Can Neurobiology Teach us Anything about Consciousness?^*

9/25/95

I Introduction:

In assuming that neuroscience can reveal the physical mechanisms subserving psychological functions, I am assuming that it is indeed the brain that performs those functions -- that capacities of the humans mind are in fact capacities of the human brain. This assumption and its corollary rejection of Cartesian souls or spirits or "spooky stuff" existing separately from the brain is no whimsy. On the contrary, it is a highly probable hypothesis, based on evidence currently available from physics, chemistry, neuroscience and evolutionary biology. In saying that physicalism is an hypothesis, I mean to emphasize its status as an empirical matter. I do not assume that it is a question of conceptual analysis, a priori insight, or religious faith, though I appreciate that not all philosophers are at one with me on this point. [2]

Additionally, I am convinced that the right strategy for understanding psychological capacities is essentially reductionist, by which I mean, broadly, that understanding the neurobiological mechanisms is not a frill but a necessity. Whether science will finally succeed in reducing psychological phenomena to neurobiological phenomena is, needless to say, yet another empirical question. Adopting the reductionist strategy means trying to explain the macro levels (psychological properties) in terms of micro levels (neural network properties).

The fundamental rationale behind this research strategy is straightforward: if you want to understand how a thing works, you need to understand not only its behavioral profile, but also its basic components and how they are organized to constitute a system. If you do not have the engineering designs available for reference, you resort to reverse engineering -- the tactic of taking a part a device to see how it works.[3] Insofar as I am trying to discover macro-to-micro explanations, I am a reductionist. Because many philosophers who agree with me on the brain-based nature of the soul nonetheless rail against reductionism as ridiculous if not downright pitiful, it may behoove me to begin by explaining briefly what I do and, most emphatically, do not mean by a reductionist research strategy.[4]

Clearing away the negatives first, may I say that I do not mean that a reductionist research strategy implies that a purely bottom-up strategy should be adopted. So far as I can tell, no one in neuroscience thinks that the way to understand the nervous system is first to understanding everything about the basic molecules, then everything about every neuron and every synapse, and to continue ponderously thus to ascend the various levels of organization until, at long last, one arrives at the uppermost level -- psychological processes. (Figure 1) Nor is there anything in the history of science that says a research strategy is reductionist only if it is purely bottom-up. That characterization is straw through and through. The research behind the classical reductionist successes -- explanation of thermodynamics in terms of statistical mechanics; of optics in terms of electromagnetic radiation; of hereditary transmission in terms of DNA -- certainly did not conform to any purely bottom-up research directive.

Figure 1 about here

So far as neuroscience and psychology are concerned, my view is simply that it would be wisest to conduct research on many levels simultaneously, from the molecular, through to networks, systems, brain areas, and of course behavior. Here, as elsewhere in science, hypotheses at various levels can co-evolve as they correct and inform one another. [5] Neuroscientists would be silly to make a point of ignoring psychological data, for example, just as psychologists would be silly to make a point of ignoring all neurobiological data.

Second, by reductionist research strategy I do not mean that there is something disreputable, unscientific or otherwise unsavory about high level descriptions or capacities per se. It seems fairly obvious, to take a simple example, that certain rhythmic properties in nervous systems are network properties resulting from the individual membrane traits of various neuron types in the network, together with the way the set of neurons interact. Recognition that something is the face of Arafat, for another example, almost certainly emerges from the responsivity profiles of the neurons in the network plus the ways in which those neurons interact. "Emergence" in this context is entirely non-spooky and respectable, meaning, to a first approximation, "property of the network". Determining precisely what the network property is, for some particular feat, will naturally take quite a lot of experimental effort. Moreover, given that neuronal behavior is highly nonlinear, the network properties are never a simple "sum of the parts". They are some function -- some complicated function -- of the properties of the parts. High level capacities clearly exist, and high level descriptions are therefore needed to specify them.

"Eliminative materialism", refers to the hypothesis that (1) materialism is most probably true, and also (2) many traditional aspects of explanation of human behavior are probably not adequate to the reality of the etiology of behavior. [6] The standard analogy here is that just as "caloric fluid" was useful but fundamentally mistaken in understanding thermal phenomena (conduction, convection, radiation) so some psychological categories currently invoked may be somewhat useful but fundamentally mistaken in fathoming behavioral etiology. Other existing characterizations of capacities may have a core of adequacy but undergo major redrawings, in something like the way Mendel's notion of "factor" came to be modified by genetics into the notion of "gene" which itself was modified and deepened with the development of molecular biology. Some categories such as "attitude" are extremely vague and might be replaced altogether; others, such as "is sleeping" have already undergone a fractionation as EEG and neurophysiological research has revealed important brain differences in various stages of sleep. Categories such as "memory", "attention" and "reasoning" are likewise undergoing revision, as experimental psychology and neuroscience proceed.[7] It remains to be seen whether there is a neurobiological reality to sustain notions such as "belief" and "desire" as articulated by modern philosophers such as Fodor[8] and Searle[9], though Paul Churchland and I have argued that revision here too is most probable. [10] The revisionary prediction too is an empirical hypothesis, and one for which empirical support already exists. [11]

The possibility of nontrivial revision and even replacement of existing high level descriptions by 'neurobiologically harmonious' high level categories is the crux of what makes eliminative materialism eliminative. [12] By 'neurobiologically harmonious' categories, I mean those that permit coherent, integrated explanations from the whole brain on down through neural systems, big networks, micronets, and neurons. Only the strawman is so foolish as to claim that there are no high level capacities, that there are no high level phenomena. [13] In its general aspect, my point here merely reflects this fact: in a profoundly important sense we do not understand exactly what, at its higher levels, the brain really does. Accordingly, it is practical to earmark even our fondest intuitions about mind/brain function as revisable hypotheses rather than as transcendental absolutes or introspectively given certainties. Acknowledgment of such revisability makes an enormous difference in how we conduct psychological and neurobiological experiments, and in how we interpret the results.

II NAYSAYING THE NEUROBIOLOGICAL GOAL

A. The Goal is Absurd (Incoherent)

One set of reasons for dooming the reductionist research strategy is summed up thus: "I simply cannot imagine that seeing blue or the feeling of pain, for example, could consist in some pattern of activity of neurons in the brain," or, more bluntly, "I cannot imagine how you can get awareness out of meat." There is sometimes considerable filler between the "it's unimaginable" premise and the "it's impossible" conclusion, but so far as I can tell, the filler is typically dust which cloaks the fallacious core of the argument. [14]

Given how little in detail we currently understand about how the human brain "en-neurons" any of its diverse capacities, it is altogether predictable that we should have difficulty imagining the neural mechanisms. When the human scientific community was comparably ignorant of such matters as valence, electron shells, and so forth, natural philosophers could not imagine how you could explain the malleability of metals, the magnetizability of iron, and the rust resistance of gold , in terms of underlying components and their organization. Until the advent of molecular biology, many people thought it was unimaginable, and hence impossible, that to be a living thing could consist in a particular organization of "dead" molecules. "I cannot imagine", said the vitalists, "how you could get life out of dead stuff".

