BEING ABOUT Chapter 2. Wide nets
Cortex and aboutnessClassical dualist mentalism has thought of aboutness as a property of ideas (or some such) in immaterial minds. The monist contemporary variant has aboutness localized to physical structures stored and processed in cerebral cortex, the most recently evolved brain structure. The wrongness of representational language, applied to mysterious somethings in immaterial minds or to structures in material cortex, is that the locution sets us up to speak as if the sentient creature is about the representational structure rather than something in the world. In the last chapter I described aboutness as relational. I said that the whole organism is structurally about its environment; that it is about many aspects of that environment at the same time; and that structural aboutness is causally complex, its structural means having been created at many times, both in the life of the individual and in evolutionary time. If we say aboutness is a whole organism's structural adaptedness built in contact with an environment, and if we say nervous systems are facilitative to this contact, and if we say that even the most recent, most exclusively human additions will have to have evolved to promote effective sensory-motor through-function, it makes dubious sense to say a cortical structure is about something in particular. And yet, having said that cortical neural form should not be thought to be about anything, especially not in the mediational way a representational object is about something, there are nonetheless reasons to consider the particular form of the central neural means. The reasons are so compelling that they have made the representationalist locution a persistent temptation.
When we dream, bodily response to the world around the sleeper is reduced; both sensors and effectors are largely shut down. And yet the dreamer is, in some very reduced way, as if about many things. What about that partial, as-if aboutness? Specifically, how should we talk about its central means, which are, more or less, at least in the moment, the whole means? Their particular form will certainly determine what the dreamer is seeming to be about. How is that decoupled cortical self-organization different from 'inner representation'? What goes for dreaming goes for the many kinds of simulation the evolved cerebral cortex enables. We are eminently able to be as if about things we are not, at the moment, in contact with. We need not take simulational aboutness as paradigmatic, but we do have to account for it. And there is a related, very awkward, question about the difference between non-sentient aboutness and sentient engagement, which we guess, for good reasons, must have something to do with differences in the particular form of the cerebral means. Can we be interested in central neural form without thinking of it dualistically, or as a representation? There is an undeniable tension between two understandings of aboutness, the version that sees it as a relation, and so must imagine organism and environment at the same time, and the understanding that central structures are critical to the organism's experience of being so related. The solution of this difficulty is partly a matter of linguistic care. We can say there is a structure that is an essential part of sentient engagement without talking as if it is about things in the world -- and without talking as if it is what we are about. We can for instance say it this way: together with the world and the rest of the body, a particular cortical form is (an essential part of) the means by which I am or seem to be about something at this moment. The challenge in what follows is to find ways to imagine the specifically cortical means of aboutness in a manner that is compatible both with its importance to the way knowing is done in developed creatures with highly developed central structure, and with the whole-body aboutness I have described. Findings and hypotheses about cortical structure that I will describe below can support revisions to our theory of knowledge that will make it more relevant to what knowing is like. Imagining the brain... a fairly simple point about causation: that it is multiple, interdependent and complex. Oyama 1993,25 We have not known enough about the brain to be able to imagine it well, and our difficulties remain even as we learn more. We are not good at imagining granular complexity, but in a cortex there are billions of relevant elements, and many of them are active at the same time. In trying to think cognitive function in terms of brain structure we also need to be aware of the difficulties of thinking about structure at all. There is a figure-ground problem. We generally imagine 'a structure' as a simple, stable figure, a relation of articulated parts against an inactive background. When we are imagining structures, as when we are seeing them, we understand one structure by ignoring others. But in the brain there is properly speaking no background, and structures are neither simple nor stable.
We have had various kinds of anatomical illustration; microphotographs of stained neural tissue at the scale of the synapse, the cell, the cell group; cross-sections of layers and bundles; slices of the entire brain; labeled anatomical line drawings; circuit diagrams of physical connections discovered or inferred between regions. These kinds of illustrations do not help us to go very far into imagining what happens as a brain is active. When we are trying to imagine a brain in action, we can try to imagine stable anatomy as background and transient physiology as foreground structure; or we can try thinking of functional changes as moving objects ('impulses', 'signals', and the like) channelled through anatomical structures. Both are inaccurate. Cortical function is also cortical reconstruction. A cortex functions by settling into one structure rather than another. It settles into one structure rather than another by being active, by functioning. We could concentrate on differences between relatively stable structures (which nonetheless are constantly being built and rebuilt) and the transient changes of structure that are, for instance, the synapse's operational changes. But even this distinction is hard to maintain, since structural alterations in the course of momentary function can result in permanent structural changes. These structural changes belong to physiology in the sense that they are functional alterations, changes that may, for instance, constitute occurrent perception. But these functional changes may also participate in long-term structural change. For example, activity propagating through existing connections propagates by means of synaptic change. These functional changes may reconstruct the synapse, so that what happens at a synapse at some very particular later time will be different because of a constellation of conditions existing at this moment. Reconstruction of this sort is one of the physical bases of memory and skill. Cortical self-construction is an instance of the distributed, connected, multiply integrational complexity of biological effects and causes. The structure/function difficulty we experience in describing cortical events is an instance of a more general difficulty we have with causal complexity. Biological causation in general requires new sorts of systems thinking, because biological entities are both self-referring -- constructing and maintaining themselves -- and massively interactive with a larger, embedding system. What happens in biological self-ordering fields generally varies simultaneously with both internal and external facts, and in ways that may be hard to separate conceptually. Understanding complex systems requires that we think in wholes at the same time as we are thinking of parts. This principle applies at different scales. If we think of organisms as parts of environments, the dynamic coupling of organism and environment is not forgotten even when we are concentrating on internal facts. If we think of individuals as parts of social groups, the social management of individual cognition does not drop out of our theory of mind. Similarly, if we think of nervous systems as parts of whole bodies, and if we remember whole evolutionary and developmental trajectories, we are less likely to think of brains as separate from the sense and muscle systems they facilitate. And if we think of the moment's cognitive structure as part of an organ doing many things at once, we are less likely to imagine bits of aboutness being processed in modules. A biological systems approach also says, don't think of individual parts as created first and then assembled into wholes. Think of co-evolution, the evolutionary emergence of wholes whose parts are mutually necessary to one another even as they are developing. Functional differentiation of cortical parts is, for instance, largely the result of activity differences that are themselves a result of differences of position relative to sensory and motor in-and out-connections. (For more on biological systems see Rosen 1985, 1991; Wilden 1980; Oyama 1985, 1993; Jarvilheto 1998a, 1998b, 1999, 2000; Kauffman 1995; van Gelder and Port 1995.)
