The Neuroscience of Consciousness [A paper on the neuroscience of consciousness delivered at the ABS conference, San Francisco, CA, August 12, 2006] [slide 1: title] [slide 2: overview section 1. Some Historic Background section 2. An Example of Reductionistic Explanation of Behavior section 3. The Functionalist Approach to Consciousness section 4. Consciousness as Subjective Experience section 5. The Neurobiological Approach to Consciousness: Is There an NCC? section 6. Combining the Functionalist and Neurobiological Approaches section 7. Meaning and the Importance of the Self section 8. Broader Implications] 1. Some Historic Background Until the early 20th century, classical mechanics, as first formulated by Newton and further developed by Laplace and others, was seen as the foundation for science as a whole. It was expected that the observations made by other sciences would sooner or later be reduced to the laws of mechanics. Although that never happened, other disciplines, such as biology, psychology, or economics, did adopt a generally mechanistic and reductionistic methodology and worldview. This influence was so great that even today most people with only a basic notion of science still implicitly equate “scientific thinking” with “Newtonian thinking”. The logic behind this hegemony of Newtonian science is easy to formulate. Its central principle is that of analysis or reduction: to understand any complex phenomenon, you need to take it apart, and reduce it to its individual components. If these are still complex, you need to take your analysis one step further and look at their components. If you continue this subdivision long enough, you end up with the smallest possible parts, the atoms or elementary particles. Atoms are thought of as separate pieces of the same hard, permanent stuff, called “matter”. Newtonian ontology is therefore is materialistic. It assumes that all phenomena, whether physical, biological, mental or social, are ultimately just complex arrangements of bits of matter. Any change, development or evolution is merely a structural rearrangement caused by the movement of the particles. This movement is governed by deterministic laws of nature. So if you know the initial positions and velocities of the particles constituting a system, together with the forces acting on those particles, you can in principle predict the further evolution of the system with complete certainty and accuracy. The trajectory of the system is predictable backwards, too: given its present state, you can in principle reconstruct any earlier state it’s gone through. 1 The elements of the Newtonian ontology are matter, the absolute space and time in which bits of matter move, and the forces or natural laws that govern their movement. No other fundamental categories of being, such as consciousness, mind, life, purpose, moral facts or esthetic facts, are acknowledged. Or, if they are acknowledged to exist, it is only as causally-irrelevant epiphenomena -- transient arrangements of particles in space and time. [slide 3: Epiphenomenalism] If physicalism is true, consciousness may well be something “epiphenomenal”, like the shadow cast by a moving car. The shadow is a real phenomenon, but it has no causal powers of its own. Newtonian epistemology is based on the correspondence view of truth. Knowledge is merely an imperfect reflection of the particular arrangements of matter outside of us. The task of science is to make the mapping or correspondence between the external, material objects and the internal representations of them (which are also ultimately just arrangements of bits of matter) as accurate as possible. A basic assumption of the Newtonian outlook is simplicity. The seeming complexity of the world is only apparent. To deal with it you need to analyze phenomena into their simplest components. Once you’ve done that, their evolution will turn out to be perfectly regular and predictable, and the knowledge you gain will be both a reflection and an instance of that pre-existing order. My purpose in giving this brief caricature of the assumptions behind Newtonian science in the early modern period is to suggest that while the science of physics itself has undergone deep changes, the methodological assumptions behind early modern science have continued to be influential to the present day. In biology, for example, physicalism and reductionism are integral to the genetic determinism that characterizes some interpretations of evolutionary theory. In the field of psychology, physicalism and reductionism had to contend with the obvious fact of our own conscious experience and our special introspective access to it, as well as with religious beliefs about the soul held by the majority of people in most societies. Modern psychology only began in the 19th century, and for the first half century or so, consciousness was a central topic. Psychophysicists like Ernst Weber (1795-1878) and Gustav Fechner (1801-1887) studied the relationship between physical stimuli and reportable sensations, relating conscious sensation to the intensity of the stimulus. Wilhelm Wundt (1832-1920), usually credited with founding the first laboratory of experimental psychology, based his studies on introspective reports of people’s conscious experiences. He and his student Titchener (1867-1927) tried to train people to make very precise, reliable observations