, that we continue the mediating functions, abstracting from
    those above, which merely (we are supposing, as a worst case) note (but do not necessarily
    experience) qualities.
168) One possibility for this abstraction would be to further refine the qualities. Another would be to
    refine the operations on the qualities, i.e., to abstract the mediating - control, comparison,
    maintenance, etc., - operations between sensorium and the corresponding representation. Another
    would be to abstract the manipulating operations, that is, the synthetic operations: those where
    other modalities employ the representations as sensorium substitutes, in effect. As to the first, this
    would seem redundant on the original abstracting operations within the modality. The second is
    certainly possible, as is the third.
169) But in fact what we must do is remain focused on the function of representation, i.e., the internal
    manipulation of the sensorium. Otherwise, the possibilities for abstraction types and functions are
    just too overwhelming. Now the speculation is this: that in manipulating a representation, in
    order to avoid the above problem, that is, to know that it has in fact been manipulated, i.e.,
    changed in some way, we must have the ability to compare present and past states of the
    representation, and that comparison must be both continuous and rather extended temporally,
    because we usually have some goal that we are tracking along with the change in the present
    state. That is, in order to see that a color is changing as we specify (consciously or not), or that a
    position is changing, we must be able to note not only the present versus the immediate past
    positions, but also to note where it is relative to where its goal position is. However, we know that
    this system is, ultimately, conscious in some sense, so we must take that into consideration in this
    analysis. So there are two considerations here: first, the manipulations must be true to our
    experience, i.e., contain moments extended in time, to avoid an eternal present, which we do not
    experience, and in addition to contain the goal in some form, even a general one. Second, these
    comparisons must be in terms of qualities, however they are at present realized, if the arguments
    above are correct.
170) But in saying that we must note the "present position," or the present state, whatever that may
    be, to have a reference to compare the changed state to in order to know that the change has in
    fact taken place, and taken place in a manner suitable to the goal, we have in some sense specified
    a viewpoint. That present state is in fact a constant viewpoint we apply to the changing
    representations. But there are at least two severe limitations on this now. First, that viewpoint is
    too constant, and second, it is too limited. The first objection is easily addressed; one can vary it as
    the representation changes, etc. The second, however, points out that this "viewpoint" could be no
    more than a computer could have of its simulations.
171) But there is another aspect to this… that is, the difference between representation and reality,
    i.e., the sensorium. Returning to the fundamental reason for representation, we must be able to
    differentiate representation from reality. Now how can this be done, especially if representation
    uses some, at least, of the CAs that the sensorium uses? In addition, to manipulate
    representations, we must, unless there is a great deal of duplication (and this goes against
    experiments as well), use to some extent the CAs employed to manipulate reality, i.e., motor
    processes, etc., to manipulate representation. That is, the representations of those motor processes
    must, just as with the sensorium, use at least some of the non-representational motor CAs. But if
    this is true, then the differentiation becomes a serious issue. In addition, there is an interesting
    aspect to the voluntary (i.e., conscious) manipulation of representation. One might at first claim
    that such manipulation requires consciousness, unless one then reflects on "unconscious"
    processes, i.e., metaphorical, etc., processes, which themselves certainly employ representation of
    both the sensorium and motor processes. Then consciousness is not required for representational
    manipulation, although it may be for representational acquisition. But what of the above
    differentiation? I do not know of any difference between unconscious or conscious manipulations
    in terms of their relationship to reality. That is, are not at least some of one's unconscious
    processes the same whether in service of representation or the sensorium? To put it another way,
    we cannot, I am claiming, make the differentiation between reality and representation merely on
    the basis of differentiating the processes manipulating these nor on their content. We must make
    this differentiation independently of the processes and of the content. Thus, I will speculate that
    this differentiation requires consciousness (in the sense I am using the term, i.e., experiencing).
172) But what of schizophrenics? What of religious mystics? Are these not people who cannot
    differentiate as above? Yes, but it must be that the human brain is hard-wired for consciousness,
    and that aberrations of this sort, while they clearly involve problems with consciousness (and that
    they do is reflected in their classification as just such problems), are not such that all-or-nothing
    consciousness could be expected here. Nonetheless, I think that these and other syndromes in fact
    support this position. When one hears nonexistent voices, one is confusing representation (of the
    sort I describe) and reality: the processes generating those voices, normally in part conscious, are
    not in that case. It would seem, also, that consciousness of all the generating processes is not
    necessary, as witness the unconscious generation of phonemes, for example; but those were, given
    the learning efforts of children (if we assume that pre-verbal children are conscious), conscious at
    some point in time. In addition, for processes that are always, intrinsically, unconscious, there is
    no issue here; since in the sensorium they are also always unconscious as well as in
    representation, there can be no confusion.
173) Consider also at the representation of the past and future, that which generates the extended
    present. Here we have an example of representation which is at best marginally conscious as such,
    and inasmuch as it is not, the past and future in fact blend and extend the present: representation
    blends with reality.
174) We can look at historical distinctions made between representation and sensorium: the relative
    faintness of the "impressions" which are the result of visualization vs. the intensity of true
    sensation, cited by most from Descartes on; is there truth or a basis for this? If one considers the
    origins of the sensorium and representation, we see that the sensorium must be driven by the
    environment in some form or other; it must be, originally, a response to something, otherwise it is
    meaningless to hold any form of ontology with an external world, or indeed even to have a
    sensorium. Whether subsequently the sensorium is influenced by internal states is irrelevant at
    this point; its function and therefore its origin is to inform us about the world (which includes our
    internal, bodily world also). Now what this implies is that it must be generated "bottom-up," i.e.,
    with what I have termed "individuate" entities, from particulars, in contrast to being generated
    from abstractions. The thrust of the whole above analysis is toward the generation of successive
    abstractions, which then may, surely, turn and influence their generation.
175) Representation, in contrast, is internally generated. But this implies that it is not generated
    through particulars, through individuate entities, but from abstractions, i.e., "top-down." To
    further support this point, if the function of representation is the internal manipulation of the
    sensorium for the purposes, in part at least, of prediction, testing, etc., then the particular goals,
    rather than the particular sensory aspects, of that representation must govern its generation. That
    is, merely to reproduce the sensorium would seem useless, wasteful of resources, and simply
    confusing. Thus, in any particular situation, representations are generated in order to fulfill some
    functions, and it is just that orientation toward some goals, however specific or vague, which
    determine the representational focus, in effect, i.e., what is either being generated or is being
    utilized at that time. Even in the case of metaphorical processes, which we may assume are largely
    unconscious, representation (which as I will argue, at this level may not be distinguishable, when
    they are unconscious, from manipulations of the sensorium) is goal-fulfilling. This function, then,
    supports the top-down generation of representation. But if these different origins and purposes of
    representation and the sensorium are true, then it is clear that there is a basis for the 1) relative
    sensory clarity and intensity of the sensorium, and perhaps the vagueness of associated goals, and
    in contrast, the 2) sensory vagueness but goal orientation of representations. One would expect, if
    this is true, that representations, especially since they must, as I have argued above, share the
    sensoria's qualitative CAs, would be less intense: they are generated from the top of the
    abstraction process hierarchy downward; they function in relation to the abstract goals, and
    therefore sensory intensity is not needed for them, although of course it is not forbidden either. In
    addition, the induction, from higher-level abstractive processes, of particular sensory processes
    (remembering that they must, due to lack of neural duplication, use pretty much the same CAs
    that the sensorium uses) would not seem to result in those lower-level processes necessarily being
    as intense as when they are driven from the sensory organs.
176) Does this mean that it is easy, then, to distinguish between representation and sensorium?
    Surely, however, there are cases in which the sensory aspect of representations is as clear and
    intense as with sensations, and cases in which abstractions are clearly present in sensation. Thus
    those general differences between sensation and representation would not seem sufficient for
    distinguishing them, at least reliably.
177) But the issue here is more general. Let us recapitulate. I have argued that the processes
    throughout the CNS, especially at the higher levels of abstractive processes, consist of CAs which
    result in dynamic emergent entities because of the relaxation constraints on the formation of CAs.
    Both sensorium and representation must consist of such processes, I have claimed. In addition,
    there must be comparison between sensorium and representation, through some sort of mediating
    processes, which must in addition, at this or a higher level, establish some sort of baseline or
    reference level in order to generate knowledge and awareness of change as such. Now in order to
    compare sensorium and representation, we must compare similar "things" in order to verify that
    the representation is accurate, etc. But this involves several entities that are not "things" in a
    normal sense. First, the sensorium has qualities, whether experienced (at this point) or not, which
    must be compared to comparable qualities in the representation; and similarly for higher-level
    abstractions, concepts, perhaps, associated with and/or attached to and/or present within those
    sensory qualities. Examples of qualities might be the green of a lawn, the experience of seeing or
    visualizing the edges of a cube, etc. (I will, pace Dennett, consider that we do indeed experience
    qualities, although it is certainly difficult, perhaps impossible, taking his Wittgensteinian point, to
    "pin down" any particular in those qualities. Why this should cast doubt on qualities as
    reasonable, if not precise, particulars and classes of experiences, however, I do not know. But then,
    I do not accept Wittgenstein's private language argument either (see also 191 below).)
178) Now those qualities, let us say, are realized neurally as CAs of the type described above. Yet
    clearly those CAs are highly variable even within either sensorium or representation. That is,
    green, say, as realized by a particular CA is surely not realized by that exact same CA the next
    time it is seen, even if the green color in the world (i.e., the light in all its complexity) is exactly the
    same as before. So the CA realizing green is not identical over time or shades of green, nor is the
    CA realizing the representation of green. The neural comparison between representation and
    sensorium, of green, then, must involve the dynamic emergent entities generated by CAs in such a
    way that those entities respond "similarly" (to be analyzed later - 218) to different CAs realizing
    the same qualities; changes in those latter CAs cannot be reacted to in a "clearly different" manner
    by the mediating CAs until those changes pass some sort of "boundary or threshold." In addition,
    the response of the mediating CAs must be "similar" to the same qualities in representation and in
    sensorium, an even greater difference, given the different sets of neurons involved in those
    processes, and the corresponding difference in qualities, described above. So this comparison
    cannot be something simple, like counting neurons, describing similar patterns of neural
    discharge, comparing discharges or patterns of neural firing to a template or even set of templates,
    etc., given the physical dynamics of the CNS and variability in stimulus, etc.
179) Creating a representation, then, can not be fully explained by its low-level components. If a
    representation is created or employed neurally, i.e., as a set, however dynamic, of neurons, then
    evaluating it is no more than any machine can do. However, if a representation is created and
    employed in terms of the real-world functions that the corresponding sensoria realize, then it
    must be considered to be treated, at least, in terms of qualities. We must thus ask just how the
    CNS must treat representations (and perhaps also the CAs in the sensorium) in order to fulfill
    their functions as representations. But this is an interesting point; after all, the CNS, or any other
    CAs, cannot pick up and rotate another CA, cannot change its color, etc. Yet on the other hand, the
    operations are performed with just those goals "in mind," and set forth in terms of those goals. The
    "system" must "understand" in those terms in order to apply the processes appropriately. What
    then is a "process"?
180) Let us see what a "rotation" might consist of. It does not seem possible that a neural net
    corresponding to some position of a visualized cube, let us say, is trained in any sense, on the fly,
    as we perform the mental operation of rotation. That operation is too easy and happens without
    error and indeed many times without effort. It may be that there is some sort of successive
    "replacement" of the CA representing a cube with CAs representing cubes intermediate in the
    path of the rotation, but this also does not seem to correspond with how a neural net functions.
    That is, there are no "images" or even, it seems, specific CAs which could somehow "replace"
    another CA, indeed, the concept of "replacing" here is very strange in neural net terms, where we
    are dealing with the dynamics of the activation of neural sets rather than with images of any sort
    (but more on this later - Edelman's simulations with pictures of shapes neurally realized and
    similar evidence, however, actually favors the functional explanation below). After all, virtually
    the same neurons, virtually the same CA, would be used to realize any possible set of rotations,
    say, for the same cube, merely because that is how the CNS, as a neural net, works (note, for
    example, the sets of neurons for angles). But if this is true, then there is really only one way that
    such a process could work: there must be stimulation of successive sets of neurons corresponding
    to the various angles of rotations of the cube. This raises some interesting questions, however.
    First, why successive? It has been demonstrated that the greater the rotation, the longer it takes to
    visualize it: the cube is literally being rotated in, let us say, "mental space." But this mental space
    must be realized through the activation of CAs which incorporate neural sets corresponding to the
    various angles. Those sets have, effectively, been shown to exist (e.g., the groups of neurons
    stimulated by lines at different angles, etc.). So why the successive activation? Why, given that one
    wanted to rotate through 90 degrees, could the sets not switch rapidly; after all, other images can?
    We conclude that the reason is that the cube is rotated, mentally. But here we are dealing,
    supposedly, with no mentality, and perhaps no qualia, only with CAs realizing such things as
    cubes. This is evidence, I claim, that the successive activation of CAs corresponding to successive angles of
    the cube is a result of the CA realization of the cube and the CA realization of the operation of rotation being
    handled as those operations are handled when we deal with the world; handled as we handle things through
    the sensorium, as qualities.
181) That is, the CA realization of the cube is not literally "handled" - clearly we cannot embrace the
    old Gestalt explanations, Edelman's simulations notwithstanding - but then, neither is a cube
    itself, as far as neural sets - CAs - are concerned. We are, as I argued above, dealing with the
    differences between bottom-up and top-down driven CAs when we speak of the general
    differences between sensorium and representation in this context. Nonetheless, given that CAs
    correspond to qualia, at least functionally, at this point, and that they are dealt with as qualities,
    seems to imply that operating on a color, a shape, etc., in the sensorium, i.e., when seen as part of
    the world, is done with processes that employ the CAs as operations in the world on qualities in
    the world. Thus if the CA corresponding to a representation of an object at some angle is changed
    by processes to a CA corresponding to a representation of the same object at another angle, this
    change is performed by the processes (or by very similar processes, allowing for neural variability
    and for the top-down shift described above) which operate on the actual object. Which means:
    performed by the processes which operate on the CAs induced by the actual object, that is, by
    motor processes, guided by visual and kinesthetic feedback, where motor processes are effector
    CAs, and visual and kinesthetic feedback are, of course, CAs induced by those sensory modalities
    as they interact with the objects. This last is very basic neurology when directly applied to the
    sensorium, but when this is transformed into representation, those same operational and CA
    parameters apply.
182) One could still argue, however, that qualities or anything corresponding to them need not enter
    into this description; that one can describe the operations with the world in terms of CAs, and that
    those can be analyzed into neural net firings with no necessary correspondence, whatever we
    actually experience, to qualities. However, the key here is the difference between representation
    and the sensorium. In order for representation to correspond (and the system to verify that) to the
    sensorium, insofar as both objects and operations are concerned, there must be a correspondence
    between bottom-up induced processes and top-down induced processes, both functionally and
    descriptively.
183) In the sensorium, as we have seen, CAs are induced and progressively generate abstractions. In
    a representation, the abstractions generate the progress (regress, in a sense) to particulars. The
    abstractions employed in representations must be those (or functions of those) generated by the
    sensorium; that is what is available. But those latter abstractions must also be what correspond to
    qualities at various levels of abstraction, since the sensorium is what generates and is employed to
    furnish qualities to provide the objects, relations and processes we experience as the world. Thus
    in order to generate and operate on representations, since top-down induction through
    abstractions of processes and qualities of more specific abstractions and qualities is employed, the
    representations do not merely correspond to the sensorium, they correspond in this particular
    way: such that they are generated from and composed of those CAs which must correspond to the
    qualities we operate with and experience in the world (which can, of course, be very abstract,
    beyond what are usually termed "qualities" to abstractions). That is, they are initially and
    primarily generated by and composed of CAs that function, internally, as qualities.
184) Now the question still remains, and can be put in this form: what of a digital computer
    simulating the above CAs, etc.? Why would not (or would it not) this simulation have mind, or
    qualia? This of course is just another version of the "hard problem," and has still not been entirely
    answered. More on this below.
185) Let us look again at the comparison between representation and sensorium. Now it is clear that
    despite the picture of representation and sensorium as operating more or less in parallel,
    connected by some sort of executive system, this cannot really be true. The overlap between CAs
    in those functional systems must be great, for reasons given above. Thus a comparison between a
    representation and a simultaneously induced sensory object (e.g., an image or perhaps even a
    memory) must be, to a great extent, a comparison between abstractions which induce particulars
    on the one hand, and some of the same particulars inducing some of the same abstractions on the
    other. But the problem here is just how that comparison might be realized. What, after all, is a
    "particular" or an "abstraction" but a set of neurons firing? How does a system "compare" these
    sets without 1) some sort of statistical matching, 2) some sort of counting operation, 3) some sort
    of output comparison, 4) an observer? But none of these seems appropriate here. This, as I see it,
    encourages the picture above. That is, to compare, in this case, cannot be to match equal sets,
    particularly of neurons, since the sets employed by representation cannot be the same as those for
    the sensorium; but it could be to match sets of CAs. Considering CAs as corresponding to
    qualities, as above, we might use the particulars to induce abstractions and compare those to the
    abstractions used to induce particulars, and vice versa. We might simultaneously do the same
    thing with memories. But we must look more closely at the nature of comparison.
186) How then are CAs compared, in general? One method might be the simultaneous induction
    from two or more sets of CAs into the same set of neurons (i.e., into a single CA), another might be
    the simultaneous induction of the same CA (set of neurons) by different CAs. Both of these
    methods require yet another CA, which notes a change in a set of CAs that is activated. The first
    method might occur by noting if there is a disturbance in the regularity and/or the stability of the
    dynamics of a set of neurons which is more or less continuously firing. If that CA is maintaining
    contact of some sort - interacting - with another stable CA configuration, a disruption of that
    stability could initiate any number of results. Thus, it requires the disturbance of a stable dynamic
    pattern, in this case, to note difference. The second method seems to be pretty much the same,
    since in order to note that "the same" CA was induced by different activating sets of CAs, that set
    of CAs induced would have to be treated as potentially different sets, and those compared
    analogously to the first method.
187) Let us consider this in more detail. If two CAs are to be compared, they are induced
    simultaneously - if they were not, we would merely be faced with a regression of this problem
    with, say, a memory (which must itself be a replication - for a "partial" memory or "symbol" of a
    memory would then have to be connected somehow with its original - how otherwise would it be
    a memory "of" that original?) of one of the CAs. That is, suppose they were not induced
    simultaneously. Then how would it be determined that they were similar? A "memory" of one
    would then have to be compared with the other. But if this memory were not a complete
    reconstruction of the original, how would the comparison determine the extent and nature of the
    similarities, i.e., whether the two were indeed different, if given this limited set, they were found
    similar? If all the sets and subsets of the CAs could be "symbolized" and this symbolization
    compared with the other, one might argue that a comparison could proceed, but what, neurally
    speaking, could a "symbolization" of a neural set consist of, except yet another neural set? Thus
    the problem regresses. Now given the configuration of the CNS, especially the cortical
    configuration, since what are being compared are sensorium and representation (in the sense I
    have discussed at length) of the same object, much the same CAs, from abstract to particulars, are
    induced. What should happen, then, if they are very similar? In this case, the induced CAs,
    downward and upward, will be closely matched if not identical neural sets. That is, the neural sets
    will literally be realized on identical or nearly identical neurons - there is no duplicate cortex, as I
    have argued above. This will either reinforce the induction or have no (or little) effect on it - no
    negative effect, at any rate (the caveat here, of course, is "given our present level of knowledge of
    the CNS"). Suppose, on the other hand, that the CAs are dissimilar. Then, because of the focusing
    and abstracting processes described above, there will be (with the caveat) mutual inhibition
    between nearly similar CAs, and both will be interfered with to some extent. Suppose the sets are
    widely dissimilar. Then there will be two largely independent sets of CAs induced, again a
    different outcome from the case of similarity. Now one implication of this kind of comparison is
    that there is some set of CAs which is noting, in some way, the interference or lack of it between
    the compared sets. We will return to this later.
188) In addition, the idea of "comparison" presupposes to some extent a prejudgment of sorts about
    the CAs to be compared, e.g., that there is a possibility of similarity. But how is this done? What is
    "similarity"? We are, after all, talking about sets of neurons, and on the face of it one neural set
    (and here I refer not to the physical neurons, which after all are completely invisible to the CNS at
    the level of coding or information - we might except the chemical level, but that seems largely
    irrelevant to the considerations here - but to the dynamic neural codes, the ongoing impulses)
    seems, especially given dynamic variability, indistinguishable qualitatively from another. But is
    this really true? Clearly, different sensory modalities occupy different locations on the cortex, etc.,
    but how can the CNS, the internal system itself within the cortex (etc.), note an absolute location?
    That seems impossible. However, relative locations are functionally distinguishable because, at
    least, functions located closely to one another will interfere with one another in the above
    similarity tests. In addition, it must be the case, given all the arguments above relating to sensory
    modalities, i.e., relating to lack of neural code qualities (no colored neural impulses, etc.), that it is
    the interrelationships, the pattern, of neural dynamics which distinguishes modalities and
    qualities within modalities. And indeed, at this point we can see how those relationships (e.g., the
    various color relationships, as realized in CAs) might be realized, i.e., embodied: as dynamic
    emergent patterns on the CAs realizing them (Or on abstractions - generated through the normal
    abstractive processes - of the original large sets of CAs. One might speculate as to whether the
    abstractive processes, operating normally on a neural level - i.e., with its units individual neurons
    - might operate with CAs as units instead, but perhaps our CNS is not large enough for this.). That
    is, this is precisely a conception of how a gestalt-like entity may in fact be created, in contrast to
    gestalt-type characteristics, which are created, it seems, during the processes of abstraction. Thus
    there is indeed, given a level of monitoring of sensorium/representation comparison which seems
    necessary from the argument above, reason to believe that modalities and (neural realizations of)
    qualities are distinguishable (from other modalities and qualities). Now given that they are
    distinguishable (and thus at least roughly classifiable), and again given that there is some
    mediating set of CAs, one could conceive that a rough comparison could start this process.
189) I would like to digress a bit here and note that it seems, at this point, that Searle's speculations as
    to the necessity of the CNS (or some kind of analog net) underlying the mental might have some
    basis. That is, given CAs of the type described above, with emergent dynamic entities realized on
    them, it is clear, of course, that conventional digital computers cannot embody such entities. Their
    physical circuitry is incompatible with the requirements for such networks. Digital simulations,
    likewise, could not embody such entities, and although they might be simulated, such simulations
    would indeed be analogous to the simulation of gasoline (Searle's example), since no true entity,
    as would be the case in a neural network, would be created. Now one might reply that such
    simulations, in contrast to those of gasoline or of asteroids, for example, would be knowable, i.e.,
    completely analyzable. But this is not true either, for reasons having to do with the detailed
    structure of the cell and with recursive analog coding that I will not go into again. However, let us
    suppose that we were to build a digital network along the lines of the CNS. Would this be capable
    of creating similar emergent entities? One problem with that network, as I see it, is that digital
    codes have, in effect, holes in them. Strings of discrete symbols have spaces between the symbols
    and the words of the code, and thus feedback arriving would have to be discrete, timed to
    coincide with particular parts of the outgoing codes or with code processing. In addition, a code
    cannot literally act on another code in a digital system, as it can in an analog system; the codes
    must be interpreted. Interpretation takes time, but worse, digital interpretation is a set of
    processes which, once started, cannot be altered. To put this another way, digital codes cannot
    directly interact; a processor must receive all of some code string and all of another in order to
    have them interact. What if, during that process, another code string arrives? Then the processor,
    in order to interact the three, must alter the result of the first process, that is, it must start over.
    The CNS does not have this constraint; codes are processed as they arrive, and, since they are state
    changes in the processor, are part of that processing. What unites such processing into an entity is
    the inclusion of the codes into the entity, so that as code arrives at a cell body, that body and its
    output alter. Given those constraints, could digital dynamic entities be realized instead of merely
    simulated? Perhaps, but it seems that at best this would be like successive stop-motion
    photographs of corresponding analog entities. Another problem here is that of the retention of
    qualities. Digital codes, like analog codes, have no qualities; they must be assigned through labels.
    We have seen, however, that analog code patterns ("dynamic emergent entities") might realize
    qualities; without these entities, the problem of qualities seems severe for digital systems without
    external interpretation (see 191 also).
190) Let us take another example from AI. Suppose we have a chess playing computer, which does
    internal computation of chess moves and then arrives at a solution and moves. Does this device
    represent in the sense I have been using the term? I must answer in the negative, for two reasons.
    First, the computer does not discriminate between the "inside" and "outside" worlds; there is no
    difference, as far as the computer is concerned, between representation and actual moves. One
    might object, however, that the computer does not create output until it finishes evaluating, and
    that surely this is the essence of the difference? I would reply, secondly, that for the computer,
    there is no external manipulation; the only manipulation it does is internal. It's "interaction" with
    the outside world consists only in receiving inputs and generating outputs. It may have a "chess
    board" in memory, but that is merely a reference point for running its processes, a memory only.
    So the "representation" which takes place "internally" is the only manipulation there is for that
    machine; it might as well be only external; there is no interaction between internal and external
    manipulative processes because there are none of the latter, so there is no representation in the
    sense I have been employing the term. I might also object to terming what a chess playing
    computer does as representation precisely because those processes do not in any sense represent.
    That is, since there is no external world (or, in a sense, there is only an external world - given that
    one can say that there is any world at all) for such a computer, as I have argued above, there is no
    possibility of abstracting that world to representation.
191) It does not seem likely that what I have been terming "mediating processes" are merely further
    sets of abstractions built on the sensorium and representation; if that were true, then why would
    they be any more separate from those than any other aspect of those processes? Yet there must be
    some degree of separation, since they must to some extent evaluate those processes, as we have
    seen, and in addition serve to some degree as an "executive process." This separation, while
    bringing up the question of an interface, does allow the above evaluation to occur without regress,
    however (I begin to address this in more detail at 270-271).
192) Another caveat here relates to the nature of what I have been terming "emergent dynamic
    entities." It would take a delicate balance between positive and negative feedback to maintain
    these entities indefinitely; also, theories of memory utilizing perpetually firing loops have been
    disproved; also, it seems that the abstracting processes described here are largely conservative;
    also, it seems that the RAS has to be employed to activate the cortex - without that activation, the
    firing (in certain areas) ends. Because of these reasons, it does not seem that any particular entity
    of this sort could be maintained indefinitely (or even for very long), without outside stimulation,
    probably. This does not mean, of course, that membrane biases cannot exist that would recreate
    them; but any theory dependent on the continual existence of particular emergent dynamic
    entities would seem to be dubious (one could see this argument as both supporting and
    undermining Dennett's arguments about qualia; if they are realized by such entities, they will
    never be entirely reproducible or accurately describable, yet they will exist and function). Why
    then do I think that they are desirable, even necessary? First, they provide a basis for the temporal
    continuity of qualities over short time periods, at least. Second, they provide a basis for the
    comparison of qualities in neural nets, without having to have a neural net attempt to simulate
    digital operations, e.g., storing particular numeric values and memory locations, i.e., a "table" of
    values, in order to compare them later to values from similar tables; in addition, there is a basis for
    this comparison being analog (an infinitely long table, in effect) rather than digital. Third, this
    same basis for comparison, then, rather than necessarily employing dimensionless (or actually,
    unidimensional, since they are numeric) qualities can, given operations on different dynamic
    entities, retain qualities (as patterns, given the argument above) and compare them directly. The
    fourth reason, then, is that the temporal continuity and simultaneous comparison of (entities
    corresponding to) different qualities is possible with these entities. Thus, these operations must
    either be fast, or must be actively sustained (see 85).
193) Now Damasio and others notwithstanding, it is not possible that representation is merely the
    evocation of parts or aspects of the sensorium, and consciousness merely the addition of one's
    sense of "self" and/or bodily sensations. It has been made clear, for example, that schizophrenia is
    a malfunction of one's self-assignment, an aspect of consciousness, certainly; but without the
    assignment of self to, for example, internal speech, one is conscious of such speech as externally
    originating (or not conscious of it as internally originating), but one is still conscious of it; a
    component has been lost, but not consciousness nor even consciousness of the speech. See also
    Capgras syndrome; this is a similar situation to schizophrenia, but emotions are not integrated
    here, yet the person is still conscious. Thus, syndromes in which data is missing do not seem to
    relate to being conscious but merely to the content of one's experiences, which is what one would
    expect if data were integrated into or experienced in, rather than responsible for, consciousness.
    Therefore it does not seem possible that consciousness is merely data-driven; it must be related to
    circuitry structure and dynamics.
194) In addition, the differences between representation and the sensorium have been gone into at
    length above. Further, as I have mentioned, there is one difference which seems easily overlooked
    by the cognitive/DP paradigm, and that is the difference between willed and unwilled, i.e.,
    consciously caused and non-consciously caused mental events. Sensory events are imposed on us;
    representations are, at least in part, consciously willed, even if at times indirectly. That is, if one
    wishes to visualize someone's face, one attempts to visualize that face on the basis of a name, a
    person, a relationship. These abstract, willed, considerations then drive the induction, with, we
    note, usually some conscious effort, of the particulars of the person's face. In stark contrast, when
    we see that person, their face is imposed on us, and their relationship to us strongly induced by
    that presentation. There may be exceptions to this, as I have mentioned: the unconscious
    metaphorical processes may not be driven at all by consciousness; although, given the results on
    attention (e.g., "Inattentional Blindness") it would seem that consciousness at some point, if
    indirectly, drives metaphor and thus representation.
195) Another aspect (and implication) of the difference between representation and the sensorium is
    that between tools and other objects. Tools must result from the kind of abstractive processes
    described above, applied to the interaction between representation and sensorium, since there are
    no tools until goals, purposes, and various types of manipulations of the environment are
    abstracted beyond the objects' immediate meanings. A tool as a shaped object is a kind of
    realization of representation on the world, as Heidegger has emphasized; and this process
    supports the kind of high-level abstractive and integrative processes I have described here, and in
    addition the difference between representation and sensorium. Now the origin of the tool,
    speaking phylogenetically, is another interesting indication, perhaps, of the origin of
    consciousness. That is, if a tool is a representation which is realized in objects, such that an object
    "takes on" some meaning which must originate in a representation and not in the object as one,
    without representation, interacts with it, then the most primitive tools are not shaped objects, but
    natural objects utilized for implements and toys, and surely this might indicate that other animals
    (than human) are conscious. Thus, one can claim, I maintain, that a test for consciousness is the
    ability to play, in which an animal understands another animal or object as something else: a play
    threat or prey, usually. If this is true, then consciousness descends into the animal kingdom
    perhaps to birds, certainly not to reptiles and fish. That is, one must differentiate "tools" such as
    those ants employ, or even the shaped parts of flowers, from objects clearly temporarily and
    reversibly assigned meaning apart from their apparent meanings, such as toys played with by
    cats, etc. When a fish or reptile builds a nest, is this nest a tool? Only metaphorically, since the nest
    becomes an object with fixed (not temporary or reversible) function.
196) Damasio (et al) also talks about "convergence zones." These must of course exist, yet the
    difference between these and the functionally similar areas within cortical modalities would seem
    to be minor, including the distance factor. That is, distance involves no more than somewhat
    longer delay in transmission between CAs, effectively decreasing the relaxation time necessary to
    form an emergent entity (i.e., making it more difficult). And indeed it is clear,
    phenomenologically, that it is easier to separate inter- than intra-modality sensory experiences
    into components. Thus it seems that these "zones" perform the same functions as the intra-
    modality convergence zones, namely a continuation of the abstractive processes within a modality
    to those between modalities, which are clearly necessary to form integrated sensations and
    representations.
197) Now there are, then, two general components involved in the transformation of CAs to qualia.
    The first, as we have said above, is a viewpoint, as an alternative to specific mental qualities or
    aspects. That is, if the problem of external observation is valid, then an external observer looking
    at a neural net cannot know just what it is experiencing, although some parameters of that
    experience may be known, as I have argued above. The problem then becomes whether one can
    describe in general what is necessary for a viewpoint, i.e., the general characteristics of neural
    circuitry responding to experienced qualities. More on this later. The other point involves qualities
    per se. I have argued extensively above that qualities are realized by singular entities, i.e.,
    emergent dynamic neural entities. The reason for this is that if the dynamics of such a CA were
    capable of being treated in terms of singular neural firings or neural codes, i.e., the dynamics any
    individual neuron were meaningfully separable (such that the individual dynamics could be used,
    as individuals, to generate - compute, if you will - the dynamics of the CA) from that of other
    neurons, then that CA would be functionally, at least, separable into its components. In order to
    counter that and to establish a neural basis for singular entities corresponding to qualities, I have
    argued for the existence, in at least some circumstances, of the above emergent dynamic entities.
    Now how is such an entity, which is, after all, a set of neurons firing, to be treated singularly? It
    would be the case that such a set were treated as a singular entity (E) if the interaction of some CA
    (D) with that set E, whether that interaction was by means of a subset of that entity (Es) or the
    whole neural set E, proceeded such that the response of Es to D was governed by the rest of E; i.e.,
    the relaxation time of any particular response of (Es) were such that the influence of the rest of the
    entity (E) on that response was possible before the end of that particular response of that subset
    (Es). In that case, the set D could not meaningfully interact with the subset Es alone, since that
    subset's (Es) interaction would be influenced, as it (Es) interacted with D, with the rest of E. This
    would seem to give us a basis for saying that there can be complex singular entities (in part,
    surely, resulting from the dynamics of "convergence zones") which are emergent from certain
    neural circuits, which I have been terming CAs. That is, it is not enough to show the possibility of
    such entities within a CA, but it must be shown that it is possible that such entities interact as
    wholes with other neural sets. Given that the argument here is reasonable, such entities are
    possible.