The brain (like an anthill) is a volume active all over. The space that can be thought of as background to activity propagated through it, is itself minutely ordered and active. Its structure changes in response to propagated changes, and it actively influences that propagation. There is really no inactive space in the brain. We have to think of certain processes as background only in the sense that we must fail to imagine them while we imagine something else. Change propagates through the brain not only by spikes zipping down axons, but also by slower processes occurring outside the myelinated axons (Bullock 1993; Bach-y-Rita 1994). The neuropile is made up of dendritic arbors, axons, and support cells in a bath of extracellular fluid that itself is electrochemically active. Electrical currents generated in all the cells in an area cross their cell membranes into shared extracellular space and are integrated as electrical fields. Neuromodulating peptides released into extracellular fluid as a result of activity in brain-stem-based circuits are taken up also at non-synaptic binding sites on neural cell surfaces, where they alter cell characteristics. The chemical presence of these neuromodulators in extracellular fluid also has field effects. Electrical and chemical fields in the neuropile are thought to operate over many micrometers (Bullock 1993, 8). It is this sort of integrated field effect that is measured by electroencephalographic and magnetoencephalographic probes, which discover slowly travelling and standing waves observable at various frequencies. Slow wave electrical and chemical fields interact with spike propagation in myelinated channels by altering synaptic and trigger zone sensitivities. When we think in terms of signal transmission we are thinking of high levels of activity sent unchanged down chains of axons, but neural response is actually being constructed at each point in cooperation with existing structure at that point. We would have to imagine it as forming the way a cloud forms, by a sort of on-the-spot condensation of brightness around invisible points in an existing ground.
Neural connections are not pipes. They do not transmit some fluid, or some symbol, or some packet of 'information'. They are physical structures whose selective response helps to put an organism in touch with its environment. There is actual movement of materials in a brain, notably blood flow, cell migration, and at a smaller scale, molecular transport to binding sites on cell membranes. But most of what we imagine as movement in a brain is instead a propagation of effect, where changes in one unit or region set up change in adjacent units or regions. It is the structural changes by which a brain is functioning that should be thought of as propagating within a volume. Both volume transmission and spike propagation occur by means of local changes that are integrated results of many other local changes. What is happening at any particular synapse, trigger zone, or extracellular site will be determined as a function of its position in the field, and will participate in determining what happens at other sites. There is continuous alteration of field structure as a result of changes which have been initiated at many points. When patterns of activity are propagated iteratively, through recurrent connections, the resulting changes in a brain can also be seen as three-dimensional standing structures -- transient three-dimensional configurations sustained over time. Some of these standing structures may include nested substructures whose cycles are longer or shorter.
Activity propagated from sensory origins usually is described as passing through hierarchized stages on its way up into and through the cortex, but this way of thinking is more suitable to a conception of neural processes as delivering a product to an end-point than it is to an understanding that what happens at any stage goes on having effects at many points simultaneously. Notions of cortical hierarchy are used in several ways. One is a social metaphor implying a pecking order in cognitive abilities, with sensory processes, the lower orders, described as operating bottom-up, and 'higher functions' described as operating top-down. A second, related way of evoking hierarchy is in descriptions of the causal relations of parts and wholes, where parts are thought to be atoms 'extracted' 'at a lower level' and then assembled into 'higher level' wholes. Phonemes would for instance be thought of as extracted first, and then assembled into words, which would be assembled into sentences. But the evidence is that phonemes, words and sentences are being perceived concurrently, and that structures active in perceiving parts are interconnected both with each other and with parts active in understanding wholes. It is probably true that different logical levels of discriminations are being made at different foci, but reentrant connection allows 'higher-level' foci to make continuous reference to what is happening at foci earlier in the path, while also allowing the effects of convergences later in the path to modify what is happening at these earlier foci. And there seems to be no one sequence followed by any train of activity. Instead there is great redundancy, many routes, all with backward and forward connections integrating effects across temporal and spatial differences. So biological systems can be understood at different logical levels, and they function in relation to what can be thought of as different logical levels, but they are not in any evident way hierarchically organized.
The prevailing paradigm of neural operation is circuits and spikes. ... The sometime analogy to digital computers is well discredited, there being no evidence of the time-slicing required for a digital device. Bullock 1993,8 It has been tempting to avoid imagining the brain altogether and imagine instead familiar sorts of electronic communicational and computational mechanism, elaborating a metaphor based on notions of information processing, signal and code. The nervous system's changing acts of spatial structure may be thought of as computation if we think of computation clearly enough. Mechanical computation is systematic consequential connection. In a nervous system systematic consequential connection has to do with coordinating an organism both internally and externally. Complex spatiotemporal coordination is what mechanics and organics have in common; it is what supports the usefulness of a computational metaphor in the first place. But there are important differences between mechanisms and organisms. Mechanisms are artifacts; organic structure is marked by its very different origin. Cortical evolution and developmentLike the rest of the body, brains are constructed as the result of events occurring in species time, in somatic time, and in immediacy. Brains evolve, they develop, and they restructure themselves in the moment of function. In its over-all organization, a brain is more like a plant than like a computer; its organs have shapes and internal textures, the way roots and petals have shapes and textures. Like plant forms, the layout of cerebral organs and regions, the differences among cell types, the folding of cortical surface, the reach of axons over small and very great distances, make sense as results of an evolutionary and developmental history. The large-scale organization of the cortex shows an evolutionary and developmental logic that follows three very evident principles. The first is the ongoing primacy of sensor-effector through-function, as described earlier for nervous systems as wholes. Whatever else happens in the cortex, it must be able to support the organism's ability to respond effectively -- to stay in touch with where it is. Everything else has to be built around that exigency. The second and third principles are increasing integration and increasing immediacy. Increased integration occurs by cross connections between through-lines, beginning with contact between nets serving different sensor-effector pairs, and proceeding to ganglia, to notochord, to brainstem, to midbrain, and then, as described below, to cortex and neocortex, retaining earlier through-lines while adding contacts that include more sensor systems, more motor systems, and more systems having to do with internal self regulation. A second aspect of integration occurs with synchronization of activity by means of these connections. Synchronization also increases throughout mammalian evolution; there is wide variance, but the tendency for spatially defined populations of cells to be coherent increases in the order of taxa, and in "more advanced levels, stages and states" of a given brain (Bullock 1993, 3). Increasing immediacy comes about by elaboration of sensory and motor systems and the resulting large increases in the number of central connections included in these systems. It also results from increasing plasticity; where the cortex is more able to reconstruct itself as circumstances change, the organism will be more finely tuned to the particularity of its present time and place.