of their own inner experience, so that a science of consciousness could be built on the basis of these atoms of experience. William James’s classic work, Principles of Psychology (1980), still widely read today, was all about trying to understand conscious experience. And, though it seems hard to believe in our 2 current climate, in late 19th c. philosophical circles, idealism (roughly the view that reality is ultimately conscious or mental) was a dominant force. [Josiah Royce. Hegel. Bradley] But all this changed in the early 20th c., when psychology, too, came under the spell of reductionistic physicalism. In 1913 the American psychologist John B. Watson argued that psychology did not need introspection-based methods, and in fact could do without the concept of consciousness altogether. [1913 article Psych. Review.] He set out to establish psychology as a purely objective branch of the natural sciences, with the straightforward theoretical goal of predicting and controlling human behavior. The work of Pavlov and Watson on classical conditioning and Skinner’s work on operant conditioning became the paradigms in psychology. Skinner was convinced that consciousness is epiphenomenal, and that its study shouldn’t be part of psychology at all. This was the general mood for many decades while behaviorism reigned in university psychology departments. By the 1960s behaviorism began to lose its influence, and cognitive psychology, with its emphasis on internal representations and information processing, became the dominant school of thought. But even within cognitive psychology, consciousness was a taboo subject for many years. This neglect of consciousness is now fading rapidly. After almost a century, a series of significant papers began to appear in the 1980s, in leading journals like Science and Nature, reporting marked progress in understanding conscious vision in the cortex, conscious memories mediated by the hippocampus, and other cases where conscious events can be compared to otherwise similar but unconscious ones. Since that time, thousands of articles on consciousness have appeared in the neuroscience and psychological literature, under various headings. There are new journals, new academic conferences, and new graduate programs, all devoted to the science of consciousness. There is little doubt that we are again looking squarely at the questions that were familiar and important to William James and his generation, but now with better evidence, better technologies and perhaps better theory than ever before. The science of consciousness is in its infancy. Seemingly intractable problems face us. But this is one of the things that makes the topic of consciousness so exciting. Each individual discovery in the neuroscience of consciousness is fascinating in itself, and tremendously important for medical reasons. But in addition to that, the effort to reach a scientific understanding of consciousness is forcing us to reconsider some of the fundamental assumptions about the universe that have been with us for centuries, since the birth of modern science. 2. An Example of Reductionistic Explanation of Behavior. I want to begin with a very simple introduction to some aspects of our current understanding of how the mind works. Looking around the universe, we notice that consciousness appears where there are minds: i.e. where there are living organisms with brains capable of producing intelligent behavior. Consciousness may also be a fundamental and therefore widespread property in the universe, as one of our speakers 3 today, Dr. Hoffman, will suggest. But organisms capable of intelligent behavior seems like the right place to begin. How does a living organism have mental capacities and conscious mental states? The answer to this question is not obvious from simple anatomy. If you open the skull and look at a brain, here’s what you’ll see. [slide 4: The Brain] Looking at a brain doesn’t give you any answers. It’s not that different from looking at an exposed heart or a kidney. In fact for many centuries the best minds of Europe debated whether the seat of consciousness was in the brain or the heart. (The heart seemed like a good candidate because it was centrally located, and connected by blood vessels to all parts of the body.) It’s not at all obvious how a piece of a body (a piece of meat, if we want to put it very crudely) can give rise to conscious experience! Let’s start with a very simple example. What’s the scientific explanation of intelligent behavior in, for example, a mouse? Does the explanation of intelligent behavior include a satisfying explanation of consciousness? [slide 5: The Morris Water Maze (Richard G. Morris. 