From the vantage point of considerable ignorance, failure to imagine some possibility is only that: a failure of imagination-- one psychological capacity amongst others. It does not betoken any metaphysical limitations on what we can come to understand , and it cannot predict anything significant about the future of scientific research. After reflecting on the awesome complexity of the problem of thermoregulation in homeotherms such as ourselves, I find I cannot imagine how brains control body temperature under diverse conditions. I suspect, however, that this is a relatively uninteresting psychological fact about me, reflecting merely my current state of ignorance. It is not an interesting metaphysical fact about the universe nor even an epistemological fact about the limits of scientific knowledge.

A variation of the "cannot imagine" proposal is expressed as "we can never, never know....", or "it is impossible to ever understand...." or "it is forever beyond science to show that....". The idea here is that something's being impossible to conceive says something decisive about its empirical or logical impossibility. I am not insisting that such proposals are never relevant. Sometimes they may be. But they are surprisingly high-handed when science is in the very early stages of studying a phenomenon.

The sobering point here is that assorted "a priori certainties" have, in the course of history, turned out to be empirical duds, however obvious and heartfelt in their heyday. The impossibility that space is non-Euclidean , the impossibility that in real space parallel lines should converge, the impossibility of having good evidence that some events are undetermined, or that someone is now dreaming, or that the universe had a beginning -- each slipped its logical noose as we came to a deeper understanding of how things are. If we have learned anything from the many counterintuitive discoveries in science it is that our intuitions can be wrong. Our intuitions about ourselves and how we work may also be quite wrong. There is no basis in evolutionary theory, mathematics, or anything else, for assuming that prescientific conceptions are essentially scientifically adequate conceptions.

A third variation on this "nay, nay, never" theme draws conclusions about how the world must actually be, based on linguistic properties of certain central categories in current use to describe the world. Permit me to give a boiled down instance: "the category 'mental' is remote in meaning -- means something completely different -- from the category 'physical' . It is absurd therefore to talk of the brain seeing or feeling, just as it is absurd to talk of the mind having neurotransmitters or conducting current." Allegedly, this categorial absurdity undercuts the very possibility that science could discover that feeling pain is activity in neurons in the brain. The epithet "category error" is sometimes considered sufficient to reveal the naked nonsense of reductionism.

Much has already been said on this matter elsewhere,[15] and I shall bypass a lengthy discussion of philosophy of language with three brief points. (1) It is rather far-fetched to suppose that intuitions in the philosophy of language can be a reliable guide to what science can and cannot discover about the nature of the universe. (2) Meanings change as science makes discoveries about what some macro phenomenon is in terms of its composition and the dynamics of the underlying structure. (3) Scientists are unlikely to halt their research when informed that their hypotheses and theories "sound funny" relative to current usage. More likely, they will say this: "the theories might sound funny to you, but let me teach the background science that makes us think the theory is true. Then it will sound less funny." It may be noted that it sounded funny to Copernicus' contemporaries to say the Earth is a planet and moves; it sounded funny to say that heat is molecular motion or that physical space is non-Euclidean or that there is no absolute "downness". And so forth.

That a scientifically plausible theory sounds funny is a criterion only of its not having become common coin, not of its being wrong. Scientific discoveries that a certain macro phenomenon is a complex result of the micro structure and its dynamics are typically surprising and typically sound funny -- at first. Obviously none of this is positive evidence that we can achieve a reduction of psychological phenomena to neurobiological phenomena. It says only that sounding funny does not signify anything, one way or the other.

B. The Goal is Inconsistent with "Multiple Realizability"

The core of this objection is that if a macro phenomenon can be the outcome of more than one mechanism (organization and dynamics of components), then it cannot be identified with any one mechanism, and hence the reduction of the macrophenonomenon to the (singular) underlying micro phenomenon is impossible. This objection seems to me totally uninteresting to science. Again, permit me to ignore important details and merely to summarize the main thrust of the replies.

(1) Explanations, and therefore reductions, are domain relative. In biology, it may be fruitful first to limn the general principles explaining some phenomenon seen in diverse species, and then figure out how to account for the interspecies differences, and then, if desirable, how to account for differences across individuals within a given species. Thus the general principles of how hearts or stomachs work are figured out, perhaps based on studies of a single species, and particularities can be resolved thereafter. Frog hearts, macaque hearts and human hearts work in essentially the same general way, but there are also significant differences, apart from size, that call for comparative analyses. Consider other examples: (a) from the general solution to the copying problem that emerged from the discovery of the fundamental structure of DNA, it was possible to undertake explorations of how differences in DNA could explain certain differences in the phenotype; (b) from the general solution to the problem of how neurons send and receive signals, it was possible to launch detailed exploration into the differences in responsivity profiles of distinct classes of neuron. [16]

(2) Once the mechanism for some biological process has been discovered, it may be possible to invent devices to mimic those processes. Nevertheless, invention of the technology for artificial hearts or artificial kidneys does not obliterate the explanatory progress on actual hearts and actual kidneys; it does not gainsay the reductive accomplishment. Again, the possibility that hereditary material of a kind different from DNA might be found in things elsewhere in the universe does not affect the basic scaffolding of a reduction on this planet. Science would have been much the poorer if Crick and Watson had abandoned their project because of the abstract possibility of Martian hereditary material or artificial hereditary material. In fact, we do know the crux of the copying mechanism on Earth -- namely, DNA, and we do know quite a lot about how it does its job. Similarly, the engineering of artificial neurons and artificial neural nets (ANNs) facilitates and is facilitated by neurobiological approaches to how real neurons work; the engineering undertakings do not mean the search for the basic principles of nervous system function is misguided.

(3) There are always questions remaining to be answered in science, and hence coming to grasp the general go of a mechanism, such as the discovery of base-pairing in DNA, ought not be mistaken for the utopian ideal of a complete reduction -- a complete explanation. Discoveries about the general go of something typically raise hosts of questions about the detailed go of it, and then about the details of the details. To signal the incompleteness of explanations, perhaps we should eschew the expression "reduction" in favor of "reductive contact". Hence we should say the aim of neuroscience is to make rich reductive contact with psychology as the two broad disciplines co-evolve. I have experimented with this recommendation myself, and although some philosophers warm to it, scientists find it quaintly pedantic. In any case, "reductive contact" between molecular biology and macrobiology has become steadily richer since 1953, though many questions remain. Reductive contact between psychology and neuroscience has also become richer, especially in the last decade, though it is fair to say that by and large the basic principles of how the brain works are poorly understood.