In Chapter 2 I described the evolution of nervous systems as demonstrating selection for increasingly complex contextualization. The evolution of the cerebral cortex can be seen as the most recent and most effective means of this increase. Jerison associates the development of cortex with alterations in distance perception: "The early reptilian visual system, including its neural components, was packaged to a major extent in the retina, and the mammalian auditory system, to expand sufficiently to perform distance sensing, had to be packaged in the brain..." (1973, 20). In early mammals, central structures (the thalamus and striatum) develop to relay activity from many sensors to many effectors through regions allowing more cross-talk. As vision and audition become more important than smell, neocortex riding over earlier structures begins to integrate developing sensory modalities with each other. [2-3 Illustration of neopallium ] In a primitive mammal such as the jumping shrew, the neopallium or neocortex is mostly sensory -- auditory, visual and tactile -- with an anterior part of the tactile area specialized for motor functions. In mammals that occasionally use their forepaws to handle objects, bears for instance, the relative placement of sensory areas remains constant, but they begin to be pushed apart by an in-fill of higher order sensory areas and multi-sensory association cortex (Critchley 1953). In the primates -- and to a much greater extent in the Old World monkeys in our evolutionary branch -- burgeoning association areas alter the layout of cerebral cortex so there is more folding and it begins to have the look of the human cortex. Cortical function in humans goes on being anchored in primary sensory and motor areas which exist also in primitive mammals. [2-4 Lateral view of human primary, secondary and tertiary cortex] These areas are significantly expanded in humans, and shoved apart by associative areas built between them, but they retain much of the character of these structures in earlier animals. Primary sensory and motor areas develop relatively early in embryogenesis, they are relatively mature at birth (dendritic and synaptic structures are organized and axons are myelinated), and their concentration of opiate binding sites does not increase after birth. They show little convolutional variation across individuals. Association areas interpolated between primary sensory and motor areas throughout mammalian evolution differ from primary areas in all these ways. Association areas in the primate brain develop later in embryogenesis, their connections are organized later, and their axons are myelinated later. Areas with delayed maturation of dendritic and synaptic organization also have delayed organization of opiate binding sites, and, in the period after birth, develop these at higher concentrations. In infants, there is more post-birth pruning of exuberant connections already present in these areas. Structure in association cortex is and remains more plastic. In young animals, lesions to these areas will not necessarily impair function. Presumably as a result of this plasticity, association areas show more convolutional and functional variation across individuals. The large increase of association area in primates occurs in the parietal lobe, the temporal lobe and the forebrain. Association tissue in the parietal lobe integrates vision and touch, so it makes sense that larger associative area in the parietal accompanies skill in handling small items. New temporal lobe tissue that supports foveal color and form vision (among other things) also supports new abilities to make social use of faces. These new abilities are examples of an increasing immediacy given by the integration and elaboration of sensor and effector systems, and by an increase of plasticity in associative cortex. The most significant innovation in human cortex seems to be the elaboration of new areas between existing association areas: instances are the inferior parietal lobule at the junction of visual, auditory and somatosensory cortex, and a related area in the forebrain. [2-5 Newest cortical areas shown in white] These newest, homotypic, associative areas are, in the individual, the last to develop, last to organize connections, last to myelinate, absolutely the most variable part of the cortex, and by far the most plastic in structure. I will suggest later that, with this latest addition of areas that associate association areas, there is an increase in ability to be unimmediate, in other words to simulate.
It is helpful to be able to imagine the development of neocortex in the individual for this reason: if it is traced from its beginning, and if it is understood without dualist metaphor, it is apparent that the development of neocortex proceeds in contact with the rest of the nervous system, that the development of the nervous system proceeds in contact with the rest of the organism, and that the development of the organism proceeds in contact with local parts of the larger world (Oyama 1993). To understand neocortical development is to understand it as many-ways embedded -- a central, but not an isolated, participant in organism aboutness. Before I go on to describe neocortex development, there are a few things to be said about development in general. Species form evolves by natural selection; the soma, or individual body, is constructed by processes that can be described as somatic selection. The mechanisms of somatic selection -- pre- and post-natal development -- are themselves part of species form, selected because they increased fitness (Edelman 1988). The construction of the individual of a species occurs by an interaction of genetic structure and world structure: in the beginning, there is a gene string and a world. The gene string can be thought of as a sort of précis of the evolved structural aboutness that is species form, but only in the sense that, given the continuing presence of an adequate environment, species form will result from the ongoing interaction of the two. This interaction has been thought of as divided into importantly different stages: differentiation and morphogenesis in the embryo; maturation and learning by experience or by instruction in the postnatal individual. Development and maturation have generally been said to be genetically controlled, and learning has often been thought to be environmentally controlled. The distinction between genetic and environmental control -- whether it is phrased in terms of a contrast between nature and nurture, or between prenatal and postnatal life -- has been generally unhelpful. First, the processes by which an individual body builds itself continue until death; development and learning are often accomplished by the same physical mechanisms. Second, the social metaphor of control obscures the interactive nature of the processes described. Although there are differences in the sort of interactivity that matter most during prenatal and postnatal life, both are interactive mechanisms from the first. Cortex is built by cell proliferation and migration, by extension of dendrites and axons to make connections with other neurons, and, at a finer scale, by the growth of synapses at points of contact. Cortical structure is refined by the death of cells, retraction of connections, and alterations of synaptic structure. During morphogenesis, which is to say, during the whole of an individual's lifetime, all of these processes are occurring at the same time. They occur always in the context of an organism already placed somewhere in the world and already partly formed. The cortex organizes itself into areas and subareas on the basis of its main extrinsic connections -- in-connections from sensory organs and out-connections to muscles and glands -- and on the basis of its important intrinsic connections to the central structures that monitor internal organs. In early development, the final position and the character of a neuron is place and gene dependent. The connections it forms, and the structure of the synapses on those connections, are determined also by the sorts of activity in its locale: by the electrical activity in its neighbours, and by chemicals diffused into intercellular fluid. When a primate neocortex is being built during embryogenesis, neurons arrange themselves into columns and layers whose architecture is at first quite similar regardless of position. Structural and functional differences which develop in these largely homogeneous neural tissues are the result of connective differences, which are set up during neurogenesis as a result of areal interactivity. Quite early in development some of this neighbourhood activity is being propagated from sensor sheets as they are being formed, and from somatosensory structures in viscera and in muscles and joints under construction. The whole of the nervous system is, after all, forming at the same time. In humans, when connective differences become apparent after about 110 days gestation (Zeki 1993, 180), the cortex has also begun to form extrinsic connections toward and from developing muscles. Neural connections begin to shape the cortex into functioning areas when the organism is very incomplete but not inactive. Motor areas are said to develop first, along with proprioceptive and skin sensation (Llinas 1987, 348), the senses most obviously relevant to the development of motor skills. A human embryo is sucking its thumb at 18 weeks gestation. Vestibular, auditory and visual connections follow in that order -- the order that makes sense for a mammal mobile in early gestation, exposed to abundant though muffled acoustic pressure waves, and only very minimally exposed to light. The organism already responding coherently must also be able to go on responding coherently, not only as conditions change, but also as its own structure changes. This may require that connections already formed be altered. It may also require that connections used to organize part of a growth sequence be pruned later -- as for instance thalamic connections between visual and auditory streams that are found in human babies.