1984)] This is a diagram of a Morris water maze. Mice (or rats) are placed in a circular arena about 6’ in diameter filled with cloudy water. The arena itself is featureless, but the surrounding environment contains positional cues of various types. A small platform is located below the surface of the water. As the mice search for this resting place, the pattern of their swimming is monitored by a video camera mounted above. After 10 trials, normal mice swim directly to the platform on each trial. Researchers wanted to know whether the mice had formed a spatial map, a representation of the spatial organization of the maze relative to the room the maze is in, or merely acquired a conditioned response. [slide 6: Which is learned?] To answer this question the experimenters next tried releasing the mice from a different location in the maze. The mice were able to swim directly to the platform from the new location on the first trial, indicating that they had formed a spatial map of the location of the platform relative to environmental cues. Mice that are missing a part of the brain called the “hippocampus” behave differently. They do eventually learn to swim to the platform directly when released, but it takes them hundreds of trials (typical of conditioning) to learn this. Then, once they’ve learned by conditioning to swim directly to the platform from the first location, if they’re released from a new location, it takes them 100’s of trials again to learn to swim to the platform from the new location. 4 [slide 7: Navigating Without a Hippocampus] So, we learn several things from these experiments: that normal mice learn spatial maps of their environments (i.e. they make representations in their brain of the spatial surroundings of individual events in their life), that this seems to require having a hippocampus, and that this is not a simple case of acquiring a skill by gradual conditioning, but is more like forming an autobiographical memory. What are the mechanisms underlying this kind of spatial learning in mice? O’Keefe and Dostrovsky discovered in 1971 that there are “place neurons” in the hippocampus of rats. These are individual neurons that fire when and only when the freely moving rat is in a certain place relative to a particular environment. (For example, a particular neuron in the rat’s hippocampus may code for a particular place in the kitchen, and a completely unrelated place in the livingroom. When the rat or mouse returns to the kitchen, no matter what direction he enters it from, the neuron codes for the identical location in the kitchen.) The hippocampus forms maplike memories of spatial locations, so that it can represent the location of objects and itself when it’s in those locations. The formation of these spatial maps is one of the mechanisms that enables maze-learning in rats and mice. But what explains how there can be place neurons that form spatial maps? What mechanism gives neurons the capacity to represent individual spatial locations? Various drugs have been administered before, during, or after maze training. Mice treated with an NMDA-receptor blocker, called APV, perform poorly on the Morris water maze, e.g., suggesting that NMDA receptors play a role in the tuning up of individual neurons. And since long-term potentiation, a biological mechanism for behavioral learning, also requires NMDA receptors, spatial tuning of individual neurons may require LTP. [There is some evidence that consolidation of spatial memory requires NMDA receptors, while retrieval of spatial memories requires AMPA receptors. Liang et al. 1994] [slide 8. LTP 1] This kind of information can be put together to give us the beginnings of a neuroscientific explanation of maze-learning in mice. [slide 9: reductionistic explanation] The basic methodology of cognitive neuroscience is reductionistic. We want to discover the neurobiological mechanisms for psychological capacities. We know that living organisms are made up of systems of cells of various types, and these are made up of molecules of certain types, and so on. Recognizing that an organism is a functional hierarchy of systems made up of other systems made up of other systems, we seek an explanation of top-level behavior by means of functional decomposition and neural localization. We first do a functional analysis of the mouse’s maze learning behavior in terms of the mouse’s homeostatic mechanisms for maintaining body temperature, which 5 lead to a goal state of avoiding submersion, which the mouse does by swimming to the platform, using the spatial map it forms in the hippocampus region of its brain. We study functional relations at the behavioral level, and then try to discover how these are realized in entities at the next level down. (And each of these is in turn realized by entities at an even lower level when these stand in certain structural/functional relations.) The goal is to understand the physical mechanisms that underlie psychological functions. Research will typically be carried on at several different levels at once. Neuroimaging techniques may show increased glucose and oxygen uptake in the hippocampus when the mouse is navigating; microanatomical studies of single neurons in the hippocampus may reveal the existence of place neurons and spatial maps formed by longterm potentiation (LTP); genetic changes that disrupt the production of certain proteins that are neurotransmitter precursors might be found to interfere with LTP; chemical changes in the bloodstream that lower the availability of Ca might be found to interfere with the activity of the NMDA receptors, and so on. A complete explanation of the animal’s behavior requires understanding the functional relations at each level, and how each of these is realized in relations among the entities of the next level down. When entities at the lower level are in a certain structural and functional relationship, they form a mechanism that realizes a function at the higher level. [slide 10. Terminology] (Let me say a word about terminology. There were several philosophical theories of the