198) Now suppose we do a little reconceptualization and conceive of representation as a more active
    kind of process than that implied above. Thus, the sensorium should, I think, be conceived as
    driven to produce representation, or as in some sense a generator of representation, rather than
    representation being the rather passive result of an "inward turning" that I had hypothesized. In
    that case, we look for the source of the driving process, and one natural candidate is the ERTAS,
    since without that the cortex shuts down. The question then arises, what about the ERTAS
    originates specific driving processes? Here we have two basic alternatives: one is some kind of
    internal reverberating circuit (because if it were not constantly running, it would have to be driven
    itself); the other is a fairly constant internal source of stimulation, the obvious candidate being the
    body, through monitoring of internal movements and processes. The advantage of the latter is
    that those internal processes, as the driving "force" for representation, would make it easier for
    representation to incorporate internalizations of manipulations; indeed, it might imply that
    representation, or its changes, was derived from such manipulations, which would fit nicely with
    speculations from cognitive science. Another question that arises naturally is that of the
    "viewpoint" I had mentioned above. Given that the body provides a constant source of stimulation
    to drive representation, some form of this constancy could also serve as a viewpoint. On the other
    hand, this latter would be as easily created by a reverberating circuit.
199) The question, again, is what exactly is inducing representation? That is, how can the sensorium
    alone induce representation; similarly, the bodily sensorium, whether a source of stimulation or
    not, seems essentially the same as the external sensorium in this regard: a set of processes which
    analyze and synthesize from the environment toward its manipulation. One might think that
    some symbolic facility, our language faculty, if that truly exists, would induce representation, yet
    why would it need to? For a computer, for example, symbolic processing alone is sufficient; why
    is it not for natural language? Yet one must concede that the pure formal manipulation of
    symbols, assuming that faculty exists, cannot relate, without some sort of mediating processes, i.e.,
    something to assign meaning to symbols which in many cases are arbitrary, to anything outside of
    that formal manipulation. It would seem, however, that this mediating faculty would be necessary
    prior to pure symbolic manipulation for that manipulation to have purpose and application at all.
    Yet surely it is an advantage for us to be able to represent, i.e., to internally manipulate the world,
    in order to plan, anticipate, etc., not to mention learning by play. So representation's origin seems
    difficult to conceive, unless it is envisioned, initially at least, as an extension of the abstractive
    processes normally occurring in the sensoria. Yet those processes are most likely passive and
    conservative, as we have seen, and representation is the opposite.
200) Now we can take animals below the consciousness threshold, as I conceive it, and note that not
    only do they not have representation but that this lack of representation results in a certain type of
    inflexibility. That is, the lack of the ability to internally model and transform the sensorium in
    order to modify the external manipulations of the sensorium limits them to learning through
    classical conditioning processes. Inasmuch as this internal manipulation can proceed, conditioning
    may be replaced by procedures which anticipate, at least, conditioning procedures and thus can
    modify those procedures. Without this ability to model, however, the organism, whatever it is,
    cannot repattern its learning procedures even to this extent. How far can this flexibility derived
    from internal manipulative ability, i.e., representation, go?
201) Suppose that we wanted to conceive of an infinitely flexible organism. Then we would have to
    be able to change a fixed set of procedures and/or patterns. But this change could only be
    accomplished through another set of procedures, or randomly; so we would have to be able to
    change the procedures for change also; but we would have to change the procedures for change of
    change also. Now how can we avoid regress here? First, we could stop arbitrarily (insects, etc.).
    Second, we could introduce randomness at some point (which is what many animals seem to do
    (rats, cats, dogs, etc.), i.e., make random choices when all else fails). Third, we could make the
    system recursive, so that the change procedures B(1) used to change procedures A(1) were acted
    on by the procedures A(2) resulting from the application of B(1), so that they became B(2), which
    then changed A(2) to A(3), and so on. It seems that this latter might if properly set up be the basis
    of a non-converging infinite series, e.g., a chaotic path around an attractor basin, which might
    actually approximately realize our initial goal of an "infinitely flexible" organism, to the extent, at
    least, of producing behavior (etc.) which holds to a pattern, i.e., moves toward or remains focused
    on a goal, while being infinitely flexible in the sense of not necessarily repeating itself. This would
    seem, actually, to be an ideal solution, since it incorporates both the achievement of the goal as a
    constant direction and the ability to not repeat the means of achieving that goal.
202) What would the neural realization of such a recursive procedure be? The neural realization of
    the procedure of abstraction is described above, more or less. We have speculated that this is
    stabilized through top-down conservative procedures; but what if we wanted to destabilize them
    in this kind of controlled manner; could this be done within this neural context, as part of either:
    the built-in hierarchy of processes, or something having to do with representation? If not, it is
    probably too ad hoc. It might be interesting to assume that classical conditioning processes work
    directly on these abstractive/stabilizing processes, altering one or both of the microscopic top-
    down/bottom-up procedures described above. But this does not seem correct; classical
    conditioning usually alters "valence": some emotional aspect of a "stimulus" or "response" so that
    it is avoided or sought; it does not change the world seen or manipulated, or those manipulations,
    in any fundamental sense, but seems instead to change the ones applied. Pigeons playing ping-
    pong have been trained to find a rather indirect environment to peck for seeds, not to behave as
    the result of representations in the sense I use the term (This is, by the way, a possible criticism of
    Damasio's inclusion of emotion in reasoning; surely there is some disjunction between reasoning
    and classical conditioning.). But this implies that representation does alter those microscopic
    processes.
203) Now we know from the above that representation must 1) invoke processes (realized in the
    CAs) that the sensorium invokes, since we have no duplicates of the CAs realizing these
    processes. In addition, 2) we must be able to differentiate those invoked processes from the
    originals, or we would not be able to differentiate representations from the sensorium. In addition,
    3) the invocation of representations must modify the processes of the sensorium, since the
    manipulations which are the purposes of the representation must be different from those
    manipulations actually going on and those which have gone on in the world, or there would be no
    point in internalizing the representations at all. That is, manipulations on the CAs employed by
    the sensorium, as representations, represent hypotheses, extrapolations, fantasies, etc., and thus
    must alter somewhat the CAs invoked by the world. Thus, we must conclude that the process of
    representation, or the processes invoking representations in the CNS, must not invoke the actual
    complexes of ongoing sensory processes, but must create others different from those ongoing. In
    addition, not only do the representations modify the CAs involved in the sensorium, but given the
    necessities of the system, there must be feedback from the sensorium to the representations. We
    have a possible basis, then, in the generation of representations, for the recursive modification of
    CAs.
204) But what is initiating the modification is not a sensory input but an internal state, and this
    internal state must, in some cases at least, be some sort of realization of a goal state, since the
    internal manipulations giving rise to the representations arise in order to find out how to realize
    that state. Yet that state cannot be realized except as an abstraction, an abstract high-level idea, in
    most cases, since its realization is a problem to be solved. The reciprocal dynamic between
    representation and the sensorium on which it is imposed must rest on differences between these
    two. But if they are different, where does the initiating pattern, goal, idea, come from to invoke the
    representation? That is, one can say that we employ "internal manipulation," but this is not
    specific enough. Certainly, there must be a goal as an abstraction, but the realization of this goal in
    manipulation must be realized through some kind of initial manipulation or abstract idea, again,
    of what that is to be. Yet if manipulation is performed through the sensorium, how can it
    originate? The sensorium either senses patterns in the environment or employs those either stored
    as CAs already formed or patterns invoked through representation, which takes us back to the
    question. Where do new patterns come from; how are they formed? It seems that one is forced
    into the position of maintaining that there must be a separate area for the manipulation of high-
    level abstract patterns, based on abstract goals, in order to initiate lower-level, particular patterns
    in the sensorium. We are forced, it would seem, into hypothesizing another area(s) of the brain
    devoted to abstract symbolic manipulation. (Another extremely speculative reason for believing in
    the necessity of such areas is that when we dream, we must actively inhibit muscular actions. In
    dreams, it seems, the sensorium is directly evoked, with the consequence that, instead of dreams
    consisting entirely of representation (in, of course, the sense I have employed the term), they
    consist in part of actual manipulations. The necessity of muscular inhibition in dreaming then
    would seem to support the hypothesis that in waking states in which we employ representation,
    we use (in part at least) something other than the sensorium, since we are not in the paralysis
    necessary for dreaming. Fortunately for this hypothesis, we do indeed have areas devoted to
    abstract symbolic manipulation (a possible implication of this is that animals which dream a) have
    some sort of capabilities for abstraction, and b) are thus conscious).)
205) In addition, it is not enough, even if we could, to be able to point to a neural circuit that
    fluctuates in time with someone's consciousness. It is not enough to be able to construct a neural
    circuit which has a "viewpoint" in the sense of reacting as a whole to another circuit which
    presents itself as a whole, although we can see, at this point, a way leading to this. There must be
    also a way of incorporating into the above conception of neural dynamics the understanding how
    a circuit actively interprets specific incoming information, at least in some general sense. But what
    is it to "interpret," on a neural level? At one level, the incoming impulses from a particular sensory
    modality are interpreted as belonging to that modality because of their input (into the interpreter)
    location (since the interpreter cannot know their ultimate origin); and in addition, they are
    interpreted as belonging to a modality because of the unique patterns of impulses from that
    modality, which also determines how they are abstracted. That is, information about color, as
    neural impulses, is combined with information about shape (etc.) to generate an object. But this is
    a passive (for the most part) kind of interpretation, brought about by the structure of the CNS.
    What of active interpretation, i.e., interpretation which has flexibility, i.e., choice, in effect;
    interpretation which can actively assign meaning to representation? What kind of neural circuit
    could either put information from different modalities (or from one) together in different ways, or
    could assign information from some set of modalities arbitrarily to information from some other
    set of modalities? What would a "higher order" or "second order" interpretive circuit look like; a
    circuit whose function (content? structure?), in a sense, is the potential to interpret? That is, not
    only do we have to take into account the specific kind of information (color, etc.) in a set of
    impulses, but we have to do this in an active way, particularly in order to have representation.
    Representation, as I have argued above, has, among other functions, that of extrapolation. Surely
    here, at least, one must actively assign and combine functions from the sensorium.
206) It is possible that we might extrapolate from manipulation toward this kind of abstract
    assignment. Manipulation, i.e., the neural realization of motor processes, considered in the
    abstract, is the creation of arbitrary and flexible kinesthetic (etc.) patterns for a variety of purposes
    which must (usually) take into account, actively, the conditions of the body and the outside world.
    That is, the variability of the world encourages flexibility in dealing (manipulatively) with that
    world, in contrast to the more passive sensorium, where the variability of the world, rather than
    actively interacted with, is interpreted through processes which deal with variability by, in many
    cases, filtering it out, in other cases by passively responding to it, and in some cases, by turning
    our "attention" to it so that we may actively interact and create new categories. In addition,
    manipulation is performed by all the animals, from the most primitive; so we have in this an
    origin of the creative ability which is not arbitrary. One can see, observing "lower" animals, that
    manipulation by such creatures consists of fairly set combinations of limited repertoires of
    behaviors. I have mentioned something like this above (e.g., 197-199). But this forces us to take a
    close look at language and logic as realized in the CNS. Since they are realized in digital
    computers, and those devices (I will assume) are not conscious, and if the above is true, and a)
    consciousness is involved in the creation of representation, and b) that creation, in order to be
    flexible, must employ areas of the CNS principally involved with logical and linguistic, i.e.,
    "purely" symbolic, operations, then we must conclude that those areas cannot be anything like
    digital computers' logic circuits (not too radical a conclusion), and in addition, that it is likely,
    given the manipulative origin of these processes (if the conclusion at the beginning of this
    paragraph and the previous ones is true), that they resemble and function more similarly to motor
    rather than to sensory or other areas.
207) Note that if this and the previous paragraph is true it is not these areas alone which enable flexibility (i.e.,
    not the formal recursive or combinatorial abilities of language or logic), but the modifiability of linguistic
    and logical patterns (“rules”?) through (recursive) interaction with the sensorium; and here also is
    where, I believe, the analog/digital difference makes itself felt, in that if particulars are encoded,
    as described above, infinitely finely grained variability (and all other effects of chaotic systems) is
    possible. One might envision, for example, a process, realized in a CA, "sandwiched" between a
    top-down representation-induced process attempting to alter it, and a bottom-up sensation-
    induced process attempting to conserve it, in a dynamic loop inducing a series of alterations in
    that CA which might alter both of the processes feeding into it.