(4) What, precisely, are supposed to be the programmatic sequelae to the multiple realizability argument? Is it that neuroscience is irrelevant to understanding the nature of the human mind? Obviously not. That neuroscience is not necessary to understand the human mind? One cannot, certainly, deny that it is remarkably useful. Consider the discoveries concerning sleep, wakeness, and dreaming; the discoveries concerning split brains, humans with focal brain lesions, the neurophysiology and neuroanatomy of the visual system, and so on. Is it perhaps that we should not get our hopes up too high? What, precisely, is "too high" here? Is it the hope that we shall discover the general principles of how the brain works? Why is that too high a hope?

C. The Brain Causes Consciousness

Nay saying the reductionist goal while keeping dualism at arm's length is a manoeuvre requiring great delicacy. John Searle's strategy is to say that although the brain causes conscious states, any identification of conscious states with brain activities is unsound. Traditionally, it has been opined that the best the reductionist can hope for are correlations between subjective states and brain states, and although correlations can be evidence for causality they are not evidence for identity. Searle has tried to bolster that objection by saying that whereas a/b identifications elsewhere in science reveal the reality behind the appearance, in the case of awareness, the reality and the appearance are inseparable -- there is no reality to awareness except what is present in awareness. There is, therefore, no reduction to be had.

Synoptically, here is why Searle's manoeuvre is unconvincing: he fails to appreciate why scientists opt for identifications when they do. Depending on the data, cross-level identifications to the effect that a is b may be less troublesome and more comprehensible scientifically than supposing thing a causes separate thing b. This is best seen by example.[17]

Science as we know it says electrical current in a wire is not caused by moving electrons; it is moving electrons. Genes are not caused by chunks of base pairs in DNA; they are chunks of base pairs (albeit sometimes distributed chunks). Temperature is not caused by mean molecular kinetic energy; it is mean molecular kinetic energy. Reflect for a moment on the inventiveness required to generate explanations that maintain the nonidentity and causal dependency of (a) electric current and moving electrons, (b) genes and chunks of DNA, and (c) heat and molecular motion. Unacquainted with the relevant convergent data and explanatory successes, one may suppose this is not so difficult. Enter Betty Crocker.

In her microwave oven cookbook, Betty Crocker offers to explain how a microwave oven works. She says that when you turn the oven on, the microwaves excite the water molecules in the food, causing them to move faster and faster. Does she, as any high school science teacher knows she should, end the explanation here, perhaps noting, "increased temperature just is increased kinetic energy of the constituent molecules "? She does not. She goes on to explain that because the molecules move faster, they bump into each other more often, which increases the friction between molecules, and, as we all know, friction causes heat. Betty Crocker still thinks heat is something other than molecular KE; something caused by but actually independent of molecular motion. [18] Why do scientists not think so too?

Roughly, because explanations for heat phenomena -- production by combustion, by the sun, and in chemical reactions; of conductivity, including conductivity in a vacuum, the variance in conductivity in distinct materials, etc. -- are vastly simpler and more coherent on the assumption that heat is molecular energy of the constituent molecules. By contrast, trying to make the data fit with assumption that heat is some other thing caused by speeding up molecular motion is like trying to nail jelly to the wall. If one is bound and determined to cleave to a caloric thermodynamics, one might, with heroic effort, pull it off for oneself, though converts are improbable. The cost, however, in coherence with the rest of scientific theory, not to mention with other observations, is extremely high. What would motivate paying that cost? Perhaps an iron-willed, written-in-blood, resolve to maintain unsullied the intuition that heat is what it is and not another thing. In retrospect, and knowing what we now know, the idea that anyone would go to exorbitant lengths to defend the heat intuition seems rather a waste of time.

In the case at hand, I am predicting that explanatory power, coherence and economy will favor the hypothesis that awareness just is some pattern of activity in neurons. I may turn out to be wrong. If I am wrong, it will not be because an introspectively-based intuition is immutable, but because the science leads us in a different direction. If I am right, and certain patterns of brain activity are the reality behind the experience, this fact does not in and of itself change my experience and suddenly allow me (my brain) to view my brain as an MR scanner or a neurosurgeon might view it. I shall continue to have experiences in the regular old way, though in order to understand the neuronal reality of them, my brain needs to have lots of experiences and undergo lots of learning.

Finally, barring a jump to the dualist' s horse, the idea that there has to be a bedrock of subjective appearance on which reality/appearance discoveries must ultimately rest is faintly strange. It seems a bit like insisting that "down" cannot be relative to where one is in space; down is down. Or like insisting that time cannot be relative, that either two events happen at the same time or they don't, and that's that. Humans are products of evolution; nervous systems have evolved in the context of competition for survival -- in the struggle to succeed in the four F's: feeding fleeing, fighting, and reproduction. The brain's model of the external world enjoys improvement through appreciating various of reality/appearance distinctions -- in short, through common critical reason and through science. In the nature of things, it is quite likely that the brain's model of its internal world also allows for appearance/reality discoveries. The brain did not evolve to know the nature of the sun as it is known by a physicist, nor to know itself as it is known by a neurophysiologist. But, in the right circumstances, it can come to know them anyhow. [19]

D. Because Consciousness is a Virtual Machine

This is the view of D. C. Dennett. [20] Like Searle, Dennett is no dualist. Unlike Searle, who thinks that quite a lot, if not all, about consciousness can be discovered by neuroscience, Dennett has long been convinced that study of the brain itself -- its physiology and anatomy -- is largely a waste of time so far as understanding the nature of consciousness and cognition are concerned. Simplified, the crux of his idea is this: humans become conscious as they acquire language and learn to talk to themselves. What happens in this transformation is that a parallel machine (the neural networks of the brain) simulates a serial machine (operations are performed one at a time, in a sequence, according to rules, which may be recursive.)

By acquiring a language and then learning to speak silently to oneself, one allegedly creates a consciousness virtual machine in the brain. Dennett explains what this by means of a pivotal analogy: it is like creating a virtual machine for simulating piloting a plane on your desktop computer by installing software, such as Flight Simulator. Consciousness bears the same relation to the brain as the flight simulation bears to the events inside the computer.

Dennett's methodological moral is unambiguous: just as we cannot hope to learn anything much about the flight simulator (it scope and limits, how it works) by studying the computer's innards while it is running Flight Simulator so we cannot hope to learn much about consciousness by studying the brain's innards while it is conscious. If one wants to know about Flight Simulator and its many properties, the best you can do is study its performance -- in a sense, there really is not anything else to Flight Simulator than its performance. We find it fruitful in talking about Flight Simulator to say things like "its altimeter registers altitude", but this does not mean that there is something in my computer that really is high in the sky or something that measures how high it really is. Such talk is simply an economical, convenient way of making sense of the computer's screen performance when it is running Flight Simulator software.