mind/brain relation in the period following the demise of behaviorism, none of them adequate. The method used in the cognitive neurosciences today, which I have characterized as functional decomposition and neural localization, recognizes the incompleteness of each of these views. The state of the question has moved forward, and both researchers and philosophers today generally agree on something like the method I’ve outlined.) The working assumption in the reductionistic science of intelligent behavior is that all this can be understood in strictly physical terms. This may seem obvious -- after all, we can make machines that do what the trained mouse can do. The reason this is possible is that information is a physical reality. It’s relational, but physical. So is information processing. Let’s look at a simple example. [slide 11. Thermostat] The thermostat guides the behavior of the furnace in response to changes in temperature in the surrounding room. This feedback mechanism is a simple example of informationprocessing. If you think of hundreds of mechanisms, like the thermostat-furnace mechanism but much more sophisticated, packed together in a single place, serving the needs of a living organism, you get a pretty good picture of what goes on in a brain. The fact that the thermostat keeps track of the room temperature and turns the furnace on as soon as the room temperature drops to 68 degrees is a pretty fancy kind of fact. But we can see that the whole process is one of ordinary physical causality. For example, the 6 reason the bimetallic strip performs its function is because the 2 metals it’s made out of have 2 different coefficients of expansion. The furnace’s seemingly responsive behavior can be given a reductionistic explanation in terms of the behavior of the fundamental particles that the thermostat and furnace are made out of, when these fundamental particles are brought together in a certain arrangement. [A reductionism methodology does not imply a purely bottom-up approach. No one in neuroscience thinks that the way to understand consciousness and other psychological capacities is to first understand everything about the fundamental particles and the molecules, then everything about every neuron and synapse, and thus to continue laboriously up the various levels of organization until one finally reaches the psychological level. The research strategy in what are commonly interpreted as prime cases of reductionist success in science, like the explanation of thermodynamics in terms of statistical mechanics, of optics in terms of electromagnetic radiation, or of hereditary transmission in terms of DNA, was not bottom-up in this way. Instead, research is conducted on many levels simultaneously. Scientific hypotheses at the various levels coevolve in a helpful way, as they both stimulate and constrain one another. ] Anytime we do get a reasonably complete reductionistic explanation of a phenomenon, it’s a tremendously satisfying achievement! It’s important medically, because it enables us to devise therapeutic interventions at multiple levels. And our intellectual curiosity is satisfied, because we understand the underlying mechanisms through which each higher level capacity emerges. We understand the How and Why of it. (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 in such a way that a certain new kind of thing happens.) So now, returning to our topic of consciousness, we have some idea of what a reductionistic, physicalist explanation of consciousness would look like. We would expect to find that consciousness has some biological function(s). That is, we would expect to find that a mental state’s being conscious gives it causal properties that it would not otherwise have. We would also expect to be able eventually to explain the fact that a certain mental state is conscious in terms of certain structural and functional relations among the lower-level entities that it’s made out of. In other words, we would expect to find the mechanisms that produce or constitute conscious experience. 3. The Functionalist Approach to Consciousness . [slide 12. Definition of Functionalism Functionalism or computationalism about the mind is the view that what makes a mental state to be a mental state is its causal relations with inputs, outputs, and other mental states. Compare: What is it that makes a bimetallic strip a thermostat?] 7 Almost everything we’ve learned about the mind, about human cognition, over the last 50 years has been learned using a functionalist or computationalist paradigm. The brain is a computer, and the mind is the software running on the computer. So it’s only natural that the first attempts to develop a science of consciousness would approach the topic from a computational point of view. Scientists often use a functionalist definition of “consciousness” as access to information. The thinking goes something like this: We know that brainstates can carry information about the environment, just as the bimetallic strip does in the thermostat. Most of these informational patterns in the brain never become conscious. But for those that do become conscious, we can think and talk about the information our nervous system has picked up. So we can define “consciousness” functionally in the following way. We say that the informational contents of brainstates are “conscious” if they’re available for verbal