208) In representation, then, we have several possible classes of processes going on. First, if we
    assume that an aspect of representation is the active inducing of sensory CAs, then that inducing
    or evocation must occur from the top down. Second, given that we can create structures and
    meanings within, so to speak, particular representations, the evocation of particular CAs from the
    sensorium must be flexible at least in that different CAs from different sensory modalities can be
    combined fairly arbitrarily; and even different CAs from the same sensory modality. We must
    assume, however, that the more closely joined CAs are into unitary objects, the more difficult it is
    to alter their components. That is, we have described above the unification of CAs into ―emergent
    dynamic entities‖ through tight feedback. This process, clearly, cannot only serve to create a CA
    as such an entity, but to unify several CAs into singular entities. I am assuming here that the
    composites are more easily separated into the original components than into other components:
    other CAs which might have formed a composite.
209) Third, to combine the last several paragraphs, we can assume that some, at least, of the above
    flexible induction takes place guided by and controlled by the high-level abstracted ―symbolic‖
    processes which require, as we seem to have found, a separate neural substrate from the
    sensorium. We seem here to have to have symbols of some sort interacting with some sort of
    entities which are not neural circuits. But how is this possible in neural CAs? There are no
    symbols there; everything is neural impulses. However, since the origin of this abstract processing
    seems likely to be kinesthetic processes, i.e., bodily processes in general, then we might be able to
    employ those to describe, at least initially, the processes underlying and giving rise to symbolic
    manipulation: language and logical thought. Given that this is true, it becomes clearer as to how
    these high-level processes can evoke coherent sets or strings of sensory CAs. Nonetheless the
    problem of the mental remains. Let us first turn, however, to the problem of symbols. One
    question is, why have symbols; what function do they serve in the sensorium/representation
    interaction system? Another question is what can the nature of the symbols be; what can serve as a
    ―symbol‖ in a neural net? Consider a process of successive abstraction on an aspect of
    kinesthetics: motion toward something. The most specific level of this which still retains this
    overall gestalt is seeing or feeling some specific object move toward another object. As this
    becomes abstracted, any kind of figure moves; and we can only know this if it moves relative to
    either a background or to another figure, preferably both. But the figures and ground can have no
    specificity other than relative motion, and in the case of the figure, some sort of object with finite
    size and some direction of motion. That is, the abstraction of motion must have, let us say, these
    elements. We might now apply this concept to some aspect of language or of logic involving
    change over time, let us say – but in order to do this, indeed in order to even speak in these terms,
    we have had to shift from the language of neural circuits to that of ―motion,‖ ―objects,‖ etc. – all of
    which are mental. This simply cannot be correct; it is too smooth and unobtrusive a transition.
210) But there is a very strange phenomenon going on at the highest abstractive level. At lower
    levels, in the sensorium, as we see, for example, an object moving toward another, we can speak of
    CAs firing, of coordination between different CAs responding to motion, the objects, and so forth.
    At higher levels, still in the sensorium, we can speak, if we wish, of small CAs, even of single cells,
    registering that motion is taking place, how, and so forth. But these are evocations of the
    particular motions, and even though they may contain very abstract information about the
    situation, they are still, in a manner of speaking, driven by it. However, if we pass to ―symbolic‖
    processes, i.e., processes which are, first, not driven by externals but internally driven, and second,
    processes which directly interact some sort of CAs which have in some manner properties,
    however abstract, of the objects and the field in which they are moving, then in some manner
    those CAs are interacting as the objects interact: they have become, in some manner, those objects.
    Now what can that mean? If these interactions were in the sensory areas, which they must drive,
    ultimately, then we could say that those areas were being stimulated in the order, so to speak, in
    which they were previously driven, with that stimulation and that order serving to realize the
    correspondence between CAs and the world. Yet aside from asking what might cause such a
    phenomena, we have concluded, above, that the most abstract processes must take place outside
    of the sensorium; and of course there is neurological evidence for this in the existence, for
    example, of distinct language areas in the CNS. Thus we must say that CAs corresponding to
    abstractions of motion (abstract objects moving toward one another on a ground, as above, let us
    say) are being ―internally‖ stimulated in the order in which the sensory CAs corresponding to
    them at lower levels of abstraction would be stimulated.
211) Now even though we could, it seems, explain this in terms of the rather passive CA to CA
    stimulation in terms of which normal distributed processing explanations of the CNS are
    formulated, there seems to be some problems with that kind of explanation. First, if we assume
    that the processes in this abstract area are indeed perfectly general (probably not a good
    assumption), they must be modified by specific information in order to make choices at branch
    points – and even if they are imperfectly general, this should be true. Now this information can be,
    to some extent, in that physical area (i.e., in that part of the CNS)or it can be in the sensorium, but
    it is necessary to employ it, in order to proceed past purely symbolic operations, which cannot be
    specific enough to run meaningful simulations, for example (i.e., induce meaningful
    representations). Thus, the passive running of abstract processes must be actively modified by
    some level of particular information. We could assume that this modification occurs also through
    these passive processes, so that analyses, language, etc., runs governed by a combination of
    general processes and specifics, again passively, but the problem with this is twofold. First, we are
    aware of choices, and choices can be made in a representation. Second, representations are run, I
    have argued, top-down. If choices are necessary, however, the information for these must come
    from particulars; at least, from more particular information than is contained in the abstract
    processes, i.e., from bottom-up. But choices for kinds of simulations are abstract ones, which can
    then be modified by particulars. Somewhere in this process, there must be a guided interaction
    between abstractions and particulars, in which both enable travel through the ―tree‖ (―net,‖ really)
    of branching choices in some flexible manner.
212) What kinds of ―choices‖ am I talking about here? First, the overall choice of a representation;
    this can be explained in a variety of ways. Second, the choice of details within that representation;
    given the possibilities here, there is more difficulty explaining these as the result of passive
    processes, but this seems possible nonetheless. Third, ongoing choices resulting from the running
    of a representation-based simulation. To take a simple example, consider the ―mental‖ rotation of
    a cube. Now in this example, the choices are few, unless we interact that process with some other
    goal. For example, if we rotate the image of a wire cube in order to model stresses in that cube,
    and visualize it breaking apart at the corners, we may modify that visualization to have it merely
    bending at the sides. I do not ask here what guides the details of the representation; memories of
    rather specific objects and events would seem sufficient to create the wire cube, its movement, its
    distortion. Yet the overall choice of that distortion (corners coming apart vs. sides bending), while
    initially, perhaps, guided by some memory, is alterable. This is the kind of critical choice I am
    talking about. Even in the ―pure‖ abstractions of processes it is quite possible, I maintain, to claim
    that such choices are necessary and intrinsic aspects. Now someone like Damasio might claim that
    emotions (i.e., not merely feelings, but emotional biases) play a part in such choices, and why
    shouldn‘t they? But to claim that these alone govern such choices is to ignore not merely the
    evidence of our introspection but that of logic also, in that we can, albeit with great difficulty, plan
    rationally, even in the matter of our biases; i.e., we can, sometimes, override or modify those
    biases, even if only temporarily. That is, I am arguing that passive processes are inadequate to
    account fully for such choices, as we make them in ―real life situations‖; and emotions (unless we
    actively manipulate them, which we indeed do to some extent) are such passive processes (note
    however that we cannot ―actively manipulate‖ them unless we are aware – conscious – of them).
213) ―Actively manipulate‖? Is this merely a subtle (or not so subtle) circularity? It may be, if this is
    assumed to require consciousness, as I do so assume. But rather than face this problem at this
    point, let us look at a possible consequence of assuming this ability. That is, can the type of choices
    hypothesized in the above paragraph, aside from consciousness, lead us further to the mental, i.e.,
    toward qualia? Are qualia necessary to make or to formulate (if that latter is necessary itself) the
    above type (which I have not even clearly demarcated, really) of choices? One of the puzzling
    aspects of the top-down/bottom-up system I have been hypothesizing above is its neurological
    realization. That is, it is easy, given the above, to conceptualize, in general terms, the interaction of
    adjacent layers of neurons on each other, not merely topologically but in terms of information
    flows. That is, if the next higher level feeds down, one can think in terms of an abstraction
    becoming, in some sense, more specific, and vice versa. However, what of neural processes in this
    system feeding down (or up) more than one level? How are the lower (or higher) levels to
    interpret the ―signal‖ or ―information‖ from levels which they do not, except in the most indirect
    way, participate in creating? And this problem becomes more severe for cross-modal information.
    Of course, one could just say, in effect, that this is the way the system is wired, and it just has to
    deal with it, in effect; probably not an unreasonable attitude for simple systems. One might also
    wonder just how much inter-level communication there is in relatively more complex systems, i.e.,
    in reptiles, fish, even small birds. Assuming there is this type of interaction in ―higher‖ animals,
    does it exist in those with smaller cortices? I do not know.
214) Yet the problem with the ―just deal with it‖ position vis-à-vis the CNS is that, while in smaller
    systems this kind of connection might not need any explicit identification: it could, in effect, ―just
    grow,‖ in larger systems, particularly in systems where there are coherent sequences of abstract
    processes, the relationship between those and the processes in the sensorium needs to be fairly
    systematic and clear-cut. That is, if the lower-level processes are truly necessary to interpret, to
    make meaningful, the abstract processes, then some sort of mutual identification of the processes
    is necessary. Now this would certainly be true if some type of rather strict Chomskian or Fodorian
    interpretation of thought were true; if there were truly a language of thought in which formal
    syntax and abstract semantics coherently rode over, in effect, the particulars of the sensorium,
    then some kind of assignment to those particulars would ultimately be necessary. However, if the
    cognitive linguists are correct, and the abstractions were derived, as I have indicated they seem to
    be, from motor processes, then the preexistence of those latter processes, it might be argued,
    would imply some preexisting assignment of particulars, even of cross-modal particulars.
    However, the counter-argument to that is the arbitrariness of the abstractive processes; if these
    are, even though derived from kinesthetics, universally applicable to any modality, and in
    addition capable, a la Fodor, of running linguistic and/or logical sequences independently, pre-
    interpretation, of particulars, then assignment, post-processing, is necessary. And indeed, the
    arguments for the existence of high-level choices I make above would also seem to support this
    position.
215) Let us look more closely at these abstract processes, then. What could they be, neurologically?
    Clearly, we cannot have ―grandfather‖ cells for all ideas, concepts, images, etc., that may be
    employed in abstract reasoning. We must, then, use CAs for this; there just are not enough single
    cells, and the CNS is just not physically stable enough, for individual cells to perform this kind of
    function. And what, then, can abstract ―processes‖ consist of? Surely if single cells realizing
    abstractions cannot serve sequentially as realizations of chains of abstract concepts, CAs realizing
    such abstractions can. We might envision CAs built on CAs, as described very early above, in the
    realization of successive levels of abstraction, and processes containing such realizations as the
    successive evocation of such CAs. Given the timing necessary to form a CA, as described above,
    the timing necessary for the realization of such abstract chains is probably rather critical and
    difficult: an additional reason for the difficulty of abstract thought, perhaps.
216) So we have a CA realizing an abstraction, within, let us say, a verbal area; and similarly, we
    have a small set of cells, perhaps a CA, perhaps not, realizing a particular, within, say, the visual
    cortex. These two (for the sake of an example) must relate in a very intimate fashion, in order that
    the abstraction a) evoke the particular, as an aspect of a representation, and b) the abstraction be
    assigned the particular, in order that it have a pathway, in effect, through the trees of choices as to
    how, ultimately and fully, to create a particular meaning and/or representation. Now one aspect
    of that relationship, as I have said, is the ―knowledge‖ of modality. That is, given the argument
    above about the arbitrariness of neural codes and the necessity of what I have termed
    ―individuate‖ information, it seems that this latter information, rather than being somehow
    carried upwards through all the abstractive levels, could be directly encountered through the
    melding, in effect, of the higher and lower (abstractive) levels: the evocation of the lower by the
    higher. That is, the abstract processes can remain so, if they ultimately encounter the lower levels
    which do retain the individuate information, in order that those abstract processes can be assigned
    (so to speak) particular values or choices of evocations. Specifically, we must, when reasoning,
    speaking, etc., most generally, ultimately know that we are reasoning about sound, sight, or
    whatever, and given the arbitrariness of the coding in these modalities (as described at length
    above), it must be that in order to make this discrimination (and other finer ones), the patterns of
    individuate information in the various sensory modalities must be different from each other, and
    abstract processes must ultimately take these differences into account.
217) To put this another way, we might assume that, contrary to the above account, the
    abstract/particular relationship is fully determined; that there is no choice of assignment of
    abstract concepts, etc., to particular meanings or interpretations; that, despite the term ―abstract‖
    usually being taken to indicate, among other things, a kind of reasoning and/or symbolism into
    which many possible particulars can be substituted (or perhaps more accurately in this context:
    with which many possible particulars can be related); that all this is not really true in a real neural
    network, and when we think of reasoning as abstract and that we have choices, we merely are
    having feelings attached to the same strictly determined processes which govern most of our
    memory and other evocations (e.g., that fire engines have black tires, that in the former phrase
    ―black‖ comes before ―tires,‖ etc.). If we make this contrary assumption, then there is no real issue
    here. The sensory processes merely culminate, in effect, in the abstract processes, even though the
    latter take place in a different location, and we are under the illusion that those latter processes, in
    any particular situation, are arbitrary and may be altered as to both their pathways and the
    choices of what particulars to apply them to. The whole system is strictly passive, from lower
    sensorium up to the highest levels. In this case, however, consciousness, as well as choice, must be
    viewed as a kind of illusion, certainly as unnecessary; this is of course an old and recognized
    position. As well, given this contrary position, the problem of qualia does indeed seem
    unapproachable, certainly by the reasoning followed in this essay. Given a materialist position on
    mind and an evolutionary perspective, however, this particular contrary position seems absurd,
    and I will take it as such; I assume that there is some (functional) reason we have evolved
    consciousness, that consciousness performs some function; similarly with qualia (―feels,‖ if one is
    uncomfortable with ―qualia‖; one‘s position on that terminological issue has no real bearing here).
218) I conclude, also, that consciousness is, on a more abstract level, due to basically the same type of
    process which occurs on a lower level involving focusing and inhibitory peripheral exclusion. If
    we combine the thesis and results of the ―inattentional blindness‖ studies with the recent MRI
    results demonstrating below-conscious processing at low levels of stimulation, then the (old)
    thesis that consciousness is involved with focusing as intensification and exclusion, just as the
    lower processes in the cortex are involved, in the production of abstractions, with the same kinds
    of focusing and exclusion, is supported, I believe, by this combination of data. In addition, this ties
    in with the above ideas about choice. We cannot choose between specific neural sets; this is an
    irrelevant and distracting point, and not consistent with the changeable nature of the connections
    in the CNS. Thus we must choose between environments, whether internal or external.
219) There is a sense in which a situation necessitating choice is created by a goal. That is, in the case
    of a very concrete choice, if I merely pick up a rock with no previous idea of what to do with it, I
    am in a sense faced with either no choices or with a choice of any action I can perform with a rock
    in my hands. If however, I pick up a rock with a goal of opening a can, I have at least the choices
    of smashing the can, of opening a small hole in it, or of attempting now remove part of the can.
    The presence of the can of course is not sufficient to generate these goals and possibilities; there
    must be motivation as well in that specific direction, so to speak. Thus even though "fitting" the
    possibilities to the goal limits them severely, they are still multiple, and perhaps some set of them
    have equal (perceived) validity or efficiency in reaching the goal. The choice is among that latter
    set, but it now seems arbitrary. And we (and other animals) do seek choices like this; if we are in a
    situation in which all actions are rote or algorithmic, we quickly become bored. So the desire to
    choose is there; and clearly we do not choose between neural sets, but between environments, i.e.,
    (the neural realization of) qualities, in effect, as I argued above. In addition, the supposition in this
    essay is that representation is a key element here. That is, when we try to decide what to do with
    the rock and the can, we do so, in part, by mentally rehearsing (which term can here include
    atemporal, i.e. ―visualizing,‖ types of representation) some, at least, of the alternatives. But the fact
    that we create these representations indicates that we do make choices. That is, if the system were
    determined, so that the flow of action merely took place with no need for choice, or that choice
    itself were determined and its feeling an illusion, why the rehearsal of alternatives? If this is in
    order, for some bizarre evolutionary reason, to merely generate the feeling of will or choice, then it
    is a waste of time and energy, commodities in short enough supply. We are, then, actually
    rehearsing various alternatives and choosing between them, and this choice cannot be determined
    in some simple and direct way. In addition, if our choice were made by actually making, say,
    different motions with a rock in the real world, as animals move at random to generate
    alternatives, we might again have some basis for arguing that choices were direct or random.
    However, given representations, we must be able to distinguish between real and virtual or
    representative rock motions; yet something internal has to react in the same way to the real world
    as to the representation.

220) So what is the next step? What would a neural net need to holistically differentiate between
    different sounds, or sound and sight? When we are dealing with a set of single cells on the lower
    levels of processing, the formation of abstractions occurs when a single cell responds to one
    narrow range from a set of ranges of responses of a set of cells. What is needed comparably, then,
    is perhaps something like this for a CA dynamic. Some single cell or small CA must respond to
    some narrow range of dynamic patterns as realized on another CA. One would not need multiple
    CAs, as one needs multiple cells, since the variability of the dynamics could substitute for that.
    Above, I argued that CAs would have to register, in some way, the different patterns in such
    neural events in order to even approach qualities, and I outlined various ways in which this might
    happen, involving, among other things, phase relationships between input to a CA and the
    internal dynamics of that CA (see 40-41, 83-85, 94; 149-152, 155-157, 162-163; 184-185; 194).
    However, the issue is complicated by the relationship between consciousness, self-consciousness,
    and qualia. It seems virtually a tautology that when an entity is conscious, it experiences, i.e., has
    qualia, and that having qualia implies consciousness, at least potentially (I avoid, I hope, the issue
    of whether there are unconscious mental processes by claiming that without the potential for at
    least some of them to be conscious, there can be none.). That is, both imply each other. However,
    self-consciousness is another issue, and it is not clear that what has been termed "reflexive"
    consciousness by Sartre or "second-degree awareness" is necessary for qualia. Thus, I would assert
    that animals can have qualia yet not, for example, pass the mirror test. That in some sense
    simplifies the problem here, in that all I want to clarify, if possible, is qualia. We have seen, I think,
    that there are neural bases for the kind of gestalt entity that qualia necessitate, yet also that this
    does not yet imply experience.
221) Now if the above is true, then the various arguments of those claiming that self-consciousness,
    i.e., some sort of consciousness, not so much of the self, but of the self as self, i.e., as the "center,"
    so to speak, of one's experiences (in contrast, for example, of being conscious of one's body merely
    as experiences like any others), is necessary for qualia, are false or misguided. That's a lot to claim,
    but I think that a) in order to have some reasonable cutoff point for consciousness, and b) that
    such a cutoff not be arbitrarily set at human beings (and indeed the great apes do incontrovertibly
    pass the mirror test, yet do not incontrovertibly have language - certainly not incontrovertibly in a
    human sense), such a claim must be made. Now as an aside here, it seems (and I have only seen
    anecdotal evidence for this) that the great apes do not pass the mirror test until they have matured
    to some extent; but surely they do not attain consciousness until such a time? Young apes, then
    (and perhaps young children), are, we assume (since they are later, if passing the mirror test
    demonstrates this) conscious, yet do not pass the mirror test. Thus that test is not a sure one for
    consciousness, a support for my general thesis.
222) In addition, one can muster various arguments, I believe, to the effect that in dealing with the
    similarity of representations to, say, the sensorium, and that in manipulating (and in choosing
    which to manipulate) both representations and the world, one must employ the same types of, yet
    different, processes, to the same types of, yet different, neural realizations of sensory phenomena,
    that what is necessary is something like qualia, in order to simplify dealing with the similar
    characteristics realized by the similar but not identical neural sets (CAs) and neural dynamics of
    representations and sensorium. That is, it seems that qualia would make good shortcuts in dealing
    with these complexities. This also, if true, would imply qualia only with representation. But I will
    not at this point bother to present the arguments that may occur to me in this regard, for the
    reason that without some sort of support in the form of at least a general idea of the neural
    realization of the creation of these shortcuts, those arguments will at best be clever
    rationalizations. But I think, actually, that the best argument for the existence of qualia is the
    functional/representational one. That is, we manipulate and in general deal the world in certain
    ways, let us say, which are for the most part (with the perhaps arguable exception of language)
    very body-oriented. Now this would be fine if there were no representations: we could just
    describe the system in terms of manipulating CAs that "stood for" the world. But if we are to, in
    addition, manipulate representations of similar "things" in the world as what we literally
    manipulate, then the best way, it seems to me, would be to devise some way of manipulating
    them similarly, while maintaining some sort of distinction. But insofar as both representations and
    the sensorium are CAs, but cannot be exactly the same CAs (if for no other reason than that
    representations are, as I have argued, top-down induced; secondly, that they must be
    phenomenologically – and functionally – distinguished from the sensorium), we must "see" them
    somehow as very similar, even identical, in order to perform, in effect, the same manipulations on
    them. After all, one might say that the Darwinian point of having representations is to test reality
    without getting killed by it. So a "manipulation," which is for example a movement in the world,
    needs to be handled virtually the same by the system which employs it both externally and
    internally to move a "world" which also needs to be sensed virtually the same when realized both
    externally and internally.
223) Also, I have barely touched on emotions in this essay, surely one of the most basic aspects of
    qualia. However, their utility seems unarguable; if one merely observes oneself being cut
    (assuming this happens while one is looking), for example, without pain, the evaluation of
    damage and importance becomes difficult. Qualia thus seem functional here in the same way that
    they do for the sensory modalities. Given that emotions have particular areas for processing other
    than the cortex, one might ask just where the consciousness of emotions takes place; but without
    some neural means of modeling this, as indicated above, the question is moot. Just as there are
    special channels for information from the different sensory modalities, there are different channels
    for the emotions; are there different patterns, as well? The fact that one can feel a painful stimulus
    as simultaneously pleasurable, for example, would seem to support this. Given the bodily
    responses to and of the various emotions, which do seem to differ fairly clearly, one can easily
    hypothesize that these are aspects, at least, of the bases for emotional generation and
    representation. We must remember, as above, that there is no painful information, just as there is
    no red information.