According to the theory, development never stops. Edelman 1988, 193 Bodies adapt. The lung capacity of the high altitude dweller, the high right shoulder of the briefcase carrier, are reconstructions on the basis of an evolved ability to reconstruct. They are ways bodies come to be about their circumstances by altering in response to them. During phylogeny "the brain is the most variable of complex organ systems", Edelman says, and in the higher vertebrates the telencephalon, which gives rise to the cerebral cortex, the most variable of brain regions (1988, 176). Developmental scheduling seems to set up a default macrostructure of brain regions, overall connectivity along with some connective microstructure. But an individual's way of life, their environment and activity, will customize this structure. And moment-by-moment cognitive function is itself an interactive reordering of cortical microstructure. In mammals this interaction begins before birth. Although they are somewhat isolated in their self-constructed developmental subniches, mammals do not develop outside the world. They form in the presence of gravity, amniotic buoyancy, acoustic disturbances, maternal chemistry conveyed through the placenta, and, above all, skin contact with the motion of a mature organism already acting capably in the world. They also form in the presence of systematic covariation among these facts. Species differ in which, and how much, of their structure requires environmental interaction for completion (Wall 1990, 142). Where a newborn lamb must be ready to follow a flock hours after birth, it is a year before human babies are able to walk. What happens in this year is coordination of phyletic scheduling and structure resulting from interaction. It is not that structure given by nature in earlier mammals is now being supplied by the nurture of environmental contact: it is that phenotypic structuring processes formerly isolated from active contact are now being codetermined in active contact, as they have been selected to do. There has been an addition of another aspect of nature, which is nurture. After birth, the cortex goes on being built as well as refined. Morphogenesis by cell division and migration continues vigorously in the first three months after birth. It has recently been found that the total number of neurons in the infant's cortex increases by a third in this period (Blakeslee 2000). From birth to adolescence, in the period of maturation when the child is most intensely interactive, billions of cells are added. New cells are created even into old age, some to compensate for cell losses but others in response to new circumstances (Eriksson et al 1998). Functional structure...the complex levels of the brain, in its manifold integrative tiers Bullock 1993, 2 I've said that, in species time, cortical construction follows three imperatives: keep sensor-effector through-function; add integration by adding connectivity and synchrony; and add immediacy by adding differentiation and plasticity. These imperatives are played out in somatic time -- in prenatal development and postnatal learning and response -- in the organization and function of basic cortical structures: streams, loops, matrices and wide nets. Important sources for the following descriptions of cortical structure and function are Edelman 1987, Ungerleider 1995 and 1998, Damasio and Damasio 1994, Fuster 1995, Gazzaniga ed 1995, Kandel ed 1991, Hebb 1949, Mesulam 1981 and 1990, Nunez 1999, Nauta and Feirtag 1990, and Tononi, Sporns and Edelman 1992.
Basic through-function -- propagation of activity through the cortex -- can be thought of as streams: streams that branch and braid, and may split into substreams with different valencies of effect, inhibitory or excitatory. Streams exist at many scales: in fine lateral connections, in intra-areal connections, in inter-areal connections, and between hemispheres. Streams at many scales also form loops. Loops may occur horizontally, forming groups across parallel streams; they may be intra-areal; or at an even larger scale they may be inter-areal. They may traverse many levels, as, for instance, the very deep thalamocortical loops from midbrain nuclei through primary and secondary sensory cortex, and on through association cortex to frontal areas and then back down to the thalamus.
Sensors are often organized in sheets at or near the surface of the body (retina, cochlea, skin), and through-routes from these sensory sheets project in parallel, through the midbrain to sensory receiving areas in the cortex. This parallel organization may be found some distance into higher sensory and association areas, but it must eventually be replaced by another kind of order, because it must terminate at motor ensembles arranged, also in sheets, but according to act possibilities and not sensory possibilities. Where streams join or cross, matrices form. Matrices contextualize in the sense that they convolve activity propagated from different sources. Matrices have been thought of as maps because response properties of individual neurons are found to vary systematically with the neuron's position in a region. But map is a representational concept, and matrix structures are better thought of as active nodes, altering and propagating patterns of activity. Matrices may vary systematically in one functional dimension, in two, or in more than two. A matrix gains dimensions by integrating activity from more sources. An element in a matrix may be thought of as multifunctional in the sense that it is integrative, but it may also be thought of as multifunctional because its response decision makes a difference to the response decisions of elements in many areas with downstream connections. Axes of variability of matrices may be orthogonal or nonorthogonal. Their organization may be irregular in various ways: there may be functional bulges by which one position has more cells, or cells that are more sharply tuned, or cells with a lower firing threshold. Matrix regions are set up in developmental stages. There is usually a first stage in which axons extending into an area from different directions form a coarse organization of contacts. The matrix is refined at a second stage, when the area has begun to receive activation through these contacts (Constantine-Paton and Law 1990). As a result of these processes, the cytoarchitecture of a matrix -- the look and arrangement of its cells, columns, bands, layers and axonal interconnections -- may differ from that in surrounding areas. It may have distinctive lines of in- or out-connections to other regions. Matrices are joined into widespread networks integrated by recurrent loops at their many scales. There is a reciprocal relation between matrix properties, cell properties and network properties. A cell's structure is altered by activity in its matrix. Matrix structure is a function of the networks that include it as well as of cell structure. And network activity depends on the operating characteristics of cells in matrices as well as on the streams and loops that connect matrices. Cortical cells, matrices and networks are all ways of functioning by restructuring, and of differentiating by integrating.
Bullock has estimated that there are tens of millions of kinds of neurons in the human cortex (1993, 8). Neurons differ in their shapes, and in the length and position and kinds of their dendrites and axons. But even cells of similar appearance can have dynamical differences in the ways they integrate the changes initiated by their contacts with other cells. A neural cell in the cortex typically will be contacted by thousands of other cells, usually through synapses on dendritic arbors upstream from the cell body. Contact in this case means that neurotransmitters released into the synaptic juncture have the effect of polarizing or depolarizing -- exciting or inhibiting -- the resting potential of the cell membrane at the point of contact. A neural cell's gain is its overall reactivity, which is the combined result of several factors. Dendritic synapses can produce more or less current in response to a given number of contacts. The trigger zone may be more or less responsive to summed dendritic current. Neurons also have a range of temporal properties (Bullock 1993). There are differences of latency: some cells respond briskly and some sluggishly. Some cells which have an intrinsic subthreshold oscillatory cycle, fire with a phase offset by some significant interval from the phase of inducing activity. There are differences of refractory period, the amount of time it takes the cell to be ready to fire another spike. There are differences in regularity or irregularity of firing: differences in whether there is a regular interval between spikes, whether the interval fluctuates, whether fluctuation is patterned. There are also differences in the speed of spike conduction down an axon. A cell's dynamic properties can be changed in several ways. Neuromodulators diffused into intercellular fluid alter synaptic structure and can thus have the effect of shifting the operating mode of a cell. A cell's trigger zone threshold may be altered simply by being active; activity increases the cell's gain, to some maximum. Cellular gain differences can be a result simply of location. Where a cell is determines its receptive and projective fields, the three-dimensional geometry of paths and loops it participates in. Its connections determine its activation history. And activation history combined with reward history (i.e. a conjunction of activity and the presence of certain neuromodulators) determines the ongoing response properties of its synapses, which are modified so that they will be more likely to respond to input configurations that have been successful at other times. Practice effects may be increased by adding more overall activity to the system. This can happen through input boosts from different parts of the brain, for different reasons. It can also come about through extracellular neurotransmitter diffusion.