report and voluntary behavior. [slide 13. The Function of Consciousness] “The biological usefulness of visual consciousness in humans is to produce the best current interpretation of the visual scene in the light of past experience, either of ourselves or of our ancestors (embodied in our genes), and to make this interpretation directly available, for a sufficient time, to the parts of the brain that plan voluntary motor output, including speech.” Crick & Koch (2003:36)] [slide 14. The Brain Produces Behavior] We can tell someone about what we’re attending to visually, for example, but not about the way our brains are controlling our posture, or the level of estrogen in our blood. The information-processing in the brain falls into two pools. One pool, which includes some of the products of high-level sensory processing and the contents of short-term memory, can be accessed by the neural systems that produce our speech, rational thought and deliberate decision-making. The other, much larger pool, which includes autonomic responses, the computations controlling movement, repressed desires or memories if there are such things, and in general the great bulk of information-processing activities of the brain, cannot be accessed by our systems for thought, speech and voluntary action. The rules for computing trajectories that our brain uses when we catch a fly ball, for example, are not available for use on a physics exam. Some part of our brain has the information and the rules it needs to compute the ball’s motion and the correct arm movement we need to catch it, but “we” as conscious persons do not have access to the rules our own brain uses. “Consciousness” is sometimes used in this informationprocessing sense, to refer to the availability of the outputs of some subsystems of the brain (but not most) to central processing. When we’re consciously aware of a piece of information, we can talk about it and use it to produce voluntary actions. Many different parts of the brain can use the information. Otherwise, not. [slide 15. Specialized Input Analyzers] The hundreds of information processors in our brain are specialized mechanisms that evolved for carrying out particular tasks. The surfaces of our body present the nervous 8 system with information in various analog formats, like spatial arrays of light intensity from retinal receptors and temporal patterns of sound frequencies from the cochlea of the ear. Our specialized information processors are input analyzers. They analyze the input from our sensory surfaces. Each is a set of specialized computational routines for converting one specialized type of analog input into the unifying, modality-independent propositional format of the brain’s central processes of belief fixation, long-term memory, planning and problem solving. Input analyzers are the intermediate-level computations that accomplish this. Each module is fine-tuned to capture specific features of environmental input, in forms that facilitate its special computations. (Each has its own special vocabulary or format, so they don’t talk to each other.) Features of the human auditory signal, for example, feed at least 3 specialized processors within the language system of the brain: one to compute the phoneme structure of speech, one to facilitate voice recognition, and one to compute emotional tone (Etcoff 1989). The visual system of humans is thought to have 20-30 different computations running off of the same stream of retinal input, each controlling a separate function like eye movement, grasping, ocular pursuit, object recognition, or balance. We have no direct conscious access to the information processing computations in our brain. But we know they’re there. If we stand with our eyes open and think about what the visual system is doing, it seems that it’s giving us the experience of a visual scene. It’s somehow producing our conscious visual experience. But in fact the spatial arrays of light intensity provided by retinal receptors are feeding separate computations of many different behaviorally-relevant features that guide several sensori-motor routines, (and they’re not, in fact, creating a complete visual scene). If you try to balance on one foot with your eyes open, it’s easy. But if you try it again with your eyes closed, it’s much more difficult! That’s because all the time our eyes are open, an certain module, unbeknownst to us, is using visual information to compute the motor instructions for keeping our balance (Bruce, Green, Georgeson 1996). Keeping our balance might seem like a very simple thing, because we don’t have to pay any attention to it to do it. But if you think of yourself trying to make a machine shaped like us that would move around and keep itself in balance as it moved, never tipping over, using movable parts of itself to make changes in its environment, etc., all without strings like a puppet, or any kind of remote control, you get some idea of the complex computations that would have to be built into the machine somewhere. In a real sense our brains are smarter than we are: the constant unconscious computations going on in our brain far outstrip any calculations we can make consciously. It’s important to keep this in mind, as we try to understand consciousness. Our fleeting conscious experience rides on top of a large and very complex set of information-processing modules in the