224) The question can be posed this way: let us suppose that we can look at various neural circuits,
    their dynamics, etc., as they happen. Now what circuit, what dynamic, would be convincing or
    even indicative as a possible realization of qualia? Certainly not a simple feedback circuit.
    Suppose we had two (or multiple) such circuits reacting to each other; would that be convincing?
    Of course not; what compels us to see the possibility of qualia being realized here? Suppose we
    had two such circuits which were each realizations of dynamic entities, as described above; would
    those, together, be convincing? Perhaps at first, but we could find, I am sure, examples of this kind
    of dynamic in nature which we would not consider realizing qualia; for example, suppose we had
    two (or multiple) interacting magnetic fields, generated by ionic currents on the surface of the sun,
    such that they had internal feedback maintaining them, and were in addition interacting with each
    other (or multiple others), or suppose we had something like Ashby's machine. Surely the
    feedback in a system like this could be such that each was a unitary entity (as described above),
    reacting to others as a unitary entity. Could this be considered as generating qualia? I do not think
    so; it is still too mechanical, too arbitrary yet too determined. What could feel or experience here?
    What if this system was recursive, built up from similar structures? Again, no qualia; why should
    there be? What is there that is convincing about qualia here? Nothing that I can see. Yet we must
    have something like this; we are a system of neurons interacting; there is no red information.
    What is missing here?
225) Let us consider a system which takes into account what I have termed "individuate"
    information, or something like it. That is, the "incoming" dynamic in a neural system is not merely
    feedback loops, but has detailed neural coding, patterning of some sort, relating, let us say, to
    some sensory modality. Now suppose that this dynamic was reacted to by another dynamic,
    which had the potential to react to multiple possible incoming patterned entities such as this
    particular one, and as part of that reaction, patterning itself, at least in part, after the circuit it
    reacts to. The patterns match, to some extent. Let us further say that as a result of this match, and
    of feedback from each to each, the two now similarly patterned circuits "lock" together such that a)
    the matching is maintained, and b) they now form one dynamic entity. Suppose that consequent
    to this, this extended entity locks itself, similarly, to whatever feeds into the previous "incoming"
    circuit, now forming, if not one entity, at least one similarly patterned, interacting, set of circuits.
    This can continue as far down the chain of abstractions as the system will allow. Alternatively, this
    could have proceeded upwards through these systems, successively "locking" higher levels of
    circuits together. Now here, I think, we have a possibility of considering that the totality of this
    dynamic realizes qualia.
226) This kind of system satisfies many requirements for qualia generation in the kind of neural
    system I have hypothesized. For example, the conservatism of the system, realized in the details of
    feedback from higher levels, as described above, is supported here by the "locking" of higher to
    lower patterning. This system will maintain itself, but, if the dynamic is right (i.e., the stability), is
    able to switch to other inputs, locking itself to those. One must hypothesize that the uppermost
    circuit or dynamic entity is capable of such plasticity, and as well is not completely "taken over"
    by one particular pattern. That is, this "uppermost" dynamic entity could consist of multiple interacting
    entities, not all of which were locked to one input (although too much variability here would result in, for
    example, the possibility of multiple responses in an environment, which is not usually possible or desirable),
    but all of which were locked, to some extent, to one another. This would allow choice, given a proper
    balance between stability and instability, and also multiple sensations, but would, given that the
    circuit was mostly locked to one input, create a qualitative focus. I will thus term this area the
    QFA (qualitative focal area).
227) Realizing that what is happening here is a kind of enforced stability, that the "locking" must be
    actively maintained, both because of "competition" from other inputs and because of internal
    variability, we must conceive of this not as a kind of passive reactivity but as an actively maintained
    stability. A more-or-less "central" QFA or rather small (relative, say, to the number of areas on the
    cortex) set of such areas (which indeed may itself be cortical) is necessary for several reasons. First,
    our human experience is of one "center" of consciousness and of a rather small set of qualia
    relative to what is actually being processed. Second, multiple "centers" would be, it seems to me,
    very non-functional in the threatening environment in which (I assume) we evolved, since
    competition between potential actions, say, caused by those centers would have to be resolved
    extremely quickly. Now I am not claiming that this system, so generally described, has
    consciousness or even qualia. But I am claiming that this type of system seems to be able to
    provide a basis for qualia, in that what is required, I think, is an internal reactivity to the
    particulars of incoming information, a reactivity which allows for this kind of flexible stability.
    That sensitivity to particulars must somehow interact with the neural bases of those particulars, in
    the cortical areas in which they are formed, yet cannot be itself based in those areas, since
    consciousness (or qualia) is retained if (some at least) of those areas are destroyed. In addition, the
    focusing or restriction of consciousness, required by the necessity for limiting responses to the
    sensorium, is fulfilled here. The downward (or upward) locking-in of neural dynamics seems to
    fulfill all of those requirements.
228) What about representation versus the sensorium? One of the issues was the "filling in" of
    specific information into the general abstract processes which were hypothesized to occur in the
    area of highest abstraction. However, given the locking-in phenomenon described above, and
    interaction of this phenomenon with that abstract area, it would seem that feedback upward to
    those processes could solve this problem. In addition, the abstract processes would play a part in
    the selection process proceeding in the QFA. What about my earlier claim that consciousness had
    to do with the interaction between representative processes and sensory processes, in this context?
    For one thing, the locking-in process, as described (roughly, and I will attempt to describe it in
    more detail below) above, does seem to be similar to the representative process, as I have
    explicated it, yet the latter leads to the internal manipulation of the sensorium. What of the
    former? But one of the problems I have faced above is the desirability of manipulating the
    sensorium as that, i.e., as mental objects corresponding as closely as possible to physical objects,
    rather than in terms of abstract operations on CAs or neural codes. The solution, as I have
    mentioned, is for the system to "have" qualia; and now we see that the similarity between qualia
    realization (as I have hypothesized it to occur) and the formation and manipulation of
    representation may be, indeed, necessary. That is, the top-down induction of sensory (let us say)
    CAs through the abstracted manipulative processes hypothesized as the generators of
    representation can now proceed by running through, so to speak, the QFA, so that not only are
    representations generated, but qualia are created through the locking-in process.
229) Why might this latter process be convincing? For one thing, if we envision the QFA as reacting
    to a great deal of input and selecting (influenced by, among other things, the abstract processing
    area) a small subset of that with which to lock-in, then we can hypothesize two further
    phenomena. First, other inputs to this area, not selected, will nonetheless be influenced by the
    locking-in process, and "colored" to some extent by the dynamic pattern assumed by the major
    part of the QFA. Second, these other inputs can serve as at least part of the background to the
    locked-in foreground. Contextual influence and gestalt effects are phenomena of consciousness
    that have to be explained, and the QFA, in dynamic stability with other potential qualitative
    focuses, can either employ those as background or context, or employ the CAs resulting from the
    locking-in processes for those purposes. For another thing, we must, as I have repeatedly
    maintained, restrict ourselves to dealing with the CNS and its neural coding; and the formation of
    "objects" such as qualia must be explained in that context; there are no little CAs shaped, colored,
    smelling, feeling and acting like trees and dogs in our heads. However, I might speculate that the
    "self-awareness" which we humans (and certainly some apes, and perhaps all mammals which can
    represent) experience arises from interactions within a large QFA, in which multiple "sub-areas"
    simulate each other to some extent so that a "locking-in" process runs between these areas, and a
    dynamic entity is formed superposed on the individual dynamic entities of each area. Given the
    inevitable competition between these areas, this would seem to be a much more unstable situation
    than with one qualitative focus. Note also that, given the ―attentional blindness‖ results, the
    locking-in process would seem to be necessary for consciousness. That is, without something like
    this process, we are not, it seems, able to be conscious of particular qualities.
230) The next problem, then, is to explain just how the CNS, or CAs, treat this dynamic patterning as
    functional unitary objects. That is, I have said that I believe it simpler, easier and faster for a
    system to react by deciding to move an arm to pick up an object, then sensing the movement of an
    arm, as an arm, picking up an object, etc., rather than by computing trajectories and working
    through grasping functions, etc. That is, qualia simplify at least the processes of decision making
    (tying them into the representing process) in interactions with the world. Given that the above is a
    start, at least, in the description of qualia, where are the objects here; how does the above simplify
    anything? To put it another way, just how can the neural systems as described above function as
    singular "objects"? The answer, I think, is something like this: the "locking-in" phenomena
    described above (and elaborated on below) is not a passive phenomena, but an active one. The
    series of CAs locked together function as an object in that when some CA anywhere in that series
    is stimulated, inhibited, i.e., acted on (neurally) in any way, that change propagates actively
    through that extended neural superset. This propagation does not have to happen with such
    timing as to create an "emergent dynamic entity" of the type described earlier, and in addition,
    given the possible complexity of the feedbacks maintaining this object, there might be a variety of
    effects of such an activation within the set of CAs combined into this object. However, given that a
    stimulation of some sort on some part of that set of CAs does propagate preferentially within that
    set, and given that the locking-in processes, actively maintained, will maintain the nature of that
    stimulation as well as the object, it is possible, then, that a "manipulation" of the qualia realized by
    that set of CAs manipulates many subsets of that set similarly and as quickly as neural
    propagation will allow. Thus interference is effectively eliminated by the locking-in processes, and
    propagation is speeded up. "Moving an arm" then can be realized at the top of the hierarchy of
    abstraction in the set of locked-in CAs, and propagated, without the necessity of any stimulation
    other than that high-level abstraction, down to very specific processing levels. There is nothing
    unique about this; it is certainly possible that such sets of CAs could be stimulated in the same
    manner without such a locking-in process, yet aiding this type of activation would seem very
    advantageous. An object, neurally, then becomes something corresponding to a "real" object, in
    that it consists (loosely speaking) of an integrated and interacting set of (neural) subsets, where
    the subsets act in accordance to their function in establishing the neural correlates of the object, as
    the subsets of an object act similarly in accordance with their functions as components of the
    object.
231) Now let us return to the speculation of 226. I said: I might speculate that the "self-awareness"
    which we humans (and certainly some apes, and perhaps all mammals which can represent)
    experience arises from interactions within a large QFA, in which multiple "sub-areas" simulate
    each other to some extent so that a "locking-in" process runs between these areas, and a dynamic
    entity is formed superposed on the individual dynamic entities of each area. Given the inevitable
    competition between these areas, this would seem to be a much more unstable situation than with
    one qualitative focus. Now I am not going to speculate here on where "self-consciousness" begins
    and ends; I mentioned above some problems with the "mirror test," but as I have repeatedly said, I
    think that the internalization of the sensorium (which is found, I claim, to some extent in animals)
    is tied in with consciousness. What level of consciousness, however, I will not speculate; but
    without this process of representation, I do not think consciousness is possible at all, and given the
    hypotheses above, this speculation seems to be becoming reasonable. However, when we proceed
    to further speculate along the lines in this paragraph, we do indeed step into uncertain realms,
    certainly as far as what "level" of consciousness we are speculating about. Now in the last
    paragraph, I argued (and I will argue this in more detail later) that there is a real possibility that
    functional objects corresponding to qualia exist neurally, but we still have not really stepped over
    the line into the mental in any definite manner.
232) Consider one of the basic ideas motivating my model: that the CNS is basically conservative, a
    system trying to find and maintain stability in a changing environment. Thus, the locking-in
    above, the formation of, let us say, proto-qualia, is for the purpose of actively creating and
    conserving hierarchies of related CAs, and boundaries between those and other CAs. That is, the
    "object" which is the sum (metaphorically speaking) of the CAs locked-in is prone to drift apart,
    and the top-down processes tend, as I have said, to conserve its components. But further, the
    locking-in process can be seen as a sort of ultimate conserving process, actively as well as
    passively maintaining the "object." Given this, the process of representation can be seen as an
    aspect of this conservatism, while the sensorium is more prone to dissolution (or simply dying
    out). In this picture, the CNS is creating and clinging to its constructions and abstractions in order
    to make sense of the world. The function of qualia, then, would seem to have to do with the
    balance of these processes. Suppose that there were a way to easily and briefly note variances in
    structure, deviations from the conserved norm. Qualia would nicely serve this function. Instead of
    having to monitor the whole neural structure from top to bottom (or the reverse), if the system
    were instead "aware" of that structure as a hierarchical set of simple "symbols," "feels," etc., then
    monitoring differences between that set of CAs as it drifted away from its original "optimal" set
    and configuration, and whatever it drifts toward and/or from, and correcting (or not) that drift
    and the boundaries between that set and others, would be, it seems to me, greatly simplified.
233) When the lower CAs send back to the top of a locked-in set, that top can react to the detailed
    information within each modality and the particulars of the stimulation in that modality. That is,
    the apical CA may in effect be a large cell-body surface (because the interconnections do no
    processing, only transmit between cells) which consists of many cell body surfaces interconnected
    rapidly. So this large surface might be capable of summarizing patterns from all stages (whatever
    reaches it) of the CAs that go into the qualia it is reacting to. This summarization, on that extended
    cell-surface, creates spatial and temporal syntheses of the received patterns; the cell networks
    constructing the qualia in that particular set are effectively unified on that extended cell-body. I
    speculate that this is the culmination, effect, of the realization of that qualia. That is, if that apical
    cell-surface is indeed fed "signals" from all parts of the large extended set of CAs making up that
    qualia (or generating the information going into it), then the patterns on that surface duplicate, in
    part, the "individuate" information in parts of the generating CAs. It may be that the apical cell-
    surface receives only from the top of the system of CAs generating the qualia; in that case, some of
    the top set must retain and/or be fed directly some of that individuate information, since we
    experience qualia, in part, in terms of that information, as I have argued above, and that part of
    that information we do experience must be sent to the top. Notice that the locking-in phenomenon
    here will not only feed back to stabilize the whole set of neural parameters; in doing this it will
    stabilize the sensorium also; in this sense the organism selects its world.
234) It would also seem, then, that a function of language and similar abstract symbolisms, and
    indeed representation in general, is conservative. That is, the use of symbols, i.e., high-level
    abstractions and abstract operations, would tend, as an aspect of that use, to maintain the sets of
    CAs already present. The employment of a symbol or abstract sequence would, by the top-down
    evocation of the corresponding set(s) of CAs, maintain and strengthen that set. The sensorium,
    then, must, in light of this conservatism, have a strong tendency to be classified in terms of
    already-existing qualia, unless reasons for changing are very cogent.

235) A pattern, whatever its type, has to ultimately be integrated into a single entity, otherwise it is
    meaningless to conceive of it as a pattern. Whether that happens over an extended cell-body, on
    one cell body, or in an extended dynamic entity, it is necessary that any pattern ultimately be
    unified, i.e. united into a single interacting entity. But it is still "unrecognized": merely an electrical
    pattern. But how can we go from something which is an electrical pattern to something which is
    not? That is, it is not that there is an ontological problem here with attempting to delineate what
    the "mental" is, so much as an epistemological problem: how can something which is physical be
    experienced in such a way as to seem not physical?
236) To put it another way, given the strong possibility that qualia, at least, are present to (i.e.,
    realized by) brains as small as those of birds and small mammals, we must conclude that the
    transformation desired from the last paragraph is a relatively simple one. We do not, it would
    seem, need the huge (?) human (and ape) cortex, prefrontal and parietal lobes, etc., to have qualia,
    though we might for self-consciousness. So there is, at this level at least, something simple and
    wholistic that can go on, in brains smaller than peanuts.
237) We might hypothesize, then, that the patterns which are reacted to as qualia are the dynamic
    temporal/spatial "emergent dynamic entities" (EDEs) described above (85, 106, 149, 156, 162, 187,
    190, 192, etc.). In addition, these EDEs can be treated in at least two different ways. First, they can
    be understood as instantiated in neural discharges within the kind of CA described above, with
    particularly suitable relaxation times. Second, it may be that it is not the discharges per se, but the
    envelopes and/or various other aspects of the discharges may be reacted to preferentially to the
    discharges themselves. That is, not merely temporal envelopes of neural discharges, but spatial
    envelopes of sets of discharges impinging on a single CA. An envelope, in effect, is a function, i.e.,
    the realization of a mathematical function on an impulse sequence, as a pattern over a CA. An
    envelope is also an abstraction. That is, in reacting not to the actual impulses but to their temporal
    and/or spatial envelopes, the system reacts to an abstraction of a set of physical events. This is in a
    way a trivial extension to the EDE concept, but I offer it to emphasize the possible variability of
    the nature of the entity.

238) The basic importance of the EDE notion is in the instantiation of pattern. This needs to be
    carefully considered. That is, if we claim that sets of neural impulses in some part of the CNS form
    a pattern, while this may be a true statement for an external observer, for the system itself pattern
    is only present if such sets are reacted to as wholes, i.e., as singular entities. A pattern is a singular
    entity. If it were not, then however "patterned" some set of neural impulses would be to an
    external observer - ranging from an observer with a computer capable of analyzing that set in
    various ways to one actually seeing the pattern (via MRI, etc.) - to the systems instantiating those
    impulses, without such a unifying analysis, no pattern would exist. Now how, in the CNS, may a
    unifying analysis be performed? Only in two ways. First, through memory, and second, through
    the simultaneous processing of the components of the analysis. Yet memory is the equivalent of
    the latter type of analysis, since the memory presents a previous (potential) component of a
    pattern as one contemporaneous with whatever it is being integrated with (in the unifying
    analysis). I briefly considered this same point above, but it is also important here. An EDE, then,
    aside from (and of course it must include) the microscopic processes on the surface of single cell
    bodies, is quite literally the only means by which functionally singular patterns may be
    instantiated and/or generated.
239) Let us consider some further implications of 238. It must be that qualities or qualia are largely
    independent of neural variation. That is, the CNS is constantly physically changing, yet red
    remains, as far as we know, more-or-less the same. Note also the recent results in which signals
    meant for the visual cortex were sent to the auditory cortex, yet the animal sees. We might, it is
    true, be dealing with a system which somehow labels the origins of neural impulses so that it
    "knows" to process information from the retina as visual; but it seems, especially given this recent
    data, much more likely that we are instead dealing with a system which treats, through the
    successive abstractive processes described above, all sensory modalities (in general) similarly, and
    so can adapt to change. However, visual qualities of some sort must have resulted from the above
    processing. Clearly structural patterns (i.e., the structures of dynamic neural patterns) were
    abstracted for the visual data. As we have seen, then, it is the patterns of what might be termed
    "abstractive computations" which at least generate qualities. Yet if we combine this idea with that
    of the EDE, we see that these patterns are capable of being literally integrated and instantiated in
    these systems. Given this, the pattern that is induced by redness is capable of not only proceeding
    through and as a result of the abstractive processes, but of being realized and integrated as an
    EDE (but see also 272).
240) When we ask how this pattern, for example, can be robust, i.e., how red can remain red despite
    variations in discharges and indeed in the neurons presenting them, we realize that the term
    "pattern" has not, and perhaps cannot at this point, be clearly explicated. However, we might
    speculate that temporal/spatial relationships and amplitudes over an extended cell-membrane
    would have ranges within which they would be responded to similarly; i.e., that the initial
    patterns on an EDE could be abstracted, analogously to the neural discharges generating them at
    the lower levels. But that of course opens the homunculus question again: "responded to" how
    and by what? I will temporarily postpone this question.
241) Let us consider some of the advantages of the above explanation as a tentative step toward the
    realization of mind. First, it is the case that qualia are based on neural discharges but are not
    neural discharges; the patterns above are patterns "of" neural discharges, but, as patterns, are not,
    especially if abstracted, actually any particular set of such discharges. Second, qualia are
    "summaries" in some sense, of the neural dynamics, but must contain the "individuate"
    information, as I have termed it. These patterns can fulfill that: one level being individuate, the
    ―envelope‖ being summary. Third, qualia are relational; no problem there, if the above is correct.
    Fourth, qualia are fairly independent of neural variation, i.e., neural connections may change
    fairly radically without changing our mental contents. Envelopes (and other functions) would be
    ideal for this. Fifth, as summaries, qualia may be processed by rather deep and simple (relatively
    speaking!) neural circuitry, e.g., the RAS and associated structures. It seems that this might be the
    case for functions of neural dynamics, if those could be reacted to or transferred to the RAS, i.e., a
    function, as a pattern fairly independent of the details of neural firings, might be adaptable to the
    RAS (or the parietal!) circuitry. In addition, the EDEs would seem able to create and establish
    identity in general without necessarily matching specific, individuate information, because of the
    robustness of the patterns they instantiate. Thus, they would be ideal for initiating the top-down
    processes necessary for representation.