The neuronal group or assembly -- an instance of which is the cortical column -- can be thought of as the basic differential element in cortical function (Hebb 1949, Edelman 1987). A neural cell assembly is a group of hundreds or thousands of similar, strongly connected cells whose synaptic properties ensure that they will react together. Mutual sensitivity of this kind may be built into groups of cells during development. Edelman calls the neuronal groups roughed-in before behavioral interaction primary repertoires: Because of the epigenetic influence of morphoregulatory mechanisms, large numbers of variant connections are formed during development in a particular brain area, connections that, at their finest ramifications, differ from individual to individual. These variant connections arise because of the dynamics of primary processes under local constraint of the morphoregulatory molecules and because various processes of developmental selection take place. This selection occurs by means of synaptic reinforcement, neurite extension and retraction, cell migration, and cell death. Moreover, because these processes are more or less local in scale and because neurites in general have restricted lengths, it is assumed that groups of more strongly connected neurons are formed in nuclei or laminae of the brain. Edelman 1988, 191 Neuronal groups can also result when synchronous response occurs under reward conditions -- that is, when synapses active together are altered transiently or permanently under the influence of neuromodulators. Edelman calls cell assemblies formed as a result of environmental interaction secondary repertoires. Connectionist models have shown how neuronal groups could form sheets displaying sharp gradients of response specificity just in virtue of local competitive interaction (Constantine-Paton and Law 1990). Lateral connections within an array of local groups provides the basis for competitive settling in response to activity propagated into the array from other areas. The neuronal group whose participation is most adaptive will be selected in the sense that the connections of that group to other selected groups will be strengthened when the groups are active together. That group will become part of a differentiated through-line. When it is located in a multidimensional matrix a neural group's response must emerge from a constellation of factors all operating simultaneously and all contributing with different relative weights. Bat neuroethologoists have for instance found neurons in periauditory cortex that may respond to a particular coincidence of a signal from the left ear and a signal from the right ear, but only where both are responses to some specific frequency. Other neurons may respond to the simultaneous presence of several or many frequencies, but only at particular amplitudes. And so on (Suga 1994). The sensitivity of a neuronal group can be very specific or quite coarse. In other words, a neuronal group may respond under a variety of conditions, or it may respond only to combinations of circumstance so particular that the response amounts to grandmother detection.
Selection of neuronal groups within arrays, occurring simultaneously at many sites located at many levels and in many regions of the brain, seems to be the basis of any cortical work performed. What happens in local regions of the brain is often a function of what is happening in specific combinations of other organs: the hand and the eye, or the left cochlea plus the right cochlea and the neck muscles. If the whole body is related to things, in whole contexts, then facilitating brain structure must be both distributed and connected. It must be highly differentiated and, at the same time, highly integrated. What is needed for the spatial and temporal coordination we see in complex behavior is thus a wide-flung temporary configuration of circuit coherency in neuronal groups at foci along streams traversing the whole of the brain. Functional imaging data from monkeys or people performing tasks have confirmed that many sites are active during any task. Activity in different pathways shows temporal dependencies and large differences in strength (Van Essen et al 1992). Path analysis of brain scan data combines what is known about anatomical connections between areas with statistical analysis of these covariance data. Regions that vary consistently during a task (negatively or positively) are taken as functionally interdependent. Sequence is taken as evidence of causal order. Strength of covariance is taken as an indication of strength of influence. Ungerleider suggests that part of normal brain function must be modulation of covariance relations among different brain regions (Mcintosh et al 1994, 663). Wide netsI have said that the momentary aboutness of an organism is its structure at that moment, insofar as that structure is its means of being related to other things. The cortical structure of the organism in the moment participates in that relatedness, which is a selection from among the indefinitely many structural states possible to a body of that species and somatic history. Having briefly described the evolution and development of cortical structure, and some of the basics of cortical anatomy and function, I am in a position to reconsider how best to imagine the overall state of cortical response at any particular moment. We haven't settled how best to talk about the building up and melting down of structured electrochemical states within and around neural connections. We are in the midst of trying out metaphors. There are spatial facts: a cell here, a synapse here, an myelinated axon stretched from here to here. But any way we are presently capable of imagining them, especially in masses and in action, is necessarily fictional. We will have to settle for whatever metaphor does least damage. Pribram (1991) talks about the construction of holoscapes. Churchland (1989) talks about thermodynamic settling into positions or trajectories in state space. Varela (1984) talks about shifting dynamical landscapes. All of these ways of speaking have the advantage that they encourage us to think of a cognitive event in physical terms. Connectionist models of parallel processes, brain imaging methods that show distributed cortical activity, the impressive connectivity discovered by physiologists using staining techniques, have all been suggesting that a network is our best image of the over-allness of cortical action. Hebb (1949) was one of the first to describe cell assemblies that are complex reverberatory circuits. Recent strong proponents of a vision of resonant networks are Mesulam (1981, 1990), Edelman (1987, 1989), Freeman (1991), Van Essen (1991), and Zeki (1997). A structure is a net in virtue of anatomical connections, levels of activity, and timing coincidences. Mesulam calls these temporary structures large-scale networks. Edelman emphasizes reentrant loops that create transiently stable coherent structures that can be both deep, reaching from brainstem to cortex, and wide, reaching across both cortical hemispheres and into all lobes.
What are the advantages of imagining distributed cortical activity as networks? The network paradigm is very rich. It easily accommodates the sort of complex systems thinking required by biological embeddedness and causal multiplicity. If the many parts of a whole body are related to things in complex contexts, then coordinating brain structures will have to be distributed and connected. It is possible to imagine a network integrating the complex aboutness of an embedded organism responding to many things about an environment. When behaviors are seen as facilitated by multi-site networks, subtasks in those behaviors can be thought of as facilitated by subnets embedded in those larger nets. Subnets can be thought of as dynamically integrated with and segregated from the larger net, perhaps by timing differences (Nunez 1999), so that subnets accomplish subcycles of a task -- the way individual finger motion must, for instance, be coordinated with and yet temporally independent of hand motion. Unlike a gadget or module, wide nets can be imagined as having temporal as well as spatial form: they can evolve over time. Parts of a network need not be seen as atomic or assembled. They can be understood as having coevolved within the context of the evolving whole. A network can easily be thought as preserving through-function from sensors to effectors while (since there are many in-points to a network) integrating effects at many points of complex covariance on the way. A network is generally envisaged as a figure against a ground, but it can be thought of as integrating electrical field and neurotransmitter influences both in and around the neural cell. Connected structure is a natural image for multiple dependence. Alteration at nodes in a network is a way of visualizing nonlinear integration as easily as linear integration. A network can easily be understood as functioning by change of structure, building itself on site. It is not easily misunderstood as simple transport of 'signals' or 'information' across irrelevant or inert spaces, through isolated and restricted channels. Since back-connections in a network allow activity to be reentrant or recurrent, a network is not easily seen as hierarchical. Sensor-effector loops at different levels -- Bullock's "manifold integrative tiers" -- can be seen as supplying alternative or ampliative depths of context, rather than shunting data through from source to sink. Most important, the network metaphor gives us a way of understanding that conceptual distinctions do not imply central dissociation of means. It allows us to imagine one structure participating in the perception of both an object and its background, and both the self and the world. Contrasted kinds of functions -- perceiving and imagining, feeling and thinking -- can be thought as performed by the same, or overlapping, structure. Thinking in terms of networks allows us to imagine perception and action participating in one another, integrated in many ways simultaneously, the scatter of nodes distributed between primary sensory and motor matrices coordinating aspects and stages of a behavior. As I will describe below, networks also allow us to imagine the means of sentient response as enfolded among, and interactive with, the means of nonsentient response. When we imagine sentient response as a network we more easily imagine it nuanced and manifold. For now, this seems to be the way to say it: any neural moment is a wide-flung net that is both a three-dimensional reentrant flow and a reverberant standing structure. It is a structure made by the cooperative activity of millions of neurons in many parts of the cortex -- a spatiotemporal structure, both spatially configured and sequenced. The rest of this chapter will consider cortical wide nets in more detail. Building the cortex in immediacy: segregation and integrationThe superstructure of an active net is the pre-existing layout of connective possibilities, which form potential through-routes that intersect other connective routes at junctures where their propagated activity may be convolved. As described above, neurons and matrices at any position on these paths already have structural characteristics as a result of their position and their history. A momentary state of wide net activity is constructed as a result of changes at many positions, over many scales: the synapse, the neuron, the neuronal group, the matrix, the stream and loop. The structure of the net at any moment will be a function of all the variables described so far: cell characteristics, which include training effects at synapses, neuronal group construction, and matrix reconstruction as a result of activity. Because changes at any of these positions and scales are integrated in the overall state of the wide net, it is not possible to say that any one of these positions or scales is responsible for some singular aspect of the organism's differentiated state. What is happening at any position or scale of integration may be necessary to the overall differentiation achieved by the organism, but -- given the competitive self-organization of neuronal groups in matrices and the inherently integrative nature of activity in through-streams -- it can never be sufficient.