Ditto (more or less) for consciousness. The brain is the hardware on which the consciousness software runs, hence looking at the brain itself is not going to teach us much about the software itself. Even as it is mistaken to suppose the computer has a little runway hidden tucked inside that gets rolled out when I press a button, so it is mistaken to think the brain really does anything like fill in the blindspot or fill in during seeing subjective motion (as in a movie). [21] Dennett believes he has shown us that there really is not so much in the way of inner experience to be explained after all. As with Flight Simulator, if you want to know about consciousness and its properties, it is performance under a variety of conditions that needs to be studied. Based on the performance you can of course infer the various computational properties of the software. And that is all there will be to explaining consciousness. Consequently, the tools of experimental psychology will suffice. The details of neuroscience might tell us something about how the software runs on the brain; that won't tell us anything about the nature of consciousness, but only about how the brain runs software. This, in capsule, is my understanding of the conviction that inspires Dennett to his book's title, Consciousness Explained.

How plausible is the Dennett story? My criticism here draws on work of Paul Churchland[22] and will focus mainly on this question: is it remotely reasonable that when we are consciousness the parallel machine (the brain) is simulating a serial machine? As an archival preliminary, however, note that Dennett's package has been subjected to intense and careful analysis. First, his claim that acquisition of human language is a necessary condition of human consciousness has been repeatedly challenged and thoroughly criticized. [23] Endlessly it has been noted that this seems to imply that preverbal infants are not conscious; that other animals, such as chimpanzees and orangs are not conscious; that subjects with global aphasia or left hemispherectomies are not conscious. Briefly, Dennett's response is this: yes indeed, nonverbal subjects are not aware in the way a fully verbal human is aware; e.g., they cannot think about whether interest rates will go down next month. Unfortunately, Dennett's response is tangential to the criticism. The issue is whether preverbal children and animals can be conscious of colors, sounds, smells, spatial extent, motion, being dizzy, feeling pain, etc. in rather like the way I am conscious of them.

Second, Dennett's according pre-eminent status to linguistic activity and his correlative "debunking" of sensory experiences (e.g. filling in), feelings, and nonlinguistic cognition generally, have been subject to a constant barrage of complaints. [24] Regrettably, I can give here only a highly truncated version of the long and sometimes convoluted debates between Dennett and various critics. The heart of the complaints is that Dennett wrongly assumes that performance is all that needs explaining -- that explaining reports of conscious experience is tantamount to explaining conscious experience itself. Dennett's core response here has been to wave off the complainers as having failed properly to understand him, scolding them for being still in the grip of bad old conceptual habits implying homunculi, ghosts in the machine, furtive Cartesianism, and kindred mistakes. Suffice it to say that Dennett's "if-you-disagree-you-have-misunderstood" stance, while conceivably true of some critics, does not appear true of all.

Is a "virtual serial machine" necessary to get a one-after-another temporal ordering? Not at all. For example, it has been well known for at least eight years that neural nets with recurrent loops can yield temporal sequencing, and do so very economically and elegantly.[25] For a recent example, beautiful work in using "real-valued genetic algorithms to evolve continuous-time recurrent neural networks capable of sequential behavior and learning" has been done by Randall Beer and other sequencing work has been done by Michael Mozer.[26] Clearly, sequencing tasks per se do not imply the existence of a simulated serial machine. [27]

Is a virtual serial machine necessary to get rule-following behavior as seen in linguistic performance? Not at all. Again, as Elman and his colleagues have shown, recurrent neural nets can manage this very well.[28] Is a virtual serial machine needed to restrict a certain class of operations to one at a time? Not at all. First, a special class of operations could be the output of one network, albeit a widely distributed network. Second, they could be the output of a winner-take-all interaction between nets.[29] And there are lots of other architectures for accomplishing this. The motor system probably functions thus, but there is no reason to think it simulates a serial machine. [30]

Third, should we assume that consciousness involves only one operation at a time? Almost certainly not. Granting that the attentional capacity is much smaller than the extra-attentive capacity to represent[31], why conclude that we can attend to only one thing at a time? When I look at a bowl of colored M&M's, can I see more than one M&M at once? Probably. Fourth, is the serial machine simulation necessary in order to enable recursive properties, such that one can be self aware (think about what one just said to oneself)? Not at all. Recurrent neural nets are powerful enough and complex enough to manage this very nicely. Indeed, recurrence probably is a key feature of various self-monitoring subsystems in the nervous system, including thermoregulation. Is there any rationale for saying, "when we are conscious the brain must simulating a serial machine?" I see none.[32] This does not entail that Dennett must be wrong, but only that there is no reason to think he is on the right track.

E. The Problem is Beyond our Feeble Intelligence

Initially, this claim appears to be a modest acknowledgment of our limitations (Colin McGinn 1990). In fact, it is a powerful prediction based not on solid evidence, but on profound ignorance. For all we can be sure now, the prediction might be correct, but equally, it might very well be false. How feeble is our intelligence? How difficult is the problem? How could you possibly know that solving the problem beyond our reach, no matter how science and technology develop? Inasmuch as it is not known that the brain is more complicated than it is smart, giving up on the attempt to find out how it works would be disappointing. On the contrary, as long as experiments continue to produce results that contribute to our understanding, why not keep going? [33]

III Tracking Down The Neural Mechanisms of Consciousness

In neuroscience there are many data at higher levels relevant to consciousness. Blindsight, hemineglect, split brains, anosognosia (unawareness of deficit), for starters, are powerful constraints to guide theoretical reflection. Careful studies using scanning devices such as magnetic resonance imagining (MRI) and positron emission tomography (PET) have allowed us to link specific kinds of functional losses with particular brain regions.[34] This helps narrow the range of structures we consider selecting for preliminary micro exploration. [35]

For example, the hippocampus might have seemed a likely candidate for a central role in consciousness because it is a region of tremendous convergence of fibres from diverse areas in the brain. We now know, however, that bilateral loss of the hippocampus, though it impairs the capacity to learn new things, does not entail loss of consciousness. At this stage, ruling something out is itself a valuable advance. We also know that certain brain stem structures such as the locus coereleus (LC) are indirectly necessary, but are not part of the mechanism for consciousness. LC does play a nonspecific role in arousal, but not a specific role in awareness of particular contents, such as awareness at a moment of the color of the morning sky rather than the sound of the lawn sprinklers. The data may be fascinating in its own right, but the question remains: how can we get from an array of intriguing data to genuine explanations of the basic mechanism? How can we get started?

In thinking about this problem, I have been greatly influenced by Francis Crick. His basic approach is straightforward: if we are going to solve the problem, we should treat it as a scientific problem to be tackled in much the way we tackle other difficult scientific problems. As with any scientific mystery, what we want is a revealing experimental entry. We want to find a thread which, when pulled, will unloose a whole lot else. To achieve that, we need to devise testable hypotheses that can connect macro effects with micro dynamics.