242) So the term "dynamic pattern" has been made a little more specific. Now we must consider in
    more detail how such patterns function. Let us consider the problem of destabilization (aside from
    the obvious: that it runs counter to what I believe to be a major organizing factor in the CNS) in
    terms of its realization. That is, suppose we have a top-down feedback process from some CA
    realizing some abstract function in the system. Let us take as an example one I used above: the
    rotation of a visualized cube. I said that in order to do this we must induce in the lower CAs the
    dynamics corresponding to the various rotational positions, and this seems reasonable in view of
    the findings that time delays for the various visualized rotations increase as the angle increases:
    we are actually visualizing and rotating. Yet again we must be cautious here and reiterate that as
    far as this explanation is concerned, we are not "minded" yet, merely describing neural circuits;
    and those corresponding to those mental rotations must be the CA subsets which fire, as aspects
    of the internal processing of the sensorium, in response to the rotations of objects as inputs. So in
    order, however this (rendering of CAs into experience) is done, to experience those rotations as
    visualizations, we must fire the CAs which correspond to the sensory rotations, in sequence. Now
    let us go yet another step and say that this specific internally directed rotation is not something
    that has happened before; we are testing some aspect of reality, say, through representation. We
    are then destabilizing the system with some internally induced (representational) novel input, in
    effect, not the normal top-down stabilizing input.
243) But in destabilizing the system, we must yet have stability. The process of representation is just
    induction by a function. That is, the rotation of the object, or whatever the manipulation is, must
    remain that rotation of that object; the object and the operation must be conserved. This brings us
    back, in a way, to the situation above, in which a pattern must be "seen" as the same pattern
    despite within-pattern variability. A function must be treated as such, but it is not enough now to
    generate a function; that function must be able to be manipulated, i.e., to have different "variables"
    (neural firing frequencies? neurons that are firing? temporal firing patterns?) entered into it, and
    indeed perhaps itself to be partially changed, yet be reacted to as the same function (we might
    speculate that stabilizing and destabilizing feedbacks might be interacting in such a way as to
    allow the system to operate within the "cracks" of the stabilizing dynamic; the variability before
    and after the stabilizing feedback might be the important aspect here). Meanwhile, to put this
    another way, we have seen (tentatively, perhaps) that when an EDE operates as a function the
    system must react to that function for it to be meaningful. But a function is not the individual
    discharges or even trains of such which generate it, but an abstraction of sorts. (What would be
    the next step toward further abstraction, further detachment from the actual (and necessary)
    neural discharges underlying a function? It seems to me that the next step would be an arbitrary
    (variable) assignment or interpretation of the same particular function. If there is more than one
    way to interpret a given set of discharges, and indeed if the system can take advantage of more
    than one possible interpretation of a given singular EDE, then not only do we have a function, but
    this function is not solely (but it must be partially) driven by the physical parameters of the
    system.)
244) Let us return to consideration of the EDE, its stabilizing function, its realization as a neural
    function, i.e., as an abstraction of a set of neural impulses, an envelope, perhaps, ―floating over‖
    the spike trains. Now suppose that the function of some EDE is to monitor the abstraction process
    of some set of CAs which are involved with responding to different angles of orientation of a set
    of figures; this EDE keeps the neural patterns fairly constant so that the same sets respond to the
    same orientations of the figures, and so that, when a figure is seen, the same sets, roughly, fire;
    and conversely, that when a particular figure is represented, the same sets fire as when it is seen.
    Now, suppose that, in a representation, one wishes to rotate the figure. One way might be to
    directly, somehow, induce firing in the sets, consecutively, of neurons responding to the various
    orientations of the figure. I have touched on this above. Another way would be, instead of dealing
    with the neural trains or neurons directly, to deal with the control function which stabilizes them,
    in order to destabilize it in such a way as to induce the necessary rotation, i.e., to induce firing in
    the successive CAs corresponding to the rotation. Here instead of sensing or monitoring specific
    neurons, the system needs to monitor (and alter) the control parameters; the neural dynamics or neural
    functions involved with stability in the system, to alter those parameters or functions systematically in
    order to systematically destabilize the image induced top-down. In addition, the highest level of
    abstraction, that which is inducing the destabilization, must in some way mimic the function it is
    destabilizing. That is, we rotate, we assume, with the purpose of rotating; the former consists of
    the systematic destabilizing described above; the latter, of an abstraction of that function. If we can
    find a way to neurally realize this, we have, I believe, while not necessarily instantiating consciousness,
    instantiated qualities.
245) Let us suppose that this variability occurs first with two separate reacting systems. That is, one
    part of the cortex, say, has processes impinging on the CA which realizes the particular EDE in
    question, senses the presence (and some values, say) of that EDE, and reacts in some way,
    sending, let us say, feedback back to the original cortical area. The same, let us suppose, is done,
    with different results, by another cortical area, either simultaneously or not, reacting to the same
    EDE in a different way. Now this does not seem so much of a jump from a "normal" (whatever
    that is in this chain of speculation) type of reactive response to an EDE. Yet even this does,
    perhaps, establish a certain variability or distance of that EDE from the normal run of neural
    interpretation, in which impulses run rather straightforwardly through circuits. Here it is
    necessary to have both macro- and micro- feedback, so to speak, where the macro- involves the
    distant area and the micro- the local impinging of that area onto the CA in question. That is, to
    "react" to an EDE requires that the neural set impinging on that EDE must, in effect, "bounce"
    impulses off of the EDE, which means bouncing them off of its neurons, of course, as they are in
    the act of maintaining that EDE. "Bouncing" here means, then, something like sending impulses to
    some neuron in the act of firing, where the neurons are connected in such a way that the
    impinging impulses are modified somewhat - as a result of the firing of the "sensed" neuron -
    when sent back ("micro-feedback") to the originating neuron (or CA containing that neuron),
    while simultaneously maintaining the EDE which is being sensed. The modification, given that
    the CA is an entity, is dependent on more neurons (in the CA being sensed) than the one being
    impinged upon. That is, when neuron A's impulses impinge on neuron B, neuron B sends
    impulses back to A which vary depending, among other factors, on whether B is currently firing
    under stimulation from neurons in the CA of which B is a member. Now that "depending" thus
    could incorporate some sort of function of the firing of B, which might include effects of other
    neurons in the CA of which B is a part, given that they are also feeding into B.
246) How exactly, then, does an EDE on a CA modulate, i.e., enter into the computation of, its
    response? To recapitulate, we have seen that there is a necessity for top-down stabilization, in the
    case of non-CA sets of neurons describing components of the next level(s)' abstract processes.
    Now that stabilization, necessary to maintain construction and discrimination in a physically and
    "informationally" changing environment, should be echoed, so to speak, on the level of CAs.
    However, CAs are hypothesized to function at higher levels of abstraction, so that a major
    function of CAs is to create multi-modal integration, and given that and other factors (size,
    distance, etc.), this does not seem a likely situation, at least in the detail necessary to maintain
    functions. We may then (hypothetically) conceive of the abstractive processing "throughput" in a
    CA as basically the same as that in the cell sets processing low-level abstractions: a more-or-less
    straight-through system (of course not without feedback, but without CA-type feedback in the
    sense described in this essay) of functional refinements (e.g., retinotopic-like refinements).
    However, given the necessity for feedback stabilization over a large set of simultaneously
    "computing" cells integrating many inputs, the EDE, as a single entity, could be an ideal check on
    both stability and on pattern integration for such a system. Note how this fits in with the "locking-
    in" processes described above (226-230). It is also possible, however, that particular cells, members
    of the CA and contributing to its processing of external (to the CA) inputs must perform a dual
    function; internal monitoring (generating the stabilizing EDE) and "external" data processing
    (using different receptor types on the same cell body?). Now it may be that this is unnecessary;
    that there are cells in CAs specialized for internal stabilizing; but this seems uneconomical, at
    least. In addition, performing such stabilizing needs to be tempered with adaptability; thus such
    specialized cell sets would seem, at higher levels of abstraction, to possibly create too rigid a
    system if they are not in the processing stream also. On the other hand, stabilizing through such
    isolated (i.e., cells which do not [primarily] process external [to the CA] input) cell sets would
    serve well, perhaps, in systems that need great stability, like syntactic processors and motor
    systems.
247) How do the EDEs ―float‖ over the CAs? If we conceive of the neural flows in the CAs as
    modulated, i.e., as their being cell-sets sensitive to such dimensions as the envelopes and various
    functions of the code flow, as I mentioned above, then those cell-sets, subsets of CAs, are, as I have
    said, sensitive to abstractions of the neural code: abstract functions. Now if they, in addition,
    operate on those functions, making adjustments to the neural flow in a given CA in order to alter
    them or merely maintain them, and in addition if those neural flows are creating EDEs (as
    described above), then those cell-sets are effectively creating and maintaining the EDEs as abstract
    characteristics of the neural flow, which flow, on a lower level (in the same CA) might be
    performing very different (abstracting) functions. The EDE in this case, then, is in effect a
    modulation of the neural flow. However, one must be careful here to realize that however abstract
    this EDE seems, it must be ultimately generated by neurons. An abstract function on a neural code
    cannot generate itself; although it might modify the effect of that code, that modification must be
    realized as modifications of some aspect(s) of the dynamics of the synapse or cell membrane. Thus
    one must be cautious of the implications of EDEs interacting with each other directly, for example.
248) Looking at cell types, the usual direction of flow for neural impulses in a CA, especially a
    sensory CA, is towards higher abstractions: the cells found in the cortex are very polarized as to
    the direction their dendrites conduct. But the problem here is the top-down flow inducing
    representation. This must use cells moving in the reverse (or both) direction. But the reverse
    direction does not contribute to the abstractive process; what then does it do, at the level of the
    CA? It must function to 1) create the EDE, and 2) create the above downward flow. But that latter
    flow must, if the abstractive processes are universal, itself serve some processing function. We
    have seen above that it serves to stabilize the system, i.e., to preserve the results of the abstractive
    processes. Now if this is the case, then representations are generated, in effect, entirely by
    processes aiming at stability. Yet this cannot be correct, if such processes serve not merely routine
    anticipation, but the production and anticipation of the unexpected. In addition, given the results
    of the ―attentional blindness‖ and many related studies, consciousness seems based, in part at
    least, on change. But the above are not, at least not essentially.
249) Might restricting the function of the EDE as I am, then, restrict its scope? I have, after all, been
    grooming it, in effect, as the basis for qualities. Yet again, as I have said above, it is just this
    conservatism that is required for qualities; green remains green in numerous circumstances, over
    variation internal and external to the CNS. The registration and conservation of pattern is now,
    given that the above is true, an essential aspect of the CA through the EDE, as a conservative
    (stabilizing) function, and it is just this creation, really, of pattern that must (as I have extensively
    argued) be the basis for qualities. But part of the restriction of scope here, it seems, is the
    restriction of physical scope. That is, an EDE employed, let us say, to "sense," create, and maintain
    a motor pattern is, by that function, it would seem, necessarily restricted in its influence to the
    motor CA it is monitoring; yet if, as I have argued above, such patterns are employed in linguistic
    functioning and in logical functioning, not to mention the integration of, on the neural side, motor
    information into, for example, visual information, and on the mental side, kinesthetic feel into, for
    example, virtually all aspects of the sensual, then it would seem that the restriction to monitoring
    one CA restricts the "movement," the communication, the integration of the corresponding
    quality, or if you will, the neural information into larger systems or entities. How is pattern, as
    such, communicated?
250) Note that if pattern in this sense is indeed transferred in some way from CA to CA, we have
    indeed taken a step from the strictly physical/neural to the mental. Previously, we have not seen
    an independent function, really, of the EDE. We might very well have asked repeatedly (and I
    did) why they were necessary; why pattern as such needed to be recognized, except that we need
    to account for how and why we do it. But now we have a reason, and it is a continuation of one of
    the functional building blocks ("necessities," if one does not wish the atomistic implications) of the
    CNS that I have hypothesized (i.e., conservation, stability). But the step from recognition and
    stabilization in a particular CA to the movement - the communication which must, clearly,
    ultimately be realized as the induction of pattern in other CAs - of the EDE's essence: its function
    and summarization of pattern, is a huge and radical one. If that does indeed happen, and we find
    it justified and functionally realizable, then we have found that a pure informational abstraction
    has been a) maintained, in some form, over different neural sets, b) communicated as an
    abstraction on top of a neural impulse base, and thus c) been assigned, as a primitive (perhaps)
    kind of meaning, to different neural sets and to different sets of neural impulses.
251) There may be two (at least) possible ways to answer this question. One is to continue the rather
    passive abstractive procedures, another is to consider the consequences of the ―locking-in‖
    dynamic. In terms of the first, we might be able to conclude that transference or communication of
    pattern is in fact unnecessary; that the same kind of successive abstraction that operates in the
    lower levels and continues within some category goes on here, in effect. That is, in the visual
    system, successive cell sets, leading to CAs, will continue processing the same visual entity, i.e.,
    angles to polygons to squares, perhaps. In that circumstance, there is continuation of the same
    abstractive line, the same kind, in the processing sequence, yet no communication of pattern;
    merely a physical continuation of the processing chain through the coincidence of mutual
    connectivity. This same structural ―forcing‖ of the processing dynamic may continue the effective
    presence of a particular EDE pattern merely through induction of the same or similar EDE in a CA
    performing the same or similar processing task. But this does not seem too satisfactory; why
    should the same function be preserved at these high levels, given the increasing flexibility of
    generalities over functions as abstraction increases?
252) The other alternative might depend on the ―locking-in‖ process (225 and following). Suppose
    that the EDEs are indeed separate, different, to some extent before this process, but similar due to
    the above considerations; after a neural ―stack‖ locks-in as a result of pattern matching (or
    alternatively, as a result of something like Edelman‘s competition for processing power), and what
    I termed the qualitative focal area (QFA) is formed, then the EDEs can be forced to combine, since
    the stack becomes something like one large CA. But how can this work, and how is an assembly
    ―like‖ a CA but not one (since the cell sets must be just too separate to form a CA in the sense I
    have described above)? In addition, the locking-in process can be a passive one if one only
    considers competition and resource allocation. However, in the case of representation, this is not
    sufficient.
253) One implication is that this process cannot end uni-modally; resulting patterns must be able to
    combine, as pattern. Why? Simply because our neural realizations of objects keep combining, in
    flexible, even arbitrary ways. That flexibility implies that there must be alternative possible inputs,
    through the sensory processes, and certainly through those of representation, into the same CAs
    synthesizing these patterns, and that implies that the patterns, as such, must be preserved in the
    inputs to those synthesizing processes. That is, the neural realization of some color might be
    conceivably the firing of one particular neuron (although given our discrimination of literally
    millions of hues, this seems wildly improbable), but given the flexibility of the CNS and of the
    sensorium, to depend on that neuron a) surviving, and b) remaining ―tuned‖ to that particular
    color seems extremely nonfunctional. In addition, we can keep proceeding up the abstractive and
    synthetic ladder here, and must at some time arrive at complexities with flexibility too great to be
    realized by any single neuron. Thus, at some level, a combination of dynamic patterns, EDEs, is
    necessary, one which results in further EDEs.
254) That is, given the necessity of flexibility in representation, what must be happening when we
    construct, for example, the visualization of a unicorn? How do we do this? I will assume that we
    do something like attempt to visualize a horse with a horn on its forehead. It is clear that we are
    ―assembling‖ those images, but what does that mean, when we are not dealing with images on
    this level, but with neural sets? To visualize or to conceive of a horse for a unicorn is to involve, I
    have assumed, the neural processes of locking-in those CAs, from the top down, relating to horse
    in general, to horses, to some specific visualization of a horse, and to its head, ultimately. But in
    addition, if this is a top down process, the idea of a horse with a horn on its head must direct,
    somehow, the total process. The basic problem here is whether we have to know what a horse is,
    what a horn is, and what it means to know these things. But there is no mental yet, just dynamic
    neural sets. That is, if we are dealing with neural sets, by neural sets, how can the choice and the
    movement of horn to horse‘s head be governed by knowledge, as such, as mental qualities or
    dynamics? If we were to fashion an explanation in passive connectionist terms, we would say
    something like: the idea of the unicorn corresponds to some particular neural activation, and
    because this idea is 1) different from that of a horse, and 2) has some, however vague, reference to
    the horn on its head, that it ―induces‖ activation in areas corresponding to the horse and horn, and
    because that activation is simultaneous (or nearly), it is ―seen‖ as belonging to the same object.
    Too many questions here.
255) Let us return to another consideration, that of pattern. Within sight, colors are combined, as are
    figures. How can this combination take place? It is easy enough to say that modules processing
    each aspect separately lead into a ―multi-modal‖ area, in which, say, the inputs are abstracted
    similarly to the abstracting processes within each module. However, this cannot work if it is seen
    as simply leading linearly to mind, for the simple reason that the basic information is not lost.
    That is, the processes within the modalities create new abstractions from the components, in
    which the basic information may be retained, or not. It usually is, however, although it is
    subordinated to the combination. Thus, if lines are combined into angles, or angles into figures,
    the angles are preferentially seen, as are the figures. With colors, the case is even clearer: purple is
    only seen with difficulty as red and blue, or whatever. Look at the difficulty artists have in
    training themselves to break down images into components they can paint, for example.
    However, this can be done, and indicates a surprising retention of the basic processes and
    information, which must be accounted for, as I have emphasized throughout. When we have
    multi-modal combinations, we have, for example, figures with color, objects with scent, and so
    forth, all of the multi-modal components of which are fairly easily separated. This argues, I think,
    either for different kinds of combining and abstracting processes than take place at uni-modal
    levels, or, and I think this more likely, the same kind of abstractive processes taking place over the
    cortex, both uni- and multi-modally, and processes of quality formation (generation) which must
    encompass large areas of the cortex simultaneously, but which ―note,‖ ―see,‖ ―react to‖ some set of the
    highest abstractive levels preferentially. This is a very strong argument, especially when extended
    to the emotions (given the location of their neural realizations), for the ERTAS theory, I believe.
    However, something like this multi-modal integration must also function, as response, at a level
    below mind, i.e., when we simply respond to the sensorium, we can do so in terms of multi-modal
    considerations, as do lower animals. Note the examples of insects and robots based on them which
    integrate leg movements. Mind, then, is not necessary for ―simple‖ responses of this type. This
    latter is, I believe, an argument for the representational theory of mind in this essay.

256) Let us speculate further on the mechanism of alteration of the contents, i.e., the patterns, of the
    CAs by the EDEs which conserve them. Given the locking-in process, an alteration of the feedback
    parameters of the EDE by another process would result in an alteration of the pattern both of that
    EDE and of the abstractive processes which it maintains. Thus this alteration, of parameters (as yet
    uninvestigated) of dynamic patterns superposed over normal abstractive processing would seem
    to be yet a further abstraction (above the EDE itself) from the base neural dynamic. In addition, the
    fact that this alteration is aimed (however it is actually realized) not at altering any specific neural
    parameter, but at altering a dynamic pattern as a whole, in order to produce a coherent change in the
    underlying neural dynamics, is the first indication, really, of processes which might be tentatively
    considered mental, i.e., oriented toward functional wholes corresponding to whole entities in the organism’s
    environment. All other processes, up to now, might be considered the result of the change in
    detailed aspects of CAs resulting passively (even if due to internal feedback) from changes which
    could functionally be described in terms of individual neural synapses; the functionality here,
    however, must be described in terms of the coherent alteration of whole patterns corresponding to
    qualities, i.e., in ways corresponding to the observed coherent alteration of qualities in the
    sensorium. Note that this is necessarily a top-down alteration, since that is where the wholes
    originate.
257) But given the interplay of memory, association (which should be mostly the result of bottom-up
    processing), and top-down control, there must be at least two types of alterations in a locked-in set
    of CAs. The above describes what might be termed ―internal‖ alterations, resulting from changes,
    wherever they originate, in processes within the CAs already part of the locked-in set of CAs.
    Further, though, the constructive nature of memory, the alterations that go on during evaluative
    thought, including representations, must result in the incorporation and exclusion of CAs from
    the locked-in set. That is, individual CAs at various levels within this set must be not merely
    changeable in the above sense, but replaceable as wholes. And more, that locked-in set as a whole
    must be able to change, as a result, again, of the various processes outlined here. That latter
    alteration, in fact, may correspond to experiences of overall alterations in the content of
    consciousness, while the other alterations, of individual components, would then correspond to
    more gradual changes of various sorts.
258) Another point is that of 160. That is, I speculated there that qualities are the result of the
    interactions between representation and the sensorium. Now, given the huge overlap between the
    two, one might argue that if the EDEs ―float‖ over the CAs in some way, then since virtually the
    same CAs are used in representations and sensorium they should exist there also. That is, the
    EDEs are not, according to this picture, induced by representation; they are present whether the
    CA is activated by representation or sensorium. However, representation and sensorium must
    also be differentiated, as I argued above, or we would blunder around in the constructs of our
    heads. Now this is an interesting possible function for the EDEs, as well as their maintaining CA
    stability. And indeed the different pattern of induction (top-down vs. bottom-up) of the same CA
    given its different function might induce EDEs that are different in the same CA for the same
    quality, and further, given their function as stabilizers, this also makes sense. The system might
    want to stabilize, not merely the dynamic pattern in the CA, but the differences between the
    pattern as representation and as sensorium.
259) Now to react to such a complex pattern in such a way as to create meaning, we must react to the
    pattern as function. A pattern merely as dynamic variation, even if associated or compared with
    other such patterns, remains meaningless. But to create function from pattern requires that the
    result be related to its precursors in such a way as to derive it as a particular consequence, in some
    manner, and to thus ―see‖ both of these as both form and function. We might note Trehub‘s (and
    also, I believe, Kosslyn‘s) assertions, for example, that retinotopic imagery determines the relevant
    contents of consciousness. The problems there are several, however. First, there is no homunculus.
    Second, there are many contents of consciousness that are, for example, imagistic, without there
    being a retinotopic (matched neural and figural) correlation: colors, for example. So there must be
    something else; however, it cannot be the case that the data, retinotopic or not, leading to the
    images is unused. It must be, then, that it is used functionally, i.e., that the relation generated
    through the abstractive processes is what is ―seen.‖ But what, neurally, is this relation? This is
    what we must investigate; however, it does seem that this leads us to the position above, in which
    I argued that abstractions: functions, envelopes, for example, of neural dynamics, i.e., aspects of
    the CNS that are not quite actual, are the basis of the mental. And if the argument here is correct,
    we have another support for that.