Reentrant integration can have important cognitive consequences (Tononi, Sporns and Edelman 1992). One is that, since trigger zone thresholds are lowered when cells are active, total activity in a loop will increase; any discrimination being made could be energized over time. Neurons with high activity levels will entrain other neurons with which they have practice-related connections, and expanded patterns of activity can accumulate. (This might be what happens in the sort of associational thinking experienced with intimate speech or free writing.) There can also be the opposite effect: with increased activity, and with converging activity from other sources excited by the reentrant loop, neural decisions can be less diffuse, more precise. Spike trains with a cyclic character, if they are propagated in loops, may auto-correlate. After several interactions, activity peaks and troughs may be reinforced. Connected areas oscillating in phase are said to be cooperative or phase coherent. Such coherency in a local matrix can have the effect of pattern clarification. Across regions, reentrant synchronization is thought to coordinate activity at different foci and so integrate response decisions of different kinds, allowing neuronal groups in one area to register differentiations made in others: to be contextual, in fact. In cognitive terms, the accumulation, stabilization and clarification of structures of neural activity in a reentrant, synchronized, wide net allows the organism to be effective in the context of many factors simultaneously present -- to be about many things at the same time.
Scalp EEG and MEG readings discover configurations of local and global frequencies such that the dynamics of local processes are partly distinct from global dynamics. Local frequencies result from differences in connectivity, neural gain, and incoming activity. Globally coherent frequencies may be set up by long-range axon delays as well as by midbrain driving centers such as the thalamus. Nunez (1999) suggests a reciprocal relation between local and global dynamics, whereby local frequencies together drive global activity, and one or more of the global frequencies entrain local frequencies, synchronizing them with other centers. But, given that it is a complexly interconnected dynamical system, how does the cortex manage differentiation along with integration? How do we manage to be about one thing rather than another while maintaining coordinated action of many parts? How does the cortex manage state shifts, transient (as when we suddenly do or think completely different things) or longer lasting (as in the operating changes that are differences of mood and waking state)? How does it manage long-term structural change whose effect is limited to the circumstances in which it is relevant? The answer in all of these cases seems to be that the cortex organizes subnets. It does so by various means: by organizing incoming activity so that it is strongly cyclic, by diffusing neuropeptides that lock into binding sites in a synapse, by permanently altering connections, and presumably by many other means as yet undiscovered.
The distributed yet integrated nature of cortical object response suggests that there must be subnets that span wide areas of the cortex but are tightly interactive. How would they be set up? Little is known here, but timing effects may have something to do with this sort of widely integrated partial segregation. At the neuron, timing coincidences matter because synapses and trigger zones have integration intervals. Neurons have functional time constants -- windows within which changes of membrane resting potential will sum. They may for instance respond to spikes within a few 10s of ms, spikes per 200 ms, .5s, or 2s. Coherence of incoming activity at many synapses will thus drive the cell to greater activity. A cell's operating properties can be reset by some of the means already described -- neuromodulator diffusion, slow wave electrical fields, and by synchronized activity itself. Setting up subnets within the wide net of global cortical state may have something to do with resetting a cell's time constants so that it is driven by activity coherent over a different interval. We can imagine this working in object perception, which involves many areas of secondary sensory and association cortex. Seeing a colored, moving object requires activity in areas of all four lobes -- primary visual cortex in the occipital, object form and color areas in the temporal, motion vision areas in the parietal, and spatial attention coordination in the frontal. This is a very wide net with many fine and large branches, but during object perception activity in all of these areas begins and continues to be driven from primary visual cortex. When we attend to an object, transducer timing broadcast into the wide net may segregate a subnet while also integrating its widespread parts. Maybe the integrated segregation that is object perception has to do with recurrent coherencies in the interconnected areas driven by the same transducer patterns.