brain, each of which evolved to meet the biological needs of moving organisms. We share many of these input-analyzing neural circuits with other mammal species. [slide 16: Probability Matching in Ducks] 9 [slide 17: Probability Matching in Humans] [slide 18. Target Tracking in Humans] The topic of consciousness in this information-processing sense lends itself to scientific study. We have learned many things about consciousness in this sense. Information from sensation and memory guides behavior only in an awake animal, so part of the neural basis of access-consciousness can be found in subcortical brain structures that control wakefulness. We know that information about an object being perceived is scattered across many parts of the cortex. So consciousness as information access requires a mechanism that binds together data in different neural networks. Crick and Koch (1995) proposed that synchronization of neural firing might be one such mechanism, perhaps entrained by loops from the cortex to the thalamus, the cerebrum’s central relay station. We know that voluntary, planned behavior requires activity in the frontal lobes. So access-consciousness may be determined in part by the anatomy of the connections from various parts of the brain to the frontal lobes. Jackendoff (1987,1994) observed that access consciousness taps the intermediate levels of information processing. Visual processing, for example, runs from the rods and cones in the retina, through levels representing edges, depths, and surfaces, etc. to a recognition of the objects and events in front of us, to longterm memories of the gist of what happens. We aren’t aware of the lowest levels of processing, which go on in parallel. Our immediate awareness doesn’t exclusively tap the highest level of representation, either. The highest levels—the gist of a scene or a message—tend to stick in long-term memory days and years after an experience, but as the experience is unfolding, we are aware of particular sights and sounds, not just the gist, of what’s happening. The intermediate level of processing produces constancies of shape, color, object identity, etc., across changes in viewing conditions, and these tend to track the environmental object’s inherent properties. So we can understand the engineering reasons for making the products of intermediate-level processing in subsystems of the brain (rather than all processing, or only the highest level of processing) available to consciousness and thus to the system as a whole. If only the highest level of processing were available, we’d be conscious of everything as instances of already familiar categories, and learning of new categories from experience would be impossible. According to one functionalist theory of consciousness (i.e. a theory that defines consciousness in terms of the function it serves), once you have a living organism like a human being that is behaving adaptively in its environment due to the operation of the information processors in its brain, there is really no difference between activity that is conscious but immediately forgotten, and activity that is unconscious. This may seem like a crazy idea at first, but think about the familiar example of driving. [slide 19. Driving on a Familiar Route] You drive on a well-known route, say to work or to a friend’s house. On one occasion you are acutely aware of all the passing trees, people, shops and traffic signals. On another day, you’re so engrossed in thinking about something that you’re completely 10 unaware of the scenery and of your own actions. You get all the way there and then realize that you drove there, unaware of what you were doing. You have no recollection at all of having passed through all those places and having made all those decisions. Yet you must have noticed the traffic signals and other things, because you didn’t drive through a red light, or hit a pedestrian, or stray into a wrong lane of the road. You found your usual route to the place you were going, and you weren’t asleep. Is it that you were conscious of traffic conditions at each moment, but immediately forgot them? Or were you not conscious of traffic conditions and just driving unconsciously? [slide 20. Post-Surgery Example. ] An outside observer would make the judgment that I was conscious, on the basis of my behavior. I felt sure that I wasn’t conscious during that period, because of my lack of any memory of it. It’s not obvious how to tell who is right. Could my behavior during that period have been produced unconsciously? Should the patient herself be the judge of whether she is conscious?] The driving example and the surgery example highlight the important connection between consciousness and the type of memory called autobiographical memory. Autobiographical memory requires a degree of self-consciousness. Consciousness without self-consciousness may be so transient and fleeting that it’s virtually indistinguishable from no consciousness at all.. I will say something more about this in section 7. In both the driving example and the post-surgery example we want to know whether the person was behaving intelligently but unconsciously, or was conscious but not remembering her experience from one moment to the next. Is there any way of being sure about the answer to this question? Is the question a legitimate question? The question makes perfectly good sense if being