260) The RAS could serve as the master switch for modularity change. That is, I have not seen a way
    that the CNS, as conceived by the connectionists, can easily change its discharge patterns, except
    passively: as a result of radical changes in the sensorium. Yet aside from external cues which
    disrupt our attentional focus, we are capable (we believe) of causing that change internally, even if
    as a correction or alteration (taking Libet into account): voluntarily. But such a change in internal
    focus cannot merely be the result of a radical ―tipping‖ of the balance of circuits in dynamic
    equilibrium to some other configuration; again, this is too passive. What of voluntary change?
    What of changes that are brought about by ―mental effort,‖ an effort which we are aware of as
    such and with which we can, sometimes, ―tear‖ our attention away from (or toward) something?
    Here we can contemplate the excitatory effects of the RAS. What if these effects were not merely
    used to turn on the whole cortex, or that part of it necessary to focus passively on some sensorial
    input, but were used as well to actively (voluntarily) alter the areas stimulated, and
    simultaneously by its withdrawal (of activation) acting to inhibit the areas not directly stimulated?
    Indeed this latter effect would serve as a good mechanism for speeding attentional change in a
    threatening environment. That is, if it were not only stimulation but an accompanying (active)
    inhibition of the areas no longer necessary that was needed for a shift in attention, that shift would
    be slower and less automatic. Utilizing the RAS in the above way would take advantage of a kind
    of ―dead man‘s switch‖ in the cortex: its inclination to inhibition. Now if this is true, then the RAS
    returns to consideration as the location for consciousness, despite the recent work with the
    parietal lobe (but see 290 on the will).
261) In addition, if the speculations above (260 and 255) are correct, then the system for
    consciousness is an interesting mix of serialism and parallelism. That is, what we seem to have is
    one set of systems that induce, from bottom or top, sets of CA activations, largely through
    association. Acting at right angles to this, in effect, is a system (either in RAS/thalamus or parietal
    cortex) which activates and/or scans the activations of the serial system, in order to both further
    emphasize subsets, and enable change, i.e., enable non-associated CAs to be activated. The
    intersection of these two systems are what is in consciousness. This seems a rather inefficient
    system for precision; one which supports multiple but related activations, reflected in our
    difficulty with maintaining focus and with multiple unrelated foci; a system more suited to flight-
    or-fight, or finding food, or emotional interactions in general. That is, something with wide
    integrative scope in one phenomenal area, but little ability to pinpoint individual CAs within an
    activation ―stack‖.
262) Now how does representation fit into this; is it just the sensorium in reverse? Let us suppose
    that representation, as I have described it, is difficult; it is, after all, an induction in the reverse of
    the normal direction (see Grossberg‘s latest article), and in addition subject to the vagaries of
    creativity and desire, supposedly. Given this, might it not need help from the RAS? In that case,
    what we have is a system in which the sensorium finds a relatively easy path up the abstractive
    ladder, monitored and aided (in parallel) by the RAS; and a representational system which needs
    activation from the RAS to function, especially at lower levels of abstraction. This would explain a
    great deal of the difficulties we have in representation; in addition, it would make theories such as
    Baars‘ and Taylor‘s both correct. That is, theories in which consciousness originates cortically
    (parietal/frontal), and in the RAS would both be correct in that both areas are essential, for
    different reasons, and that qualities, initially realized in the cortex, must have the RAS to activate
    them (or at least to keep them activated). In addition, it would seem that consciousness itself lies
    somewhere in the interface between RAS (and/or whatever is directing RAS activation) and the
    cortical processing areas (which fits with the very early speculations, above, about the interface
    between representation and sensorium, although this is more complete), another reason both
    would be correct. That is, it is clear that we can have cortical activation without consciousness; it
    happens under anesthesia and when performing various activities (such as speaking, for example,
    where we are not conscious of grammatical processing). A decisive experiment here would be to
    find out whether any cortical activation without feedback to or from the RAS is experienced.
    Another would be to see whether we can have RAS activity without consciousness and if that
    activity does not tie to cortical activity. However, this experiment would only support or
    disconfirm this particular realization of this theory; it is certainly possible that the RAS is merely
    an activator, and that the interface is between the frontal or parietal and the other areas of the
    cortex, directing activation by the RAS. The theory would be the same, with some neurological
    structures functioning somewhat differently.
263) Questions that naturally arise here are, how is the RAS directed (or does it direct itself)? It must
    ―know‖ what it is activating, but how? How exactly are downward impulses induced, and
    induced coherently (to related CAs)? Is there actually a role for EDEs here? The answer must go
    something like this. In order for the RAS to be directed by downward processes initiating
    representation, so that it can ―turn on‖ the cortex in the appropriate areas, there cannot be specific
    ―messages‖ or ―pictures‖ sent to the RAS; we are dealing here with neural impulses. What must
    be the case, then, is that in the RAS/thalamus there is a map of the cortex, very small; a set of
    areas in the RAS corresponding to the areas in the cortex that it can activate. The ―controlling‖
    module(s) then must activate those areas in the RAS, which in turn must switch on the cortex,
    through its activating processes. Now, how do the controlling areas (the logical processing area(s),
    parietal cortex, prefrontal cortex, all the above???) activate the appropriate areas in the RAS? They
    in turn must have a map (or maps, if multiple areas) of the cortex, writ small, in effect, in
    themselves. We have then a very interesting system consisting of small cortical (and thalamic?)
    maps (in the ―controlling‖ areas) which must interact through a variety of feedbacks with each
    other and with the map in the RAS, and the cortex, to activate the cortex in order to produce
    representation. Where in this does mind originate? I still favor the RAS, for the reason that it is
    closer to the body. That is, the logical processes are late and not so important for us as sensing and
    operating our bodies and our emotions, and the RAS sits right down there. But this still does not
    answer the question of how mind, the mental, originates or is generated. Yet we approach closer,
    because in this system of maps activating maps activating cortical areas (and thalamic areas,
    perhaps, although we do not seem to do very well with emotional representation, in the
    generative sense in which I use the term), the imprecision of the smaller maps must control
    cortical areas and in turn be controlled by the precision of the activation of those areas.
264) But the problem with the RAS as ―the‖ center for consciousness is this: we experience too many
    details. That is, if what we experience is somehow either moved from the cortex to the RAS or is
    duplicated in the RAS, then how is it that we see the lines making up figures; hear the individual
    sounds in chords; how is it that smart people grasp, i.e., hold in consciousness, significantly more
    than the stupid? The RAS of all of us cannot hold or generate that detailed and varied
    information, that is the function of the cortex, and why duplicate that? So we must be ―seeing‖ the
    cortex, and the RAS must be illuminating that, i.e., making it possible, without ―containing‖ it.
    One might cite the ―gating‖ function of the RAS as an objection to this; the RAS/thalamus system,
    after all, seems to select, at least before conscious control, the sensory experiences we have. Does it
    not make sense that what is let through the gate is then experienced on its other side? But if the
    speculations in the above paragraphs are true, then what is happening is that cortical processing is
    processed through the thalamus, and rather broad areas of it are selected, and subsequently that
    selection reflects back to the cortex to re-stimulate or maintain the stimulation in the areas just
    selected so that they can be experienced. There must, for one thing, be a reason that we experience
    a ―conscious effort‖ in keeping our concentration on one topic: this must be an overriding of the
    RAS/thalamic gate by, probably, prefrontal processes determining RAS selection (again, see 290
    on the will).
265) Yet the counter-objection to the above is that destruction of the cortex, or of large parts of it,
    does not result in loss of consciousness but of conscious content. This would imply that the RAS
    stimulation is necessary, but that the ―center‖ of consciousness is indeed in the RAS/thalamus
    (ERTAS) area. Yet, against this, we find that the recent results in schizophrenia, a disorder of
    ―ownership‖ of conscious contents, i.e., probably the lack of integration of low-level bodily
    experiences with higher-level cortical/sensory experiences, result from thalamic disorder, but
    again do not entail loss of consciousness, but alterations in its content. But RAS itself might still be
    implicated.
266) But all the above neuroanatomy is irrelevant to the overriding question of how experiences are
    generated, whether cortically, in RAS, or both. And what seems to be unavoidable are the above
    considerations of pattern awareness which I explored at length earlier. That is, wherever the
    neural dynamics are in the CNS that are utilized for experiences, we experience them as one field,
    indeed as unified. That is, objects are not, without effort, resolved into their various components.
    Consciousness is a gestalt phenomenon. Now it is certainly possible to model this as a neural
    phenomenon even now in neural net models. But what do these model, really? The phenomenon
    of a small set of neurons responding to a ―figure‖ as a whole is certainly absolutely necessary to
    have and to understand, yet we do experience the object‘s components (which are also wholes, of
    course) as well as the whole, i.e., even if with effort, we can resolve objects into components,
    which components are also gestalt dynamics. So we have two considerations. First, neurally, we
    must have both components and wholes realized in neural dynamics. Second, all of those are
    unified when we experience, even if the set is reduced, i.e., even if the RAS/thalamic gate limits
    the ―number‖ of these components. Let us assume that these components are generated
    separately; the results of the ―attentional blindness‖ studies indicate that this is at least partially
    true. But as our eyes, for example, rove over some restricted pathway in a scene (but remember
    that we do not just see on the fovea), the aspects of that pathway are unified, probably through the
    reentrant firing of the ―old‖ trace. The length of self-stimulation of cortical circuits is, I recall, long
    enough for that; and this phenomenon in fact supports my speculations above as to the nature of
    the EDE. But the visual experience is only part of our total conscious experience at some instant.
    But in addition it must be the case that we have multiple instances of this unity, since in order to
    control it, i.e., to plan, to play, to inhibit, those areas responsible for such overriding control must
    have at least a simplified version of this pattern and in addition a map of where to apply it, i.e., a
    functional map of the cortex. It is in the interaction between the cortex and the maps of it which
    must provide the basis for the mental, in the generation of ―shortcuts‖ for activation and
    inhibition which make the maps actually representations, in some sense. But in what sense?
267) Also, there is the possibility that the focusing (hypothetically) produced by the RAS/thalamic
    gate cannot encompass all experience, because of the figure-ground phenomenon. That is, it may
    be that what passes through the gate, what is emphasized by the RAS, what ―wins‖ the
    Baars/Edelmann ―competition,‖ is merely the foreground, or, since background context
    influences the foreground, even the background in Gurwitch‘s sense, which does not influence the
    foreground (a point with which I disagree), or influences it minimally. That is, if the ―peripheral‖
    experiences we have are not the result of the gating, then it must be the case that the ERTAS is not
    the generator of consciousness.
268) Another point: the current hypothesis that the hypothalamus (HP) ―contains‖ short-term
    memory, which it then transfers to the cortex, cannot be the case. The HP is too small, and again,
    there is no point in duplicating the cortex. Instead, what must be happening is that the HP has a
    map of the cortex, which it employs in two ways; first, this map is partially induced or configured
    by cortical stimulation, and thus ―stores‖ the ―location‖ of information or the latest stimulus in
    the cortex, without the necessity of reproducing the cortex. Second, this map then must be
    internally reinforced and then stimulate the cortex in order to ―burn in‖ or allow the GABA
    receptors to create the long-term changes necessary for permanent memory. How does the first
    work? There must be, initially, a rough cortical map in the HP. Intense enough stimulation, or a
    particular type of stimulation following enough attention and/or the ―desire to remember‖
    something then must induce cortical areas to activate corresponding HP areas. These latter areas
    cannot do abstractive processing, so the connections between the HP and the cortex must be at the
    end-points of cortical processing. What is a cortical processing endpoint? It must be involved with
    the top-down feedback from, say, the prefrontal areas. Thus that latter feedback creates,
    functionally, the endpoints of cortical columns, which must in turn be connected to the HP, both
    ways.
269) We must generate neural dynamics which realize these cortical components as singular patterns,
    across all cortical (etc.) processing we experience as a unity at any given experienced instant. I
    have outlined above some general characteristics of such patterns; now we need to go into more
    detail. When we learn actions, ideas, languages, etc., we must be conscious of the minutest details
    of the processes until we learn them. This would seem to indicate that learning behaviors that are
    not instinctive requires consciousness, or at least mind; and conversely, that instinctive behaviors
    do not require mind, and thus that animals which have nothing but instinctive behaviors that can
    be only modified in simple ways are not conscious (I will pass over the issue of whether
    consciousness is necessary for mind, that is, whether any mental operation had to be conscious at
    some point, if only when originated: Searle‘s contention.). This ties in with the above observations
    about play. That is, actions whose meanings can change require mind. Actions whose
    modifications do not change their meanings do not require mind. But the meaning of an action is
    related to its function; in fact, it is in a sense the awareness of its function. Mind must then be
    involved with maintaining and changing functions. But the function of some neural dynamic
    relates to how it combines with other neural dynamics, that is, to the process of the formation of
    what I have been terming abstractions, and the top-down processes involved with ―locking-in.‖
    But we have seen that abstractions ,in order to be experienced as wholes, must be integrated into
    singular dynamic neural patterns. That is, it is not sufficient that some set of neurons fires in
    accordance with some specific rule or in response to some specific sensation, such as a rectangle.
    In order for a rectangle, ultimately, to be experienced, it is necessary that more than some single
    neuron fires as a result of the set processing the rectangle‘s components, however necessary it is
    for that set to exist and to induce the firing of neurons leading from it, to enter into other sets. The
    firing of one neuron, however specifically it is induced or it is responding, cannot contain or
    realize qualities. The qualities must come from the pattern, the function (where the term might be
    used in something like its mathematical sense), inducing that neuron to fire. But for qualities to
    come from that pattern, it must be apprehended, ultimately, as a singular entity, its function.
270) Mind, then, is involved in the establishing and change of that entity, if it is not hard-wired. But
    for it not to be hard-wired, for it to be able to change its function, implies that the sets of cells, the
    neural dynamics, are very flexible in that some sets of cells, creating/realizing some function, can
    be changed, i.e., the function can change, and/or in addition that set can be assigned to some
    other set of cells as its input: that the output of a CA may be redirected to another set of cells,
    another CA. But learning a skill is a top-down process, usually; we start with the goal in mind:
    that which we want to learn; we must in that case be conscious or have in mind that goal, even if
    that goal involves merely avoiding pain. Sensations are consciousness because they must be
    opportunities to learn and to have raw material to modify, but also because they must be the
    material of our representations: we must literally be able to modify our sensations; that is, when
    we behave in response to something, our sensory field changes as a result of that behavior; we
    have modified our sensations, which we understand as modifying our surroundings. But as long
    as we do not modify how we modify our behaviors it is not necessary that it is conscious: that we
    have mind. But the question is, why does mind arise with the modification of assignment of
    function?
271) What may relate to this follows from the above point. That is, given that some set of cells,
    however widely distributed, acts as a CA and realizes some mental quality (we assume), then that
    CA must, functionally at least, be operated on more-or-less as a unit. That is, however distributed
    the dynamic neural pattern is that realizes red, that pattern, when we use red in representations,
    must be both flexible in what it can attach to and stable in that over varying attachments it
    remains similar enough to be ―understood as‖ red. What does this imply? That the CA realizing
    red must act as a unit to a great extent, i.e., that it must have variable inputs and outputs; it must
    be able to be ―wired‖ to different CAs. This is an interesting speculation, and it seems to imply
    some sort of differential treatment of flows to and from CAs, and the flows within CAs. Yet, on
    the most abstract levels, at least, CAs must be themselves flexible, and indeed the point of
    representation is to create different CAs from more ―elemental‖ CAs which themselves must be
    flexible to some extent. Yet if the bias of the whole CNS is toward stability, as I have speculated,
    then this property of CAs must go against the grain, so to speak, as I have mentioned above. This
    would indeed lead to stability for CAs, and problems in creating new representations, as indeed
    we do find. And this should be the case; the opposite would, I would think, make coherent
    thought extremely difficult. So here we have a reason to treat CAs, locked-in or not, as functional
    units; another step toward mind.

272) Here‘s how the above might happen, and can be conceptualized. When we hear words, we are
    taking an analog input, over a short time interval, and classifying it into particular categories
    (words) which have rather sharp boundaries, functionally. That is, within any word type, there is
    fairly high variation between tokens, but once a type is perceived to change, the token variation
    now takes place within a radically different class, the other word. So boundaries are created
    within a continuum in order to a) separate word tokens, then b) separate word types. Suppose,
    now, that we take patterns of synapse stimulation over a single cell, and speculate that these may
    be classified, functionally, by the cell into patterns (as defined by output) corresponding to word
    tokens and types. If we can have internal feedback, pre-discharge, with both positive and negative
    feedback on the cell surface, this might be possible. Now extend this to the CA. Surely here it is
    possible to create the functional category boundaries that words demand. Thus, it is possible,
    certainly over a CA, to regard inputs to that CA as functioning like words, in the sense that they
    are segregated, functionally, into token classes, then into type classes. This gives this analysis
    another dimension, beyond the EDE, above, for the creation of abstractions. These classes are such
    abstractions, where their meaning is a functional one, defined in terms of the output classes
    resulting from the various input classes. But the segregation of both of these classes according to
    rather sharp boundaries, analogously to word boundaries, is necessary for this. It is also possible,
    then, given that this takes place over a CA, to conceive of word outputs from a single CA as well
    as word inputs. This is more difficult to understand in terms of single neurons; if a word requires
    a spatial/temporal distribution of coding flow, then one neuron, with one axonal output, unless it
    branches extensively and controls, on the branchings, the spike train distributions as functions of
    the processing of the input, cannot output words. CAs can, however.
273) Let us now return to the considerations at the very beginning of this essay: transduction and the
    problem of symbolic anchoring. I determined that, logically speaking, there was no reason for the
    CNS to prefer any type of correspondence between neural code and quality dimension. But after
    reading Pattee, we might note one point. His article on the Physics of Symbols (and others) seems,
    really, a kind of cop-out, to the extent that, at the very end, the only real difference between
    computer emulations of the brain and the analog processes of the CNS is the practical difference
    in the difficulty of the emulation (and of course the fact that it is an emulation – but that does not
    seem so relevant to him) vs. the reality of the realized operations in the CNS. Now we might
    answer in terms of the analog basis of operator gradients: operations between the standard, well-
    defined operations in the digital world. However, the relevance here is different; Pattee points out
    that biological systems have evolved to make certain measurements, in effect, certain choices as to
    stimuli to take note of, over the course of time. We might claim, then, that this resolves the
    transduction dilemma, in a way. That is, what is, when arbitrary, i.e., the choice of what
    dimension of the stimulus to have the neural codes be analogous to, an issue that is unresolvable
    and irrelevant to the CNS becomes, perhaps, quite relevant when we assume that evolution has
    made those choices, in effect, non-arbitrary, i.e., that the CNS is hard-wired, willy-nilly, to create
    certain structural correspondences. Now if this is true, then the symbolic anchoring problem, as
    Pattee no doubt realized, has a solution in the selections that the CNS has, over the course of
    evolution and perhaps purely by chance, made. The question is, then, given that this is a) a
    solution to the symbolic anchoring problem, and b) that the CNS has created particular solutions,
    whether this in any way creates a door into the mental, or whether it merely provides a solution to
    the relationship between the higher-level digital coding and the lower-level analog coding.
274) In the cortex, cells for the specifics of, say, tongue movement fire, leading to a cell or small set
    which fires to unite and abstract those firings. But it must be the totality of that set that is in
    consciousness, because we feel the individual parts of the tongue and the overall motion. So those
    firings must be united; but they are not united on the cortex; the single (say) cell that fires does not
    unite the lower cells, just abstracts from them. Let us assume that there are inputs to the RAS.
    Now, the thalamus must ready part of the map in the RAS of the cortex to fire; that has to be how
    the selection, at least in the preliminary stages, works. Also, there must be inward filtering of the
    cortical inputs. Yet all the inputs for this tongue movement, in order for it to be experienced, must
    have processes entering the RAS. There they are channeled (passively?) to the RAS map of the
    cortex. Now here they must be unified; yet what is being unified is not the cortical information.
    This is very important to note. What is being unified is not the cortical information, but only
    information about where on the cortex the activation is taking place; the type of information is
    then discovered in the RAS through its own internal system from its map, activated by the cortical
    inputs. Then the RAS must feed back out to the cortex, to activate the same area in order to keep it
    firing and to produce the ‗locking-in‖ phenomenon. Thus consciousness, as unifying
    phenomenon, must take place in the RAS. Consciousness, as detailed experiencing, must take
    place in the cortex (or wherever the details of the experiencing, as pattern, are activated). But these
    two must be unified.
275) But what this all comes down to, every time, is the reaction of one neural circuit against
    another… or several against several, to produce an internal response. It simply does not matter
    whether we consider these circuits special, complex, simple, distributed, coordinated temporally
    or spatially in some way, however exotic. If there is no homunculus, if we are materialists, if we
    have one reasonably unified consciousness, even moment-to-moment, and if we cannot have
    ―information‖ differentiated as to quality (i.e., ―there are no red neural impulses or codes‖), we
    must bite the bullet and consider the internal aspects of neural circuit reactivity. But the problem
    is, at this point, not really so much how we might do this, but the initial problem of how to
    differentiate the internal neural reactivity (INR) producing qualities: the mental: whatever term
    you want, from internal reactivities that do not. Take the lowly thermostat, a nice feedback device
    with, therefore, internal reactivity. We have those maintaining that thereby it has ―feels.‖ Yet this
    must be absurd, if we differentiate mind at all from the mindless; unless we do indeed want to
    embrace an all-encompassing pan-psychism, which I definitely do not, then we must maintain
    that such a simple construct has no internal mental properties. Fine, but at what level of
    complexity, then, do we find them, and why? I have analyzed, above, various types of feedback,
    and concluded that if any can demonstrate the mental, it must be the recursive structures. But that
    analysis was based solely on a intuition about complexity, and did not bring us to the mental; and
    further, the recursive depth of the CNS surely is limited; further, suppose that I am right; the next
    step, the decisive step, is still missing: turning that recursive INR into the mental.
276) Here is another possibility. Suppose that the initial determining factor for the generation of
    qualities is, in fact, their literal physical realization in the brain. Aside from speculating, first, on
    whether this is arbitrary – which it must be, given the arguments early in this essay – or second,
    on how detailed this is, one certainly can claim that the general categories of space, temporal
    succession, and causality are realized in the physical substrate of the brain and neural codes. Then
    we can ask how these characteristics are ―read out.‖ But that question leads to asking what
    ―reading out‖ is, and thus what it is to ―be‖ mental. However, it seems that we might have
    separated the issue into two questions at this point. Thus, the question of where the bases for
    qualia come from might be answered by referring to the reading of the above (Kantian) categories;
    and this leaves the question of why there is ―mental‖ at all, which ties to the question of what
    ―mental‖ really is. Let us assume that there must be a homunculus, however strange that seems.
    What then is a homunculus? Something which sees patterns. What can see a pattern, as such? That
    must be a pattern itself. What is a pattern? A pattern must be a set of parallel processes, but in a
    special way, that they must interact with each other in the manner described above, i.e., with
    below-relaxation time interactions. But we have to be careful here, because the problem becomes,
    now, to explain what it is to ―see.‖ That is, even with below-relaxation time response, reacting to a
    pattern only produces a pattern if there is a unitary phenomenon reacting to it. Another way is to
    take passes through the data, and add the elements of each pass to the pattern: have them modify,
    successively, the pattern. What if the data, the changes in the pattern, can only be made in certain
    ways, ways compatible with the basic limitations on the qualities, i.e., spaces cannot overlap,
    times are successive, etc.?