Because of their evolutionary centrality, primary sensory and motor areas are not as plastic as association areas, but even in primary visual cortex, neuronal sensitivity can be modified by structural changes to synapses and trigger zones. We know of several kinds of deep reentrant loops that are able to act as comprehensive organizers because they change the operating state of neurons in many areas simultaneously (Mesulam 1990, Ungerleider 1998). Global operating states may be set up through loops between brain stem nuclei and cortex. There are a number of important nuclei, each of which releases a different neuromodulator into the intercellular fluid surrounding synaptic contacts in the thalamus and cortex. Their projections can be very diffuse or quite specific. Different systems project into different areas; different regions have different kinds and densities of receptors. [2-7 Neurotransmitter systems] These systems can make global changes in the operation of all parts of the cortex. They can also make large changes to the rest of the body, to the spinal cord, to the stomach, to the immune system, to glands. Mode changes that result from these changes may be felt as mood or emotion, and are part of what we mean by motivation and drive. They change what we notice and how we deal with it: they are saliency systems. Mode changes include shifts between waking and sleep or alertness and reverie. Each of these modes shows characteristic slow wave frequencies; it may be that basal areas change mode by pacemaking as well as by chemical diffusion. The PGO loop (pons, geniculate, occipital), for instance, is thought to have something to do with setting frequencies associated with rapid eye movement during dreaming sleep (Hobson, Pace-Schott and Stickgold 2000) Edelman points out that changes of these sorts can be widespread, integrative, and yet task-specific: Because saliency-related diffuse projections can affect virtually the entire cortical mantle, entire cortical states and all of the cooperative interactions that lead to their establishment can be selected during reinforcement. This results in synaptic changes in many different pathways, including some whose involvement in the task at hand may not be immediately obvious ... In a task that requires several specialized areas, the system can allocate a large number of connections in different pathways. Thus, the model reconciles distributed synaptic changes in learning with functional segregation while avoiding the error of 'equipotentiality'. Tononi, Sporns and Edelman 1992, 332
The transient stability of the network configuration of a moment must sometimes, but not always, have long-term structural consequences. We learn very selectively -- how does the wide net organize relevant adaptation? Memory also seems to be a mode change directed from basal centers: learning is a mode change that organizes long-term alterations of structure; recall is a mode change that organizes reconstruction by means of these alterations (Fuster 1995; Ungerleider 1995). Changes in synaptic potential can last for moments, hours, months, years. Priming is an immediate, transient effect shown primarily in subsequent speed of response. Long-term potentiation occurs in stages. Short-term associations are established over minutes or hours, but long-term memory consolidation may need four weeks (Ungerleider 1995). The subnet active in setting memory changes seems always to include the hippocampus and nearby temporal lobe structures deep in medial (or internal) cortex. These structures have reciprocal connections with many sensory and motor areas and so are considered to be in a unique position to help establish these links (Ungerleider 1995, 773). Sentient aboutness and wide netsIn relation to sentience it is as if there are two different sorts of things to be explained. One is how we come to be aware of one thing rather than another. The second is something about the bare fact of sentience -- how one physical entity can see or feel, or seem to see or feel, another, even in the simplest way. The sorts of brain theory we can see coming still have more to say about the first question than about the second. It looks as if we have to take basic sentience as a given, for now, and hope we know more about how to talk about it when we know more about how we do the things we do sentiently. We know that when we are perceiving and acting consciously structures of activity in the cerebral cortex are necessary to that perceiving and acting. We know perception and action mediated by central structures are not always conscious. We also know that, in a nondualist theory of cognition, neural events organizing sentient or conscious function must be imagined as spatiotemporal forms. These forms are in some way the event of sentience -- the sentient moment -- at the same time that they are consequentially connected to other spatiotemporal forms in the brain. We can escape the omnipresent invitations to call those aspects of physical structure that are involved in sentient function representations by saying there are structures that are the central means of our momentary aboutness, and some of them are means also of our momentary sentient relatedness to things.
temporary fine-time coincidence of high levels of activity Edelman 1987 We know that the difference between sentient and non-sentient cognitive function must be a physical difference, but we do not know what that difference is. Many theories are afloat. Damasio (1999) thinks it has something to do with location: sentient experience is based solely on activity in early sensory cortex and in subcortical structures reentrantly interactive with them. Crick (1992) thinks it has something to do with a specific cyclicity of neural firing: he suggests 40 Hz is the magic number. It may have to do with the activity of some neuromodulators and not others. Pribram (1991) has been suggesting it is an effect of electrical fields present around dendritic microprocesses. Freeman (1991) notes that monkeys making conscious olfactory discriminations show well-defined spatial patterns of carrier wave amplitude in the olfactory bulb; when the odorant discriminated is changed, there is rapid establishment of a completely different spatial pattern. I have described a wide net of propagated and standing activity, a web reaching into many parts of the brain, transiently stable, changing with task and environment. Edelman thinks of this wide net as made up of two concurrent kinds of subnet (Tononi and Edelman 1998), one of which is crucial to sentient function. In Edelman's formulation, global mappings are the nets responsible for discriminations and coordinations we don't make consciously. They must be both distributed and integrated, because we can perform very complex behaviors without awareness. Global mappings would include structures formed in response to internal as well as external conditions: they can be thought of as forming an active background or cognitive milieu that primes sentient activity and is its always-active context. Global mappings would include what Damasio (1994) calls reactive dispositions. Edelman thinks global mappings include portions of the thalamocortical sensor-effector through-system, with parallel loops through basal ganglia, hippocampus, and cerebellum. Edelman's dynamic core hypothesis is that sentient doing is organized by a distributed but tightly connected subnet. Like a global mapping, the dynamic core is made up of neuronal groups selectively activated within arrays. But neuronal groups participating in the moment's dynamic core would be more strongly interactive with each other, more strongly synchronized through reentrant connections and thus more energized as networks -- a brighter web within the bright web that is the neural system's total selection of state. Edelman hypothesizes that the dynamic core differs from the global mappings that are its ground and context in three main ways. Global mappings probably exist in areas where dynamic core activity seldom or never appears. Dynamic core structure changes more rapidly; it has a processing period of tens of milliseconds, whereas global mappings are formed and shift over a scale of seconds rather than milliseconds (Tononi and Edelman 1998, 1851). Neuronal groups within the dynamic core are more highly synchronized in their activity, and thus are more energized and more integrated. The complex functional cluster of neuronal groups constituting a dynamic core is capable of abrupt simultaneous shifts in functional connectivity; because it is so tightly interconnected, the subnet shifts much more rapidly than the wider net of which it is a part. A hyperenergized subnet, making selections within the broadly contextual global discriminations occurring at the moment, could supply the focus needed to behave in relation to particular aspects of an environment while taking into account, as an active background, many aspects of the environment not focalized. A hyperactive core subnet could be constructed by a combination of mechanisms. One would involve selective participation of subcortical drive systems projecting to just those thalamic and cortical areas thought to be most active in the dynamic core. Neuromodulators diffused into specific thalamic and cortical nuclei would boost their reactivity. Inhibition organized from drive systems would also be important: damping some areas would help to hype others. Subcortical systems are also thought to provide strongly cyclic activity that helps to organize and synchronize the core net. Probably most important would be activity arriving from sensor surfaces, supplying strong exogenous periodicities. But even this strong exogenous source of core dynamic selection can be overridden by basal-forebrain loops which are the means of deliberate attention. Damasio has hypothesized cortical structures he calls convergence zones (1994), that would act as switches evoking core subnet structure. Sejnowski (1989) talks about skeleton filters in similar terms: skeleton filters would be embedded net components, neuronal groups presumably, whose synaptic characteristics enable them to set up rapid shifts in the connectivity of surrounding neural groups. Neuronal groups in certain areas could be part of a global mapping or could participate in the dynamic core and thus be involved in conscious processes. Global mappings are thought to be more stable than the dynamic core, but the two partially segregated dynamic processes could also interact. Activity in global mappings could prime core dynamic activity by setting operating characteristics of cells and circuits. The hyperactive core would presumably also recruit. It may be that, as it stabilizes over time, activity in global mappings could reach threshold levels and be incorporated into the hyperactive core subset of the moment. Old ideas of mental association would fit here. Neuronal groups with trained connections to groups in the dynamic core could be included if the dynamic core were stable in their neighbourhood for those associations to be called up. Like Damasio, Edelman includes the thalamus in the core of activity which is the means of sentience, but he thinks it extends also into posterior and anterior cortical regions that are not early sensory regions.