Boiled down, what we face is a constraint satisfaction problem: find psychological phenomena that (a) have been reasonably well studied by experimental psychology, (b) are circumscribed by lesion data from human patients as well as data from precise animal microlesions, (c) are known to be related to brain regions where good neuroanatomy and neurophysiology has been done and (d) where we know quite a lot about connectivity to other brain regions. The working assumption is that if a person is aware of a stimulus, his brain will be different in some discoverable respect from the condition where he is awake and attentive but unaware of the stimulus. A auspicious strategy is to hunt down those differences, guided by data from lesion studies, PET scans, magnetoencephalograph (MEG) studies, and so forth. Discovery of those differences, in the context of neurobiological data generally, should aid discovering a theory of the mechanism.

The central idea is to generate a theory constrained by data at many levels of brain organization -- sufficiently constrained so that it can be put to meaningful tests. Ultimately a theory of consciousness will need to encompass a range of processes involved in awareness, including attention and short term memory. Initially, however, it may target a subset, such as integration across space and across time. Whether the theory falls to falsifying evidence or whether it survives tough tests, we shall learn something. That is, either we shall have ruled out specific possibilities -- a fine prize in the early stages of understanding -- or we can go on deepen and develop the theory further -- an even finer prize. In any case, the trick is to generate testable, meaty hypotheses rather than loose, frothy hypotheses susceptible only to experiments of fancy. The trick is to make some real progress.

B. Visual Awareness

What plausible candidates surface from applying the constraint satisfaction procedure? Interestingly, the choices are quite limited. Although metacognition, introspection, and awareness of emotions, for example, are indeed aspects of consciousness, either we do not have good lesion data to narrow the search space of relevant brain regions, or the supporting psychophysics is limited or both. Consequently, these processes are best put on the back burner for later study.

Visual awareness, by contrast, is a more promising candidate. In the case of vision, as Crick points out, there is a huge literature in visual psychophysics to draw upon, there is a rich literature of human and animal lesion studies, and relative to the rest of the brain, a lot is known about the neuroanatomy and neurophysiology of the visual system, at least in the monkey and the cat. Visual phenomena such as filling in, binocular rivalry, seeing motion, seeing stereoptic depth, and so forth might reward the search for the neurobiological differences between being aware and not being aware in the awake, attentive animal. This may get us started, and I do emphasize started.

1. The Crick Hypothesis

Immersed in the rich context of multi-level detail, Crick has sketched an hypothesis concerning the neuronal structures he conjectures make the salient differences, depending on whether the animal is or is not visually aware of the stimulus. [36] Integration of representations across spatially distributed neural networks -- the unity in apperception, so to speak -- is thought be accomplished by temporal 'binding', namely synchrony in the output responses of the relevant neurons. Very crudely, Crick's suggestion is that (1) for sensory awareness, such as visual awareness, the early corteces are pivotal (e.g. visual areas V2, V3, V5). This makes sense of lesion data, as well as recent PET data (Damasio, Grabowski et al. 1993; Kosslyn et al. 1993) and single cell data (Logothetis and Schall 1989). (2) Within the early sensory cortical areas, pyramidal cells in layer 5 and possibly layer 6, play the key role.

What good is this idea ? Part of its appeal is its foothold in basic structure. In biology, the solution to difficult problems about mechanism can be greatly facilitated by identification of critical structures. Crudely, if you know "what", it helps enormously in figuring out "how". On its own, the Crick hypothesis can be only a small piece of the puzzle. If we are lucky, however, it, or something like it, may be a key piece of the puzzle. This is not the time for a fuller discussion of this hypothesis. Suffice it to say that true or false, the Crick hypothesis provides a bold illustration of how to approach a problem so tricky it is often scrapped as unapproachable.

2. The Llinas Hypothesis

Another promising entry route is suggested by the differences -- phenomenological and neurobiological -- between sleep/dreaming/wakeness[37] (SDW) states. [38] This entry point is attractive first because there is the familiar and dramatic loss of awareness in deep sleep, which is recovered as we awake, and is probably present also during dreaming. The phenomenon is highly available in lots of different subjects and across many species. Second, MEG and EEG techniques reveal global brain features characteristic of different states. Human and animal lesion data are important, especially as they concern deficits in awareness during wakeness. Here again I note the significance of research on blindsight, hemineglect (tendency to be unaware of stimuli in various modalities on the left side of the body), simultanagnosia (inability to see several things simultaneously), anosognosia (unawareness of deficits such as paralysis) blindness denial, unawareness of jargon aphasia (of not making sense) and so forth.

Third, we have learned a great deal from abnormalities in and manipulation of the SDW cycle and the link to specific brain properties. Fourth, some of the global changes in state in the SDW cycle seen by macro techniques have been linked by micro techniques to interactions between specific circuits in the cortex and subcortical circuits, especially circuits in several key structures in the thalamus. Fifth, and more specifically, MEG data reveal an robust 40 Herz wave form during wakeness and dreaming.[39] The definition and amplitude is much attenuated during sleep, and the amplitude is modulated during wakeness and dreaming. Analysis of the wave from by MEG reveals it to be a traveling wave, moving in the anterior to posterior direction in the brain, covering the distance in about 12 to 13 milliseconds. Cellular data suggest that these dynamical properties emerge from particular neural circuits and their dynamical properties.

What does all this add up to? Based on these data, and mindful of the various high-level data, Rodolfo Llinas and colleagues (1991; 1993) have hypothesized that the fundamental organization subserving consciousness and the shifts seen in the SDW pattern are pairs of coupled oscillators, each of which connect thalamus and cortex, but each connects distinct cell populations via its own distinctive connectivity style. (Figure 2) One oscillator 'family' connects neurons in a thalamic structure known as the intralaminar nuclei, a bagel-shaped constellation of structures whose neurons reach to the upper layers of cortex to provide a highly regular fan-like coverage of the entire cortical mantle. The other oscillator 'family' connects neurons in thalamic nuclei for modality-specific information (MS nuclei) originating for example, in the retina or the cochlea, with modality specialized cortical areas (e.g. V2, S2). During deep sleep, the intralaminar neurons projecting to cortex cease their 40 Hz behavior. During deep sleep and dreaming, external signals to the cortex are gated by the reticular nucleus of the thalamus. Joseph Bogen (1993) also hypothesis a crucial role for intralaminar structures, noting especially the wide fan out from those nuclei to cortex, and the heavy connection to striatum.

Ever so crudely, the idea is that the second oscillator 'family' provides the content (visual, somatosensory etc. ) while the first provides the integrating context. In deep sleep the oscillators are decoupled; in dreaming they are coupled but the MS oscillating circuit is largely nonresponsive to external signals from the periphery; in wakeness, the oscillators are coupled, and the MS circuit is responsive to external signals.