277) Let us suppose that there is an organism – a self-maintaining complex, if you wish: a ―virtual
    homunculus‖ (see below) – in our brains; that this complex is comprised of dynamic neural
    patterns, i.e. discharges, potentials, ionic flows, but not neurons themselves. This complex
    attempts to maintain itself, but dies out fairly easily; it utilizes inputs from many brain areas and
    can sometimes be very spread-out over the brain, sometimes more localized. It ―feeds on‖
    sensations (including emotions) and, I think, consists of and perhaps ―excretes‖ representations,
    primarily; that is, those, as fairly constant inputs, are necessary to maintain it, and are transformed
    internally into its self-maintaining dynamics. This sounds pretty far fetched, doesn‘t it? Something
    like an amoeba made of neural dynamics living in our brains. But this would begin to answer
    many of the questions. The two key concepts here are that of the entity made of neural dynamics,
    and that of its internalization of self-maintenance and thus various goals (see below). These enable
    the mind-body gap to be bridged, since they necessitate an internal viewpoint, a dynamic
    homunculus. There is a definite similarity between Kant‘s idea of the correspondence between
    internal and external teleologies and this idea.
278) But let us combine the above with a concretization of the idea of teleology. Teleology may be
    understood in the following fashion, starting with an illustration: given a simple organism like an
    amoeba, this entity consists, in part, of a set of structures which embody, realize, the dynamics of
    food-seeking. This is a fairly self-contained dynamic, driven, say, by concentration gradients of
    certain chemicals in the solution, which start reactions in receptors in the membrane (which I will
    term ―micro-dynamics‖), which start changes in cellular structure; all of which result in the
    amoeba moving toward food. Then what? If this were the only system, the amoeba would
    nonetheless starve. But when it contacts, or gets close enough to the food, another dynamic starts,
    which involves engulfing the food, digesting it, and converting it to amoeba. Now, either of these
    dynamics can be explained in terms of microstructure (molecular and atomic structures of
    receptors, etc.) and microchemistry (i.e., individual reactions and chemicals). But the micro-
    dynamics of food-seeking stops before, and does not continue into, the micro-dynamics of eating.
    The two are not necessarily connected, only joined by likelihoods. Yet without ingesting, food-
    seeking is meaningless. The opposite is not (strictly speaking) true; an amoeba could come upon
    food by chance. These divisions are not arbitrary; an amoeba seeks food without ingesting it, and
    ingests without seeking it; in addition, different chemical and physical events take place in each.
    There is then a dependence on an abstraction, and a direction of that dependence, i.e., a teleology.
    Food-seeking is literally given meaning by ingestion, and not by food-seeking‘s internal micro-
    dynamics, nor, indeed by ingestion‘s internal micro-dynamics. That is, ingestion cannot be
    explained as a set of micro-causes of food-seeking; nor can food-seeking be explained as a set of
    micro-causes of ingestion. Only considered as systems can the interaction of the micro-dynamics
    be explained. In contrast, every aspect of food-seeking is explained by every other aspect: once the
    sequence is initiated (and it seems to be running almost constantly) it runs automatically.
    Similarly for ingestion. Thus, teleology consists in a necessary (usually directional) relationship
    between systems which does not exist at a level beneath the system level. This gives meaning, as
    well, to the concepts of emergence and of supervenience. Now, what similar situations are there in
    the CNS? Neural micro-dynamics are just as fixed, automatic, algorithmic, as are the micro-
    dynamics of amoebic food-seeking, and just as meaningless.
279) Considering the simplest feedback loop, we must conclude that a single loop is not teleological
    in the above sense. That requires not merely two connected loops, but loops which are connected
    but independent, in this sense: that they will run without the other, but in addition, that they are
    functionally dependent. Now one might compare the above to looping and branching in computer
    programs: a while or if loop going until a specified input interrupts it. Does a computer, then,
    have teleology? Why not, we‘ve built them. What of a robot, or even more interesting, perhaps, an
    ―intelligent agent‖ or bit of ―artificial life‖ in a program. What of ―food seeking‖ vs. ―ingestion‖
    within a program? Well, we built all that too. The question, however, is not about the teleology,
    really, it is about the significance of that teleology in the generation of mind.
280) But if we take the example of 277, where the entity is living in our heads, comprised of EDEs,
    then the situation becomes more interesting. We can not only talk about teleology a realized by
    neural dynamics, but we can talk, perhaps, about teleological reflexivity.

281) Now we must go back to the beginning, and contemplate transduction. The original intent of the
    transduction analysis was to ―anchor‖ symbols, i.e., to enable arbitrary symbols or symbolic
    sequences to have meaning, by taking some aspect of the world they were symbolizing and
    literally incorporating that into the symbol, or at least into its generation. Thus, frequency can be
    incorporated into the neural representation of pitch, in the frequency of neural impulses. But there
    are two problems with this. One of them I described above, very early in this overly long essay:
    that this incorporation is itself entirely arbitrary, and that arbitrariness leads, in effect, to an
    infinite regress. The other problem was this: just what is a symbol, anyway? It is easy enough to
    glibly talk about symbols and ―strings of abstract representations‖ or some such, but what are
    those, either neurally , or even in a digital computer? Let us take the computer as an easy example.
    A register which performs symbolic manipulation will take a string of bits, 1s and 0s, since that is
    literally all a computer can work with, and in addition an instruction set. This instruction set does
    not consist of symbols either; it must consist of burned-in circuitry which directs the manipulation
    of the 1s and 0s according to rules analogous to Boolean rules, although they do not have to be
    Boolean, or even consistent. A computer program then must consist itself of a pattern of bits
    which can be read as meta-instructions to activate parts of the instruction set, by matching the
    incoming instructions to patterns of bits which activate parts of the instruction set, which in turn
    consists of the above circuits. These must find data at particular coordinates in RAM and operate
    on that data… and so forth. Now where in all this are symbols? All of what is going on here is
    symbolic, in some sense; surely strings of bits are symbols; but in what sense? All the bit strings
    do is activate circuits or serve as markers to be matched to other markers. The real operations here
    consists of the Boolean manipulations, which are themselves performed by ―logic‖ circuits, which
    themselves consists of AND, OR, NOT, etc., ―gates‖, which in turn are small simple circuits which
    react to multiple electrical inputs with an automatic output. The variable elements in this system,
    the memory and the program, are bit strings inputted by the programmers which control which of
    those circuits operate on which strings. To put it another way, nothing represents, in this system,
    and it is in a sense even strange to speak of ―symbol manipulation‖ here. Bit strings are not
    ―manipulated,‖ they merely serve as inputs to logic devices which then output other bit strings,
    where the set and sequence of those devices is determined by yet other bit strings. What is it that
    makes these entities ―symbols‖, if anything? One might say that it is only when they are
    interpreted by human beings that they acquire meaning, and thus become symbols, but the
    problem with that is that there are operations in the CNS going on which, if not the same as those
    in digital computers, are equally asymbolic.
282) In computers, the only thing that could create or realize a symbol is functional dynamics. But
    what is that? We must, I think, say something to the effect that there is a structural - in the sense of
    operational structure, i.e., dynamic structure - identity between the operations on the bit strings
    and what they are ―supposed to‖ correspond to. If an alien from an other galaxy looked at an OR
    gate, and if that alien had knowledge of logic, it would understand that this device could possibly,
    at least, be an element in a logical machine. Is the gate itself, the physical entity, a symbol? It does
    have a particular structure which can only result, if properly employed, in one outcome, but this is
    not enough. That is, the structural correspondence to something else is the beginning of what
    creates a symbol; but this structural correspondence itself does not exist, in a sense: it does not
    matter, and indeed can be hidden in other structures (see the essay on transduction above),
    without that correspondence being utilized. That is, there may be innumerable such structural
    correspondences in the world (the example of programs in the random configurations of plaster
    on the wall); unless they are actually, actively, used to mimic, i.e., as substitutes for that which they
    correspond to they are not symbols. This then answers the question above as to whether a circuit
    itself is a symbol: only if it is employed, as a circuit, to realize a function corresponding to some
    other dynamic. Now how precisely does this circuit correspond? It corresponds to a single
    operation in a type of algebra, in which two, say, inputs are combined according to the rules of an
    OR truth-table and a particular set of outputs results, depending on the inputs. What if this
    correspondence is coincidence, the random working-out of thermal patterns in molecules in
    plaster? Again, the correspondence, even the dynamic correspondence, is not enough. That
    dynamic must be used as a substitute.
283) We can recast that question as: what does ―intended to be used‖ consist of? Certainly, on the one
    hand, the computer is the embodiment of its builders‘ intentions. On the other hand, what of the
    CNS, or a computer/robot which is now carrying out its own goal-seeking? We must take ―use as
    a substitute‖ then to mean not merely the activation of the potential dynamics of the structures
    that correspond, but the activation in order to realize that correspondence. The dynamics of the
    representation is manipulated instead of the manipulation of the reality. Thus, the establishment
    and the manipulation of the symbol – the structured entity - as a substitute for the reality to which it
    corresponds - the comparison, in this indirect sense, of that symbol and that reality - is of
    paramount importance, and is what, really, establishes that structural correspondence as, in fact,
    an actual correspondence, rather than a coincidence or fortuitous happening. Now this is not an
    original conclusion, but we have arrived at it through a consideration of the simplest case, that of
    the computer, where we must first acknowledge that symbols are manipulated, and second, that
    nonetheless they are arbitrary and meaningless internally, yet, third, are still manipulated as if
    meaningful.
284) The point of these last paragraphs, first, is to really establish what ―symbol‖ means; second, to
    establish that it is not enough to say that a concept is just a stable or recurrent set of neurons or
    neural dynamics (as has been recently said), i.e., that that this set must interact dynamically in
    some way corresponding to what it symbolizes. In addition, this relates both to the teleological
    digression above, in that such a set, because of such interactions, is such a realization of teleology;
    and further, this relates to the very earliest discussion of transduction. That is, it is possible to see
    a symbol, in these terms, as a kind of reverse transduction, similar to the ―top-down‖ activation I
    have been talking about, in which the structural operations (which will be further investigated
    below) help to determine the meaning of the symbol. It may be, however, that in addition, the
    bottom-up, original meaning of transduction, in which actual physical characteristics of the signal
    play a part in the symbol‘s meaning interacts with this top-down structural aspect to determine
    both meaning and quality.
285) But the problem of qualia remains. A symbol in these terms is the function of a set of neural
    circuits, and this function is identical with the corresponding function in some sensory or effector
    circuits. But the symbol, to have meaning, i.e., qualities – because meaning in purely functional
    terms is no more than an abstraction of its function, that any computer possesses – must, again, be
    reacted to as a whole, a single entity. The meaning of a symbol, internally, must be that reaction:
    the reaction to the symbol‘s function. But a symbol‘s function cannot necessarily be a whole in the
    above sense: in the sense of an EDE, simply because they are too flexible and unstable, and the
    dynamics of such functions are realized in large and relatively slowly-finishing dynamics. The
    whole, then, must be in the EDE which reacts to the function, rather than to one realized in the
    function. We must ask, then, why is there something reacting to the functions of symbols, and
    how does it react? In addition, we see that one reacting dynamic is not enough for control, only for
    evaluation at the most. Thus there must be two, one to react to the first. Now in addition, this can
    be seen as continuing the abstractive processing that is hypothesized to be the basis of all CNS
    processing. The symbol per se is at the top of such an abstractive hierarchy, but it must consist of
    functions, i.e., the dynamic interrelations and induction of various responses. However, the stage
    which reacts to this in order to turn it into a whole, i.e., which uses it as input to modify and
    generate the activity of an EDE, can be seen as continuing the abstractive process. Then we must,
    as I said, have another EDE reacting to this first one.

286) Now let us contemplate feedback again. Suppose that a set of neurons has to change in response
    to environmental change, better adaptation, or whatever. I have assumed that the set above this
    one, the more abstract set, will use top-down feedback to stabilize the lower sets. But if there is
    enough pressure for change, then that feedback must be overridden by the strength of the signal
    from below, its, variety, other reinforcing systems, etc., and the next level of abstraction must be
    modified to create, in effect, a new figure from the changed components. Alternatively, what
    might happen is that a new or another set of higher-level neurons could be selected, i.e., the old
    ones could remain and the next level of abstraction of the given lower set could just switch from
    one set to another. This could also happen top-down; if a set was generated at a higher level, it
    could either create a lower-level set or it could recruit some set from the existing lower-level sets.
    Thus the lower-level set now has a choice of which higher-level set to activate, and there must be
    feedback stabilizing this choice, or in some cases destabilizing it. This must be the neural
    equivalent of the counterfactual, controlled by a switching feedback rather than an amplitude
    feedback system. What signals the switching? We might, first, ask the same question about the
    amplitude feedback; there is a sensor set to amplitude. In switching feedback there must be a
    sensor set to type, i.e., to a differential response depending on the type of output from the
    particular neural set. Here we have an indication of the necessity for a qualitative sensing and
    maintenance system in the CNS, i.e., one which deals with content rather than magnitude.
287) The next question then is whether there is structural correspondence in the dynamics of an EDE
    with real-world entities, a problem similar to Harnad's "symbolic anchoring", but taken up to that
    higher level of abstraction within the system. Then there is the problem of the interrelation of the
    EDEs, which I think might be approached through a kind of realization of the "functional micro-
    teleology" (above, and see Sun‘s recent ―Teleology of Consciousness‖). To put this another way,
    multi-modal integration must give meaning to modal driving, and the will, i.e., symbolic
    alteration of that integration, must be the other component.
288) That is, more explicitly, and especially given the phenomena of colors, it becomes clear that we
    are not anchored at all to the ―world‖, very simply, because there not only are no ―colors‖ in the
    world, there are no hues, no color combinations, no ―primary‖ vs. ―secondary‖ colors in the
    world, nor anything corresponding to them. We might say the same kind of thing about touch. We
    experience only our own ―codes‖, ―representations‖, ―transductions‖. Now suppose that we were
    contriving a system which was to successively abstract the transductions of the world, i.e., the
    neural reactions to it, in order to find regularities and patterns. But further, in order to identify a
    pattern as such, the system clearly can not react to merely a part of that pattern; it must react to
    the pattern as a whole. The successive sensory abstractions are clearly doing this.