If aboutness involves bodily directedness and interactional contact, and is a relational state of whole organisms, then its means include the whole body and the environment too. The necessary conditions of sentient aboutness similarly include both internal and external structure. Taken in isolation a neural structure cannot be about anything. If aboutness is relatedness, and if relatedness occurs by deep co-occurrent means, but if, at the same time, sentient relatedness requires subnet organization, and partial or simulational sentient relatedness can occur without the immediate presence of entire environmental means, how should we think about a sentience subnet -- without calling up a representation metaphor? The exact form of the sentience subnet determines -- is that the word? -- something. What? Don't say 'experience' or 'consciousness' or 'phenomenology', and then ask how neural form 'is related to' these things. That locution sets us up to think of looking from one thing to another and comparing them, looking for correspondence, the way we could compare a scene with a picture of a scene. The moment's contact (contact meaning the whole of co-present action, self-talk, and -- what to call it -- detailed felt judgment) is not scenic: even vision is not scenic, since it crucially requires motion (Gibson 1966; Noe 2001). The cortical means of the moment's contact have spatial and temporal form, and this form matters, but it is not pictorial form. It does not look like anything; it is electrodynamical. Aboutness and its means cannot be compared, correlated, mapped. The organism's aboutness cannot be compared with a subset of its means. This subset cannot correspond to, or be isomorphic with, a subset (the 'conscious' part) of the moment's relatedness. And yet the particular form of the subnet is responsible for the particularity of sentient relatedness. Are parts of the subnet responsible for parts of the relatedness? Are there "specific nerve energies"? Is it possible to say this part of a core dynamic net is the means of my relatedness to the face I am seeing? No, because relatedness does not have parts. An organism is related in different ways to different things, which are parts of the situational/environmental whole. A related state is a state of a body that has parts. But the relatedness does not have parts. And yet the brain does have specialized regions. Although an external and internal circumstance is the means of my relatedness to a face, we find that a local part of the brain is specialized for differentiating faces; we have learned that reach and grasp are organized by partially segregated networks anchored at different nodes in parietal and frontal cortex. There is no end of these observations. Like other parts of the body, the cortex functions by structural differentiation. Learning about the brain is learning about these differences, as we are doing from single-cell readings, from PET and MRI and other kinds of imaging, and from function studies after lesions and reversible deactivation. But we can understand localization results in network terms rather than in representational terms. We can think of single-cell readings in visual cortex as sampling integration at a point, rather than as locating 'an image'. We can think of lesion or reversible deactivation as disrupting one or more streams through an area, rather than as damaging 'a module'. Using representational metaphors, we look for a "convergent distillate" 'constructed', 'stored' or 'transformed' in a locus. Imagining a network, we can think instead of a distributed structure most effectively accessed there (Mesulam 1990, 599). Thinking in terms of network effects is in agreement with multi-site distributions found in imaging studies for any complex behavior; with the fact that activity levels peak over nodes but tail off evenly rather than dropping off sharply (Martin, Ungerleider and Haxby 2000); with the fact that lesions result in slowed or faulty performance but seldom result in entire loss of behaviors; and with the connectivity found by tracers. If central means are not thought of as 'a representation', then there is less push to think of parts of the net as being the means of seeing or seeming to see parts of a scene. If we think in terms of networks it is easier to remember that parts of responsive structure are constructed as a dynamic result of occurrences in other parts. When we tell a dream we talk about parts of a place: the road, the river, the hill, the person in a cloak. We talk about moving and acting in relation to those parts. Imagining a net, we understand that using separate names for walking and road does not imply that the means by which we seemed to walk and seemed to see a road were distinct sub-parts of the whole instance of dreaming. Or that the means by which we dream are distinct from the means by which we tell a dream. A sentient subnet can't be thought of as a mosaic, because its zones are interactive, mutually determined. What it means to talk about neural form without talking about it in terms of representations is this: structures that are part of the means of sentient engagement, are not about anything. They are part of the means by which the person is, or seems to be, about something (Wittgenstein 1956, 281). The particularity of the form matters, but not because it corresponds (or not). It is difficult to think this clearly: the means are not mosaic, objects and contexts contacted are not mosaic, the proximal medial field is not mosaic, relatedness cannot be mosaic, and yet the specifics of every part of the subnet matter to the sentient relatedness of the moment. I find I have to keep saying this. In sumnot like a computer ... or a telephone exchange; more like a vast aggregate of interactive events in a jungle Tononi, Sporns and Edelman 1992, 69 Imagining the brain is at this point a fictional exercise. We are not yet structured to do it right, so we must acknowledge our metaphorical liberties. If we honour its evolved origins and continued participation in organic nature, we could perhaps be bolder and more visual than we have been. [2-10 Brain drawn to resemble a flower] We could say the brain is like a plant that has grown lawfully from a point, or maybe like a garden that has flourished from many points, each yielding its characteristic texture. Or that it is like an earth composed of many sizes and colors of grain, penetrated by very soft rootlets, traversed by chemical seeps. We could say cortical structure is like stream bed silt scored by twining rivulets into precisely fluid channels that take account of every shape in the terrain. Or that it is like a cloud that reforms constantly, at many points at once, condensing around elements that are charged and mobile but invisible until the white vapour marks them. Or, most accurately perhaps, we can imagine the brain as a sea traversed by precise and fluid streams of electrical and chemical activity, in effect by streams of change. We can imagine some of these streams as originating at sensory surfaces and propagated through brainstem and midbrain into and throughout the cortex. They can be imagined splitting into streamlets, diverging, reconverging, joining streams with other origins. Unlike the streamlets scoring sand or silt, these streams circle back on themselves, form loops at many scales -- minute loops within a cortical column, larger loops connecting regions in a lobe, and even larger inter-organ loops from midbrain or cerebellum to cortex. Connected streams that loop back onto themselves can be imagined as three-dimensional nets or webs. Since they are nets of electromagnetic effect, they can be imagined as nets of light. Since they may reach from many sources through many centers, at all levels and in all regions of the brain, they may be thought of as parts of even wider nets pegged at many points. We could imagine the many origins of the flow of light, some always active, some dying out as others erupt. The streams that fan out from these sources join each other at many centers. At any moment the net can be dense in some places and loose or absent in others, and it must be imagined as changing shape, shifting suddenly or slowly, dancing into new areas, maybe strobing or flickering. We could also imagine a smaller but brighter net, based in the larger net but moving at a different speed, maybe more stable, maybe flickering at a different rate, and shifting about in the areas brought alive by the global net. The wide net figure can help with thinking the nuances of knowing. That is the main thing. I want to imagine the immediate and central means of sentience integrated and distributed because understanding them that way gives me a way to think about for instance linguistic or musical or even mathematical knowing as they are experienced. Here's an example: linguistic predication. To predicate is to use linguistic forms in such a way that people are organized to think of some thing in a particular way. The berries are ripe. When we say that, we are organized to think, imagine, see, regard -- to be related, or as if related, to -- berries in terms of ripeness. Should topic evocation and predication be thought of as separate acts? Do they happen in different places? Probably not. There is a dynamical sequence. If we think of a wide net evoked by a name and then modified by subsequent language, we can imagine the effect of a sentence as accumulating modifications of the form of a net, and of the over-all effect of the sentence as an achieved dynamic result. Where name-effect is not as specific as we need it to be, we use a predicate to modify it. Structure is evoked, and then that structure is modified. Whole paragraphs can be thought of as modifying the effect of an evoked topic. In this way, a wide net metaphor can make Wittgenstein's seeing as a central instance of linguistic effect, rather than a representational puzzle. (More about language in Chapter 6.)
Part II. Presence and simulation |