What are the effects in humans of lesions to the intralaminar thalamic structures (bagel)? First, it is highly unlikely that a lesion should occur exclusively to the intralaminar nuclei, with sparing of other thalamic structures. Bearing that caution in mind, I note that the main result of small unilateral lesions believed to reside mainly in the intralaminar nuclei, is neglect (unawareness) of all stimuli originating the opposite body side. (Watson and Heilman 1979; Watson, Valenstein and Heilman 1981). Bilateral lesions apparently result in general unarousability, meaning roughly that the patient initiates no behavior and responds very poorly to sensory stimuli or questions (Castaigne et al. 1981; Guberman and Stuss 1983). The few animal studies that exist are consistent with the human data (Henderson, Alexander and Nalser 1982; Watson, Miller and Heilman 1978). Bogen (1993) also points out that the human lesion data, together with the connectivity patterns, makes the intralaminar nuclei a candidate for "the where of awareness" . These data are important and provide a useful starting point, but further studies, especially using functional MRI to locate areas of relative low and high activity, are desirable. Lesions to modality-specific regions of the thalamus, by contrast, lead to modality specific losses in awareness -- visual awareness, for example will be lost, but awareness of sounds, touches etc. can be normal. Intriguingly, the MEGs of Alzheimer's patients who have degenerated to a state of inanition show a dilapidated 40 Hz wave form when it exists at all. Obviously these data are not decisive, but at least they are consistent with the hypothesis.

Do the Llinas hypothesis and the Crick hypothesis fit together ? Minimally, they are consistent. Additionally, they are mutually supporting at the neuron and network levels. One encouraging point is this: the two families of oscillators (MS and intralaminar) richly connect to each other mainly in cortical layer 5 (Figure 2). From what we can tell now, those connections seem to be the chief means whereby the oscillators are coupled. The possibility entertained here is that the temporal synchrony Crick hypothesizes in neurons carrying signals about external stimuli may be orchestrated by the intralaminar-cortical circuit. Connections between brain stem structures and the intralaminar nucleus could have a role in modulating arousal and alertness.

Figure 2 about here

Many questions now suggest themselves. For example, how do the pivotal structures for awareness interface with behavior? More specifically, what are the connections between the intralaminar nuclei and motor structures, and between layer 5 of sensory corteces and motor structures; do the projections from the intralaminar nuclei to the cingulate cortex have a role in attention? These are questions motivated by independent data. Convergence of hypotheses is of course encouraging, but it is well to remember that it can also encourage us down the proverbial garden path. Wisdom counsels guarded optimism.

But is something missing here ? Probably. As Kant might have said to Hume, the brain will not produce awareness unless the nervous system also generates a representation of self -- a representation which carries what we would call "a point of view". And this is indeed precisely the hypothesis tendered by Antonio Damasio (1994). According to the Damasio perspective, the neurobiological mechanisms for visual awareness, for example, are essentially interconnected with the mechanisms for representing oneself as a thing that has experiences, that feels, remembers and plans; as a thing occupying space and enduring through time. To suppose that visual awareness can be understood independently of the self-representation is like supposing evolution can be understood independently of environment.

Damasio's ideas on this score have emerged from many years of observing brain damaged patients, and reflecting on the ways in which awareness is related to self-representation and how that in turn is related to body-representation (For the details of his hypothesis, see his book, Descartes' Error 1994). Against a backdrop of basic neuroanatomy and neurophysiology, Damasio sees representational complexity and interdependence as key elements in explaining consciousness. Crudely, synchrony of firing in certain visual circuits may be a necessary condition for visual awareness, but a sufficient condition it surely is not. Constructing a plausible hypothesis to capture a necessary condition is hard enough; identifying the additional neural conditions that jointly are necessary and sufficient will be harder still. Even so, Damasio's central idea is both powerful and reasonable: body-representation, which systematically integrates bodily-stimulation and body-state information, provides a scaffolding for self-representation, and self-representation is the anchor point for awareness -- modality specific and otherwise. Moreover the adaptive advantages of an integrated body-representation in life are pretty obvious (see again Damasio 1994). One needs only to think of subdominant vervet monkey males sneaking around at night to mate with females while the alpha male dozes, to realize the importance of "knowing which way is up", to put knowledge of one's body-self in its most general form.

Whereas Kant, a dyed-in-the-wool anti-reductionist, was convinced that the nature of the self was forever unresearchable empirically, Damasio finds places where scientific progress is possible; whereas Kant thought of the self in terms of a highly mysterious "transcendental unity of apperception", Damasio gives it a reassuringly concrete base in terms of neural representation of the body: the skin, muscles, joints, viscera, and so forth. Like other brain representations, the nature of body-representation is researchable with the combined techniques of neuropsychology, neurobiology, and neural network modeling. And if Damasio is right, then the neural mechanisms of self-representation are researchable too.

The three broadly neuroscientific approaches -- that of Crick, Llinas, and Damasio -- I see as complementary strategies for addressing different but overlapping segments of a large and puzzling problem. Each has particular strengths viz-a-viz some aspect of the problem, and each presents questions and challenges for the other -- the very paragon co-evolutionary discovery.

IV CONCLUDING REMARKS

References

Beer, R. D. (1995b). On the dynamics of small continuous-time recurrent neural networks. Adaptive Behavior 3:471-511.

Beer, R. D. and J. C. Gallagher (1992). Evolving dynamical neural networks for adaptive behavior. Adaptive Behavior 1:91-122.

Berthier, N. E. , S. P. Singh, A. G. Barto, and J. C. Houk (1992). A cortico-cerebellar model that learns to generate distributed motor commands to control a kinematic arm. In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 611-618.

Bickle, J. (1992). "Revisionary physicalism." Biology and Philosophy. 7: 411-430.

Block, Ned (1993). "Consciousness Ignored? Review of D.C. Dennett's Consciousness Explained." Journal of Philosophy , 90;4: 83-91.

Bogen, J. E. (1993). "Intralaminar nuclei and the where of awareness." Society for Neuroscience Abstracts.

Castaigne, P., F. Lhermitte. A. Buge, R. Escourolle, J. J. Hauw and O. Lyon-caen (1981). Paramedian thalamic and midbrain infarcts: Clinical and neuropathological study. Annals of Neurology. 10: 127-148.

Churchland, P. M. (1988). Matter and Consciousness, 2nd Edition.Cambridge, Mass.: MIT Press.

Churchland, P. M. (1993) "Evaluating our self conception". Mind and Language. 8:211-222.

Churchland, P. M.. (1994) "Betty Crocker's theory of the mind: A review of The Rediscovery of the Mind, by John Searle." London Review of Books, May, 1994.

Churchland, P. M. and P. S. Churchland (1990). "Intertheoretic reduction: A neuroscientist's field guide." Seminars in the Neurosciences. 4: 249-256.

Churchland, P. S. (1986). Neurophilosophy. Cambridge, Mass.: MIT Press.

Churchland, P. S. (1987). "Replies to Comments. Symposium on Patricia Smith Churchland's Neurophilosophy. " Inquiry. 29: 241-72.