289) What then must we conclude, to fix the mental to the physical? We must return to 277, above, and
    conclude that we, our consciousness, is what might be termed a ―virtual homunculus‖ (VH) (see
    Tannenbaum‘s article [Tannenbaum, 2001 #299] on consciousness as a ―sense‖; see also
    [Armstrong, 1970 #300], and ―internal scanning‖). That is, it must be that the internal
    manipulations of some ―central‖ neural dynamic entity, perhaps an EDE, will mirror those of the
    ―external‖ embodied entity. In order to reach, mentally, for an object our VH must make the same
    reaching motions; which as we have seen are the neural impulses, the neural dynamics, of
    reaching. It must activate these dynamics as we reach, i.e., not explicitly, but through the desire to
    reach an object. But the object is again a virtual object in the same sense. More on this below.
290) To briefly mention the ―will‖: consider the needs, the teleology of the animal, the VH. Given that
    this is maintained by the CNS, indeed is self-maintained, it must then have internal goals having
    to do with this self-maintenance, and these goals, I claim, interacting with its manipulations of
    ―internal reality‖, and our interactions with our real-world environments, become our willed
    choices. One can see, then, how ―willing‖ something might be an effort, and one of which we are
    conscious. If the VH proceeds against the sensory stream or against some conditioned bias in
    order to achieve its internal, overriding goals then that overriding must be felt as ―willed effort‖.
291) Suppose those goals incorporate an abstraction of the system‘s goal of stability. What form
    would that abstraction take? We must keep in mind that so far none of this can be, strictly
    speaking, mental. That is, thinking in terms of ―goals‖ must be done very carefully, so that
    Tannenbaum and Anderson‘s mistakes, that of taking a ―sense‖ of one‘s internal workings as
    implicitly producing phenomenal consciousness, is not made. That is, a purely neural feedback
    and monitoring must be kept in that realm of discourse until it becomes very clear that it is
    appropriate to speak of the ―mental‖. Now, given that, how indeed can we speak of the ―goals‖ of
    an ―organism‖, complex, or what-have-you, situated in the CNS, ―floating‖, so to speak, on the
    neural currents? Those goals must be put in terms of the needs of that complex, as in 290. Thus,
    we might speculate on self-maintenance, to start.
292) If there is such a dynamic neural organism, self-maintaining neural dynamic, VH, or what-have-
    you, then, continuing the assumption of self-organization in a varying (neural) environment, it
    must variously accept and/or reject inputs or stimulation of some sort. Some neural inputs to this
    VH might aid it‘s self-maintenance, and some might, we assume, harm it, i.e., cause it to become
    unstable, to dissolve, to shrink in some manner, or perhaps to be restricted in its domain (in the
    neural flows). Thus it must discriminate between ―helpful‖, let us say, and ―unhelpful‖ neural
    dynamics. Leaving aside for now exactly what those might be, we must note that this presents us
    with the necessity for recognizing that neural dynamics be divided into categories which are
    dependent functionally on pattern. ―Good‖ patterns do one thing for our VH, ―bad‖ patterns do
    another. Could this be based on neural microdynamics and not on ―pattern‖? It could, but that
    would be irrelevant. That is, previously, I have been attempting to find a way to make pattern
    functionally intrinsic to the system; yet if we consider that it is not, in this case, actually the
    ―pattern‖, the neural dynamics as either ―individual‖ potentials or trains, or the dynamics as some
    abstraction of those dynamics, which is really relevant. What is relevant is the effect of whatever
    input the VH has, on its own dynamics. This effect is the functional equivalent of the dynamic
    pattern, and, as an effect on the stability (etc.) of the VH, must be reacted to as such. So we are able
    to bypass, in effect, the whole ―pattern‖ problem, since it now takes the form of the mechanics or
    functional realization of the VH‘s maintenance of self-stability. Note that this could not happen in
    the CNS per se, without some sort of VH, simply because the CNS cannot afford to reject stimuli
    unless they are extremely noxious, i.e., harmful to the body. This takes that one step up in
    abstraction, in effect, except here the VH ―body‖ and its ―CNS‖ are identical.
293) To clarify, what hypothesizing the VH does is necessitate the reaction to pattern in a way that
    the EDE did not. The EDE is dependent on what might be fragile relationships between neural
    timings, while the VH does not seem to necessitate those. But note that we have not yet taken the
    step into the mental realm. We are now able to speak, in effect, in terms of an organism, a unified
    neural dynamic, and types of wholistic responses comparable to those of non-conscious animals,
    at any rate.
294) Now the next step would seem to be relating the fact that the neural environment of the VH is
    not merely an environment, but one composed of codes for a reality external to that environment,
    i.e., for a reality which does not, per se, impinge on the VH. Yet for the VH to make ―us‖
    conscious of the external environment, or at least to create the mental, those codes must interact
    with the neural environment of the VH. The simplest overall discriminations, which might serve
    as a starting point here, are modal differences, i.e., the sensory modalities, and/or emotional
    discriminations. And indeed the latter seem even more fundamental; if we look at some, at any
    rate, developmental studies (i.e., Stern), we see that infants‘ early ―life-worlds‖ are discriminated
    more by emotional valences than by sensory distinctions (an interesting comment on the latest
    work on the importance of emotions for consciousness). Of course, at this stage an ―emotion‖ is
    nothing more than a dynamic neural pattern in some particular location(s) in the CNS. It looks as
    if we must then assume that the VH is intrinsically biased toward and away from different subsets
    of these dynamics; or it might be the case that some subset ―enhances‖ the VH in some way and
    others ―diminish‖ it. I don‘t think it matters too much how the biases are initially set up at this
    point.
295) So what we have, then, are biases both ―without‖ and ―within‖ the VH toward and away from
    certain neural dynamics corresponding (ultimately) to emotions. Now what if the VH
    manipulated the neural currents in order to alter those dynamics? Before we consider that,
    however, we need to look more closely at the assumption I made above. To assume that the VH
    manipulates ―neural currents associated with emotions‖ in order to alter, ultimately, those
    emotions, is to assume that the VH is able to react to patterns of currents, and not merely to the
    microstructure of currents. Now can we justify that? It is easy, certainly, to justify it if we assume
    that these patterns, per se, affect the ―well-being‖ of the VH. Then normal sensing and effecting
    dynamics, which can be seen in amoebas, even, can take over. But why should patterns as such
    effect the VH? Here we have to assume that at least part of the VH is indeed composed of EDEs. If
    that is true, then it will be affected by pattern, willy-nilly.
296) Let us return to another question which might clarify the one above. Where does the VH come
    from? It does seem unlikely that there happens to be such a stable EDE just arising from neural
    dynamics, however necessary it seems to be. However, if we contemplate the CNS‘s necessity for
    monitoring the body, where that takes place, and what it involves, it seems that this monitoring,
    maintaining, and manipulating would be a good candidate for the origins of the VH. Consider
    that the body is basically a self-contained system, independent, really, as far as structure,
    maintenance, manipulation, and so forth, of one‘s sensory surround: of the external world.
    However a la mode it is these days to speak of the fuzzy boundaries between the body and mind
    or the body and the external world, it is nonetheless true, especially from the neural standpoint,
    that the body might be considered largely independent of sensory (aside from internal sensations
    from the organs) input; and further, that consciousness is not really necessary to this maintenance
    and internal sensation. We are not, after all, really conscious of our heartbeat, the movement of
    our intestines, and so forth, under normal circumstances, as a rule; and that consciousness (or,
    more precisely, ―the mental‖) is certainly not necessary to their function. But this is a radical
    difference between that and the sensory systems, and even more between that and abstract
    thought.
297) Further, a VH arising from the bodily maintenance functions would also involve the emotions;
    certainly, at any rate, the more primitive emotions of pain and pleasure. And again, the mental
    does not seem necessary for those to operate functionally: we see for example that there are
    animals fairly clearly not conscious (i.e., insects) that operate, to a certain extent, as if they felt
    pain, fear, pleasure, and attraction, without, probably, feeling anything at all. So the common
    thread here is the lack of the necessity for the mental for the routine, at least, operations of these
    aspects of the body; and this is just what we want for the VH, for here we need a system that can
    start consciousness (or, at least, the mental) without involving us in a regress.
298) So this gives us a starting point for the consideration of the nature of the organism, the VH, as a
    self-contained system: the system parameters, so to speak. The neural dynamics of the lower
    levels of the CNS (brainstem, cerebellum, RAS (?), periaqueductal gray (?), and so forth), which
    embody in neural code the body, become considered as an organism in itself. Now we must think
    about the following: first, this VH cannot and should not ―see‖ neural impulses as such. Second,
    there must be something like a ―sensory interface‖, where then ambient impulses react against
    some system which translates or transduces them into compatible streams; we might tentatively
    consider the hypothalamus, ostensibly for memory, as such a device.
299) Now the initial insight having to do with consciousness was at 116: ―But the idea of the
    integration of sensory and bodily information has another aspect, that of representation. Suppose the
    processes that were to control and manipulate the body and the external world were redirected internally,
    toward the results of those above sensory abstractive processes, rather than externally toward the
    body. Then those processes would in fact be the bases for internally manipulable representations.
    We have here, I speculate, a reason for the internal direction of consciousness, that is, to create and
    control representations. In addition, this gives us a criterion for consciousness, or at least for
    proto-consciousness. That is, it is not an organism that does not have representations in some
    sense (which can include the abstractive processes) which is not conscious, but an organism that
    cannot manipulate those representations as if they were objects, which cannot be conscious.‖ We
    have performed something like a Hegelian turn here, back to the original with new meaning.
300) But let us look at 298 again. We see that the VH cannot see neural impulses as such, because it is
    itself made of such impulses, and further because the cells on which they are based do not see
    impulses, but process them spatially/temporally into what we term ―codes‖. Thus it must be that
    the processing within the VH determines how it treats the incoming impulses: as analogous codes,
    or preferably, as the spatial/temporal integration functions. Now the internal codes are seen in the
    same terms that the whole system is built on: self-maintenance and stability functions. Yet at the
    same time, those functions are built of the abstractive processes resulting from the sensory, etc.
    inputs and interacting processes. One might speculate on high-level and multimodal integrating
    processes for the purposes of internalizing stability functions.
301) Let us look at 296-7 again. While it is true that there must be bodily maintenance functions, it
    seems more likely, at this point, that the VH is more a duplicate or replication of the cortex than I
    previously thought. For one thing, we see that we can tap into the sensory flow at fairly low
    levels; not from the absolute bottom, but still fairly low. How can we do that, without a parallel
    system of some sort; that is, the VH cannot operate purely top-down. Second, the neural coding
    of the various sensory modalities is virtually identical, on the micro level. That is, neural impulses
    are similarly processed, wherever they originate, as I have noted above. But if this is true, then the
    only way to monitor, say, aural processes is to run a duplicate of the aural cortex. That is, if neural
    impulses fed from that cortex into a general processing area, and were joined by visual, etc.,
    processes, how would they be differentiated, except by being sent to different sub-areas? But this
    is the same as ―duplicating‖ the cortex, in small. This is why, at the highest levels, specific sensory
    information must be lost; i.e., when the abstraction gets high enough so that there is total inter-
    modal integration there must be something like general logical processes. Returning to the
    previous point, we see then that the VH must consist of small areas which correspond, map, the
    medium to high sensory areas in the cortex, without, however, performing all their abstractive
    processes; otherwise the cortex would be exactly duplicated, which is absurd for a variety of
    reasons. So the control and monitoring areas of the VH map the cortex; but they must also
    perform some very high-level abstraction.
302) In addition, we have seen that one of the most basic functional aspects of the CNS is
    conservatism: the maintenance of neural parameters and stability; this is the original function of
    the top-down feedback, and it must also be the original function (integrated with memory
    retrieval) of the VH. So we now have the basis for modeling: the VH duplicates the cortex in order
    to discriminate sensory (and other, perhaps) modalities; and so within that VH we could construct
    small sparse models of possibilities.
303) But before we look at that, we might contemplate the nature of multi-modal feedback/uniting
    processes. That is, in order to form objects, the different modalities within an object must be
    integrated. One speculates that since it is done in insects and reptiles without mind, then this must
    be another low-level set of processes. Nonetheless, we probably find the beginning of nested
    recursive processes here, especially in higher forms. After all, an insect, etc., does not need much
    integration. To put this another way, in order to form a high-level multi-modal object, where there
    are abstractions formed from the inter-modal interactions, we must go beyond the insect level, at
    least. But here the interesting problem is the integration of the different ―codes‖ for the different
    modalities. We might even consider this within a complex modality such as vision, where colors
    must be integrated with shape, for example.
304) The important insight here, I think, is that when different modalities are conjoined in the
    formation of a low-level multimodal neural realization of an object, two things must be true
    (usually). First, one modality is dominant and ―drives‖ the other(s). Second, in order to conjoin
    elements from two or more modalities, the objects must be operated on as they are in reality. That
    is, to conjoin modalities too freely would be incorrect and jarring. Now we may contemplate the
    first condition. If there were two modalities conjoined, say touch and vision, and both were
    equally present and determining, then the vision corresponding to the touch, and vice versa,
    would have to be determined by some third process. Here we have a possible opening for control
    processes. But at a low level, in which much of this multi-modal conjoining takes place (to form
    objects, for example), this is not the case; indeed, these processes are not conscious. So one must
    drive the other, i.e., there must be a dominant modality, for that instance. This of course enables 1)
    lower animals to have such multi-modal conjoinings, 2) makes possible unconscious conjoining,
    and 3) provides a function for conscious manipulation of such conjoinings.

305) Now, what is it to say that ―objects‖ must be ―operated on‖? We must remember that we are
    talking about neural circuits. Let us contemplate the nature of feedback once more. Consider that
    feedback need not be either ―positive‖ or ―negative‖, but may instead, especially in a neural
    network, monitor functioning for particular patterns (i.e., as in mathematical functions), and
    maintain and restore those patterns through the application of other patterns. In other words,
    feedback can be qualitative, not just quantitative, i.e., explicitly qualitative, as when a function is
    modified to produce another function. For a simple example, suppose we have a circuit which
    needs to oscillate sinusoidally, and it begins to output jagged curves. We may have a feedback
    circuit sensitive to deviations from the sinusoid function which then corrects some other function
    back to that original. Here is an example of a feedback which is qualitatively negative, qualitative
    in the sense of looking for deviation from a particular pattern and negative in the sense of
    reversing that deviation. Surely dynamic circuitry as described above (EDEs) could accomplish
    this. Other mechanisms, involving induction of memorized patterns in modules specialized for
    such, where those patterns are not ―retrieved‖ but employed for the modification of the inducing
    (or related) circuitry in the manner indicated here could be employed. Here, all in all, I believe, is
    the basis for qualitative differences. Now suppose that there is another circuit monitoring the
    complex above, which looks for accuracy of the correction. Here is an example of feedback
    recursion, in which one part of a loop monitors another whole loop.
306) Now consider the above discourse, in which I assume that one of the bases of consciousness is
    the internal modeling or recreation of the external world. What of the function of qualitative
    feedback here? Surely, in order to model unrealized external events, i.e., to plan and anticipate,
    the higher-level qualitative stability monitoring must be modified, in order to create new patterns,
    but according to simple and/or systematic variation of the old. Here perhaps some simple
    variations of the monitoring processes would work, and some limited degree of qualitatively
    positive feedback.

307) Now again we take another tack. The recent article in Nature emphasized the fact that the CNS
    does not use coding in any conventional sense. That is, to coordinate the visual and aural systems
    of the owl, there is point-to-point correction and modification of the aural spatial map by the
    visual map. Must this be contrasted with the fact that consciousness is integrative, i.e., that we see
    the scene as a whole? No, that is not necessary, at least at that level.
308) There are two interesting things about consciousness. First, it is interesting and puzzling that we
    see what is there, at the lowest level of the CNS processing. Why should this be, except for the
    obvious evolutionary necessity? Why, from a physiological point of view? Second, there is a sense
    in which that last comment is not correct. We see what is there, synthesized over time. That is, for
    the aural input, for reasonable frequencies, we hear the time-integration of those frequencies. But
    the interesting thing here is that this is also true for the visual system. That is, the results of the
    Mack et al experiments and the experiments showing that we see at any instant only a very small
    fraction of the visual field, demonstrate that the same processes are going on in the visual system
    as are going on in the aural, and no doubt in other, systems. There is a system which is very much
    slower than the rest of the processing in the CNS, which is synthesizing, over time, inputs to that
    system.
309) This indicates several things. First, since this system works differently from the rest of the CNS,
    since that latter processing and abstraction is very fast (and indeed has to be), it is a separate
    system (at the very least functionally separate). Second, that separate system does not work with
    the same kind of neural abstractive processes, based on single-neural integration (i.e., neural
    convergence), that seems to be the case in the rest of the CNS. Third, that system cannot then be
    merely a part of the cortex, sitting on top of the abstractive hierarchy. Fourth, it cannot be merely
    something like parietal integration. Thus it must be some sort of overall integration. Yet that has
    the same integrative problem, since merely an extended neural network does not demonstrate
    experienced integration, however it is time-synchronized. This leads even more directly, then,
    than the above arguments to the conclusion that some sort of EDE is involved in consciousness.
    That is, that the extended network must be united through quick enough large-scale feedback
    processes (described far above), which however must slow down the processing because of the
    size of the network over which they function. This does indeed give us time integration, but how
    does it give us the consciousness of low-level processes, i.e., the individual frequencies responded
    to; the individual lines and angles in the visual system? It is interesting to note how the
    phenomenologists intuited this fundamental temporal integration (i.e., Husserl‘s ―running-off‖,
    etc.).
310) But before we address that latter question, let us look at the EDEs, not as potentially ―mental‖,
    whatever that means, but as what, ultimately, must ―stand for‖ (again, whatever that means), or
    realize, objects, in at the very least a functional sense. If EDEs, these large loops, are objects, then
    they must take inputs from the cortical areas abstracting to those objects, and in addition from
    memory, etc. Also, let us assume that these loops are, being so large, unstable. What then is
    necessary to stabilize them? Surely this must be input from the RAS, as focusing, and memory, as
    detail stability. That is, the RAS might function on a large scale just as the excitatory/inhibitory
    fields of individual neurons work, in order to create and maintain bounded sets feeding into the
    EDEs. We see then that these supra-cortical objects take multiple inputs, stabilize them, and that
    we use them as objects. But what does that mean, exactly? That is, how do we manipulate them as
    objects? Obviously, they are neural dynamics, not actual objects that correspond in properties to
    what they are realizing. How then do we manipulate them as the objects they realize, except to
    use them to downward induce cortical firing (and induce memory)? So as memory, i.e.,
    hypothalamic, etc., firing and RAS firing stabilizes these loops – against resistance, since they are
    unstable – so must we in addition destabilize them in order to combine and alter them. Now that
    combination, at least, must take place downward from the loops being combined, as they induce
    cortical firing; and then it must move upward into the loops, and combine them in some manner
    to be realized as a new object. We see then that ―manipulation‖ of mental (to jump for a moment)
    objects does not entail the actual manipulation even of EDEs, but instead variation, through their
    destabilization, of their downward cortical induction, and the subsequent upward restabilization
    of the loop. The next problem then is how this controlled destabilization takes place.

311) It is time to start on the paradigm of mapping correspondences. Not only does the note 307,
    above, indicate that this is indeed used in the CNS, but we must consider the problems with what
    has been termed the ―signal detector‖ paradigm (SDP), exemplified by the above long analysis.
    Here we have found that this works extremely well in analyzing one type of input and forming
    abstractions from it. But how can it coordinate between different types of analyses of the same
    input, and between different modalities of input, if all the functioning of the CNS is in the
    refinement of the detectors, of the filtering and abstracting, in effect? This is also the problem with
    Grossberg‘s analyses. I am proposing that what goes on, in addition to the detection and
    abstraction formation above, is the initialization, maintenance, and alteration of mappings
    between different areas of the CNS, and even within areas. And indeed that this mapping
    alteration is the basis of internal modeling. Consider that there is no ―information‖, per se, in the
    CNS, as I have said above, comparable to the information in a digital computer, and that what
    does go on is analog manipulations. The disadvantage of analog manipulations is that they are
    modality or function-specific. An analog device will only perform manipulations within a
    particular structural coding correspondence, and cannot employ universal symbols to switch their
    meanings between modalities, as computers can. Yet we are just analog neural nets. Thus we have
    this inflexibility, which isolates processes in this manner. But this isolation, while functional in the
    abstractive processes, is not universal in the CNS, or we would have no inter-modality
    combinations. How then can various analog signal detectors, filters, and abstractor, meaningfully
    interact, when they are performing those processes simultaneously on different types of signals?
    We must also consider that the various areas are by and large retinotopic maps, of various sorts,
    given the above processing.
312) What if we had a ―retina‖, in effect, which was an internal processing surface which took the
    inputs of several areas as its sensations, and combined them analogously to sensed combinations
    on the retina? This is a straightforward extension of the SDP, but it is also a mapping
    correspondence paradigm, in effect. Now, instead of thinking of how such signal combinations
    can be abstracted through such processes, can an equivalent analysis be done in terms of mapping
    correspondences being formed and altered? To put it another way, the SDP lends itself to a
    treatment as a passive set of processes, where active alteration of the SD functions is not the norm.
    Can we combine SD and a kind of switchboard picture?
313) Yes. Given the survey of binding by Humphries (2001), we must conclude that mapping
    correspondences are indeed fundamental aspects of the visual and thus of other systems. Oddly
    enough, he does not speak in those terms; instead using the term ―binding‖, which I find both an
    odd metaphor and indeed another one implying passivity. Thus on this picture, the SD and
    abstractive processes end within modules, in effect, and what takes over, even at very low levels,
    are the processes creating, modifying, and stabilizing mapping correspondences. But in this
    picture, we do not have the above extension of the SD paradigm (in 312). Objects are formed by
    the creation of correspondences between the maps created in turn by the SD processes, and not by
    a ―retina‖ which synthesizes signals into new patterns.

314) Aside from the transient resonances envisioned by Grossberg (and myself, with EDEs and top-
    down stabilization; and now by Crick & Koch in their recent article) there must be a large,
    continuously going resonance which tracks and stabilizes the small resonances (or EDEs). This
    must be the homunculus-equivalent. In other words, it must be not only that the small local EDEs
    participate in larger groups, but also that there has to be one very large, top-down, EDE which
    tracks, stabilizes, and relates all the small ones, on a continuous basis.

315) But here is the solution. That is, what we are, and do, is literally manipulate analog models of
    the world. The brain instantiates, to various degrees of abstraction, the world, concepts, etc. The
    brain manipulates, to various degrees of abstraction, those instantiations just as one part of an
    analog device manipulates another part. This approach, where there is a literal model of the world
    in our heads, yes, a literal model, is the only one that can solve, and integrate, all the above and
    more. All the mappings, then, are relevant to the unification of these models, and to their
    manipulation. The brain does not transfer information, it reaches out, from module to module, to
    make changes. Look at a low level, where the cortex instantiates space. To move ―something‖
    through space, we can cause the corresponding neural firings to shift across the cortex. But what is
    that but literally pushing something through space? The neural firings are that thing, or at least
    that aspect of that model which is space-related. And that‘s where the mapping coordination
    comes in. Time is easy to model; we instantiate that fairly directly. Look at the homunculus: there
    is one in the cortex and parietal lobe; that is, I will bet, the instigator of literal motions of the
    model of the world, as those motions flow through the space in the cortex. And there has to be at
    least one more in the cerebellum.
316) There is another wrinkle to the last paragraph, however. The EDEs, or Grossberg‘s ARTs, or
    whatever, are not sufficient as they stand. They have to be varied, within limits, by the RAS. The
    stability of those patterns is what the CNS is after, as I have said, but it is not enough for
    consciousness. The ERTAS activation has to ―jiggle‖ them, produce a bounded variability, so that
    we have volitional choices, different meanings, and information extraction through the variations
    in the attractor basin. Freemen thus seems more correct than I thought… it seems that in order to
    have consciousness, we have to realize a good deal of not merely the EDE but the attractor basin
    which contains the *possible* EDEs for that set. One EDE or ART resonance, or Leharian ―bubble‖
    is not sufficient for consciousness. Given that ―jiggle‖, there are many problems that are resolved.
    Volitional possibilities, both as vague feelings and as clear intuitions; the ―gap‖ in the TOT
    phenomenon; creative variability and feelings; boredom and the *loss* of the consciousness in
    monotonous situations; perhaps hypnotic states; and other phenomena. Thus, the objects we
    experience are not merely the neural objects described above, but those objects when they are
    ―activated‖, i.e., varied within some boundary, or jiggled by the ERTAS. We might look at some
    disorders of consciousness as disorders of this induced variability also.

317) I keep coming back to the necessity for representation, in effect. That is, consider play, or
    creativity, or flexibility in general. If the above neural percepts are handled just as they present
    themselves, then we have no need for consciousness. However, if we must alter their functionality
    from their normal presentations, as in play, symbolism, or creation, then we need additional
    feedback loops controlled by top-down concepts of some sort. In other words, all alterations of
    this sort are at least proto-representations. Play is the beginning of symbolism, because the play
    object stands for something else. So even a kitten playing thus has the basis for language… and
    also for consciousness. No play, no consciousness. So the question is, where does that come from?
318) The next point, about the homunculus. The classical separation between the homunculus and
    the world, leading to the infinite regress, is not necessary in the cortex, because the thoughts of the
    homunculus literally are its objects: its world, its sensory surround. This must be the case because
    of the above arguments about the intersection of bottom-up and top-down representations. That
    being the case, there is no separation between the world and thoughts for the homunculus, and
    thus no gap and no regress. That is, the homunculus literally is these sensory and cognitive
    thoughts, realized as EDEs, as well as the EDEs for its body. This leads to a strange situation,
    however, since we do experience this separation, and further, we cannot freely manipulate our
    thoughts. What we must conclude, I believe, is that first, we are not aware, obviously, that we are
    the homunculus and are ―seeing‖ EDEs. Second, it must be that although those are both the
    thoughts and the environment of the homunculus, because they are bottom-up driven as well as
    top-down, there are times – probably many times, because of the strength of the bottom-up
    impulses - that the homunculus‘ control is overridden. Why not? So although its thoughts – its
    body - are in part its world, they are still not completely in its control. An interesting paradox, in a
    way… here we have the beginning of the ―gap‖, which is not a gap.
319) Going back to 317, then, we must consider that even this homunculus, interacting with one or
    two others, as I have said above, will not produce consciousness unless it is able to alter the
    relationships presented to it in its thoughts, i.e., in its cognitions and sensations. Consider that
    ―representation‖ for a net of, say, three homunculi, must have to do with the alteration of function
    of an EDE presented to, say, the cerebellar homunculus by the cortical homunculus. What then
    would happen? First, what does ―presented to‖ mean? Remember the mapping paradigm above.
    It must be that a cortical EDE is effectively a mapping location for other parts of the CNS. Thus
    this EDE is a map evoker, a locator, for a corresponding EDE in the cerebellar map (and of course
    this is a temporal dynamic, so the EDE moves, and so must the corresponding one in the
    cerebellum, in this example). So what must be happening is that the cortical homunculus is
    altering that dynamic mapping relationship for that EDE.