Churchland, P. S. (1988). "Reduction and the neurobiological basis of consciousness." In: Consciousness in Contemporary Science. Ed. A. J. Marcel and E. Bisiach. 273-304.

Churchland, P. S. (1994). "Can neurobiology teach us anything about consciousness?". Presidential Address to the American Philosophical Associatiojn, Pacific Division. In: Proceedings and Addresses of the American Philosophical Association. Lancaster, PA: Lancaster Press. 67-4: 23-40.

Churchland, P. S. and T. J. Sejnowski (1989). "Brain and cognition". In: Foundations of Cognitive Science. Ed. M. Posner. Cambridge, Mass.: MIT Press. 245-300.

Churchland, P. S. and T. J. Sejnowski (1992). The Computational Brain. Cambridge, Mass.: MIT Press.

Churchland, P. S. and V. S. Ramachandran (1993). Filling-in: Why Dennett is wrong. In: Dennett and his Critics. Ed by B. Dahlbom. Blackwells:Oxford. 28-52.

Crick, F. H. C. (1994). The Astonishing Hypothesis. New York: Scribner's Sons.

Crick, F. H. C. and C. Koch (1990). "Towards a neurobiological theory of consciousness." Seminars in the Neurosciences. 4: 263-276.

Damasio, A. R. (1994). Descartes' Error. New York: Putnam.

Damasio, H. (1991) "Neuroanatomy of frontal lobe in vivo: A comment on methodology". In: Frontal Lobe Function and Dysfunction. Ed. H. Levin, H. Eisenberg, and A. Benton. New York: Oxford University Press. 92-121.

Damasio, H. and A. R. Damasio (1990). The neural basis of memory, language and behavioral guidance: advances with the lesion method in humans. Seminars in the Neurosciences. 4: 277-286.

Damasio, H. , T. J. Grabowski et al. (1993). "Visual recall with eyes closed and covered activated early visual cortices. " Society for Neuroscience Abstracts (658.4)

Dennett, D. C. (1991). Consciousness Explained. Boston: Little, Brown and Co.

Elman, J. L. (1991). Distributed representations, simplle recurrent networks, and grammatical structure. Machine Leaning. 7: 195-225..

Farah, M. J. (1994). "Neuropsychological inference with an interactive brain: A critique of the 'locality assumption'". Behavioral and Brain Sciences. 17: 43-104.

Feyerabend, P. K. (1981). Philosophical Papers. Vols. 1 and 2. Cambridge: Cambridge University Press.

Fodor, J. (1990). A Theory of Content and Other Essays. Cambridge,MA: MIT Press.

Flanagan, O. (1992) Consciousness Reconsidered. Cambridge, Mass.: MIT Press.

Flanagan, O. (forthcoming) "Prospects for a unified theory of consciousness, or, what dreams are made of". In: Scientific Approaches to the Question of Consciousness: 25th Carnegie Symposium on Cognition. Ed. J. Cohen and J. Schooler. Hillsdale, N. J.: L. Erlbaum.

Grush, Rick (unpublished). Emulation and Cognition. Ph. D. dissertation for UCSD.

Grush, Rick and P. S. Churchland (1995). "Gaps in Penrose's toilings." Journal of Consciousness Studies. 2:10-29.

Giles, C. L., C. B. Miller, D. Chen, G. Z. Sun, H. H. Chen, and Y. C. Lee (1992). Extracting and learning and unknown grammar with recurrent neural nets. In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 317-324.

Guberman, A. and D. Stuss. (1983). The syndrome of bilateral paramedian thalamic infarction. Neurology 33: 540-546.

Henderson, V. W., M. P. Alexander, and M. A. Nalser (1982). Right thalamic injury, impaired visuospatial perception and alexia. Neurology. 32: 235-240.

Jain, A. N. (1992). Generalizing performance in PARSEC - A structured connectionist parsing architecture. In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 209-216

Kosslyn, S. M. , N. M. Alpert, W. L. Thompson, V. Maljkovic, S. B. Weise, C. F. Chabris, S. E. Hamilton, S. L. Rauch, and F. S. Buoanno. (1993)"Visual mental imagery activated topographically organized visual cortex: PET investigations." Journal of Cognitive Neuroscience. 5: 263-287.

Lange, T. E. (1992). Dynamically-adaptive winner-take-all networks. In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 341-348.

Llinas R. R. and D. Pare (1991). "Of dreaming and wakefulness." Neuroscience 44: 521-535.

Llinas, R. R. and U. Ribary (1993). Coherent 40-Hz oscillation characterizes dream state in humans. Proceedings of the National Academy of Sciences. 90: 2078-2081.

Logothetis, N. and J. D. Schall. (1989). Neural correlates of subjective visual perception. Science. 245: 753-761.

Lycan W. G. (1987). Consciousness. Cambridge, Mass.: MIT Press.

McGinn, C. (1990). The Problem of Consciousness. Oxford: Blackwells

Mozer, M. C. (1992). Induction of multiscale temporal structure. In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 275-282.

Mozer, M. C. and J. Bachrach (1991). SLUG: A connectionist architecture for inferring the structure of fine-state environments. Machine Learning. 7: 139-160.

Penrose, Roger (1994). Shadows of the Mind. Oxford: Oxford University Press.

Pinkas, G. (1992). Constructing proofs in symmetric networks. (1992). In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 217-224.

Pollack, J. B. (1991). The induction of dynamical recognizers. Machine Learning. 7: 227-252.

Schaffner, K. F. (1993). "Theory structure, reduction, and disciplinary integration in biology. Biology and Philosophy. 8: 319-348.

Searle, J. R. (1992). The Rediscovery of the Mind. Cambridge, Mass.: MIT Press.

Singh, S. P. (1992) The efficient learning of multiple task sequences. In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 251-258.

Sumida, R. A. and M. G. Dyer (1992). Propagation filters in PDS networks for sequencing and ambiguity resolution. In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 233-240.

Sutton, J. P., A. N. Mamelak, and J. A. Hobson (1992). Network model of state-dependent sequencing. In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 283-290.

Vergehsi, P and D. G. Pelli (1992). The information capacity of visual attention. Vision Research. 32: 983-995.

Viola, P. A. , S. G. Lisberger, and T. J. Sejnowski (1992). Recurrent eye tracking network using a distributed representation of image motion. In: Neural Information Processing Systems 4. Ed. by J. E. Moody, S. J. Hanson, and R. P. Lippmann. San Mateo CA: Morgan-Kaufman. 380-387.

Watson, R. T. and K. M. Heilman (1979). Thalamic neglect. Neurology 29: 690-694.

Watson, R. T., B. D. Miller, and K. M. Heilman (1978). Nonsensory neglect. Annals of Neurology. 3: 505-508.

Watson, R. T., E. Valenstein, and K. M. Heilman (1981). Thalamic neglect: Possible role of the medial thalamus and nucleus reticularis in behavior. Archives of Neurology. 38: 501- 506.

Figure Captions

Figure 1

Schematic illustration of levels of organization in the nervous system. The spatial scales at which anatomical organization can be identified varies over many orders of magnitude. Icon to the left depicts the "neuron man", showing the brain, spinal cord, and peripheral nerves. Icons to the right represent structures at distinct levels: (top) a subset of visual areas in visual cortex; (middle) a network model proposing how ganglion cells could be connected to "simple" cells in visual cortex; and (bottom) a chemical synapse. (From Churchland and Sejnowski 1992)

Figure 2

Schematic diagram of the circuits between the thalamus and the cerebral cortex proposed to serve temporal binding. (A) Diagram of two different types of circuit connecting thalamus and cortex. On the left, specific sensory nuclei or motor nuclei of the thalamus project to Layer IV of cortex, producing cortical oscillation by direct activation and feed-forward inhibition via 40Hz inhibitory interneurons. Collaterals of these projections produce thalamic feedback via the reticular nucleus (a kind of rind covering the thalamus). The return pathway projects back to specific and reticular nuclei via Layer VI cells. On the right, the second loop shows nonspecific intralaminar nuclei projecting to Layer I of cortex, and giving collaterals to the reticular nucleus. Layer V cells return oscillation to the reticular and the intralaminar nuclei, establishing a second resonant loop. The conjunction of the specific and nonspecific loops is proposed to generate temporal binding. Connectivity between the loops is seen chiefly in Layer V. (B) Schematic diagram showing the intralaminar nucleus as a circular neuronal mass (stipple shading). Other parts of the thalamus are shown in hatched shading. The intralaminar nucleus projects widely across the cortex, to Layer I. (From Llinas and Ribary 1993)

*. This paper is based on my Presidential Address to the American Philosophical Association, Pacific Division, March, 1993, published in Proceedings and Adresses of the APA (1994). I am indebted to many people for discussion and criticism: Francis Crick, Rodolfo Llinas, Antonio Damasio, Hanna Damasio, Joe Bogen, Ramachandran, Robert Van Gulick, Owen Flanagan, Dan Lloyd, and especially Paul Churchland.

1. See our discussion in The Computational Brain, Churchland and Sejnowski (1992).

2. For concordant opinions, see also Francis Crick (1994); Paul Churchland (1989) Daniel Dennett (1991); Owen Flanagan (1992); William G. Lycan (1987); John Searle (1993).

3. As P. S. Churchland and T. J. Sejnowski argued in (1989)

4. For an outstanding discussion of reductionism that includes many of the complexities I am not worrying too much about here, see Schaffner (1993).

5. P. S. Churchland (1986), Neurophilosophy.

6. See Paul Churchland's characterization and defense of this view in (1981), reprinted in P. M. Churchland 1989.

7. See P. S. Churchland, Neurophilosophy (1986)

8. Jerry Fodor (1990).

9. John Searle (1992)

10. Op .cit.

11. See Churchland and Sejnowski (1992); Paul M. Churchland (1989)

12. Or, as we have preferred but decided not to say "what makes revisionary materialism revisionary" (P. S. Churchland 1987). See also P. M. Churchland (1993). For a related but somewhat different picture, see Bickle (1992).

13. Ibid. See also P. M. Churchland and P. S. Churchland (1990).

14. For example, Colin McGinn (1990).

15. See for example, Feyerabend (1981).

16. See also Owen Flanagan (forthcoming).

17. In the following discussion, the ideas are mostly owed to Paul Churchland (1994) . For his discussion, see "Betty Crocker's Theory of the Mind: A Review of John Searle's The Rediscovery of the Mind. The London Review of Books .

18. Paul Churchland made this discovery in our kitchen about eight years ago. It seemed to us a bang-up case of someone's not really understanding the scientific explanation. Instead of thinking the thermodynamic theory through, Betty Crocker just clumsily grafts it onto on old conception as though the old conception needed no modification. Someone who thought electricity was caused by moving electrons would tell a comparable Betty Crocker story: "voltage forces the electrons to move through the wire, and as they do so, they cause static electricity to build up, and a sparks then jump from electron to electron, on down the wire."

19. See P. M. Churchland (1993).

20. D. C. Dennett, Consciousness Explained (1992).

21. For a criticism of Dennett on filling in, See P. S. Churchland and Ramachandran (1993).

22. Paul Churchland ,The Engine of Reason; The Seat of the Soul (1995).

23. See especially Owen Flanagan (1992) and Ned Block (1993).

24. Ibid. Also John Searle The Rediscovery of the Mind (1992).

25. See for example, Singh (1992); Mozer (1992); Sutton, Mamelak and Hobson (1992).

26. Beer and Gallagher (1992); Beer (1995 a), Beer (1995 b).

27. See our discussion of various nets in Churchland and Sejnowski (1992).

28. Elman (1991). See also Mozer and Bachrach (1991; Pollack (1991); Giles et al. (1992), Jain (1992); Pinkas (1992); Sumida and Dyer (1992).

29. Lange (1992).

30. E.g. Viola, Lisberger and Sejnowski (1992); Berthier, Singh, Barto and Houk (1992).

31. Verghese and Pelli (1992) conclude from their studies that the capacity of the attentional mechanism is limited to about 44 15 bits per glimpse. They calculated the pre-attentive capacity to be much greater -- about 2106 bits. Their data on the attentional capacity is consistent with paying attention to (and being aware of ) more than one thing at a time.

32. Rick Grush has a brilliant exploration of the idea that brains emulate aspects of the world, including the body, in his PhD. dissertation, Emulation and Cognition. For an approach to the problem using recurrent nets, see also Dan Lloyd, "Consciousness: A Connectionist Manifesto" (unpublished).

33. See Owen Flanagan's convincing and more detailed discussion of McGinn's naysaying (Flanagan 1992).

34. See especially H. Damasio and A. R. Damasio (1990); H. Damasio (1991); A. R. Damasio (forthcoming); Farah (1994).

35.One possibility explored by Penrose (1994) is that consciousness is a quantum mechanical phenomenon, produced in organelle structures called microtubules. For a criticism of that hypothesis, see Grush and Churchland (1995).

36. This point is made in Crick and Koch (1990) and in Crick (1994).

37. 'Wakeness' is a bit of a neologism. I prefer not to use 'wakefulness' inasmuch as it connotes some difficulty in sleeping, or being easily awoken, as in "the baby was wakeful when he had a rash." I prefer it also to the phrase 'being awake' which is both rather cumbersome and fails to preserve parallism with 'dreaming', 'sleeping'. Clealry one cannot use 'waking' as synonymous with 'being awake' . It is a quirk of English that methought to remedy by 'wakeness'.

38. See also my discussion in P. S. Churchland (1988)

39. See Llinas and Pare (1991)