conscious means that our brain produces a movie of our external environment and then we’re in the brain somewhere looking at the movie. (Philosophers call this mistaken idea “the Cartesian theater”.) [slide 21. The Little Man in the Brain] But we know that isn’t the case. There’s no little audience-person in the brain. The brain produces an interpretation of environmental input that guides its production of motor activity. Whether this process is conscious or not is not always easy to tell. So in the driving situation, you might ask yourself: “Was the red light I stopped at in consciousness, but then forgotten? Or was it never in consciousness?” One philosopher, Dan Dennett, rejects this as a mistaken question. On his multiple drafts theory of consciousness, all kinds of cognitive activities, including perceptions, emotions and thoughts, are accomplished in the brain by parallel, multi-track processes of interpretation and elaboration of sensory inputs, and they are all under continuous revision. Like the many drafts of a book or article, perceptions and thoughts are constantly revised and altered, and at any point in time there are multiple drafts of narrative fragments at various 11 stages of editing, in various places in the brain. There is no little man in the brain who is watching all the information-processing. There are only multiple, parallel, unconscious, ongoing interpretations of input. The sense that there is a single narrative stream or “flow” of consciousness comes about when one of the streams of unconscious information-processing is probed in some way—for example, by someone asking a question or requiring some kind of response. Some of the multiple interpretations of input the brain produces at each instant are used in controlling actions, or producing speech, and some are laid down in memory, but most just fade away. Take the example of a bird flying past your window. Your conclusion, or judgment, that you saw the bird is a result of probing the unconscious steam of multiple drafts at one of many possible points. There is a judgment in response to the probe, and the event may be laid down in memory. But there is not in addition to that, some bare conscious experience of seeing the bird fly past. According to Dennett, mental contents arise, get revised, affect behavior and leave traces in memory, which then get overlaid by other traces, and so on. But there is no actual fact about what I was actually experiencing the moment the bird flew past. There is no right answer, because there is no little man watching a movie in the brain. There are no fixed facts about the stream of consciousness independent of particular probes, so it all depends on the way the parallel stream gets probed. If something leads you to say something about the bird, or to do some action in response to the bird, you will have a conscious experience of the bird. [slide 22. Unconscious or Forgotten?] [slide 23. Amnesia] This type of theory of consciousness has implications for our understanding of the self, or the soul. On this view, the self as a single unified subject of consciousness is an illusion. What happens is that as the contents of unconscious information-processing in the brain are fixed by probing the streams of processing at various points-- as we make judgments and as we speak about what we’re doing or what we’ve experienced -- so the illusion is created of there being a unified self. What we call the “self”, in Dennett’s view, is a sort of “center of gravity” of the accumulating narratives that spin themselves together in the brain. Just as the “center of gravity” of a physical object is not a real physical entity but only an abstraction, so the idea of a “unified self” is not a real physical entity but only an abstraction—a “narrative center of gravity”, the center of gravity of the loose bundle of narratives and memories put together by our brain. We can ask ourselves a related question: Do we have a single unified stream of consciousness? We’re not conscious in dreamless sleep. But even when we’re awake, are we conscious all the time? William James asked the question, way back in 1890, whether consciousness is continuous, or “only seems continuous to itself, by an illusion?” (1890:200) Whenever we ask ourselves, “Am I conscious now?” the answer always seems to be “Yes”. We cannot catch ourselves not being conscious, and when we do find ourselves being conscious, there seems to be one me, and one unified experience. But 12 what is it like the rest of the time? When I’m not asking myself whether I’m conscious at the moment, am I conscious? One possibility is that there is nothing it is like most of the time. Rather, there are just multiple parallel streams of unconscious processing going on. Then, every so often, we ask, “Am I conscious now?” or “What am I conscious of?” or in some other way we reflect introspectively about what is going on. Then, and only then, is a temporary stream of consciousness concocted, making it seem as though we have been conscious all along. At these times, recent events from memory are brought together by paying attention to them, and the appearance of a unified self having unified experiences is created. As soon as attention lapses, the unity falls apart and unconscious processes carry on as normal. Just as the refrigerator light is usually off, and the door is usually closed, so we are usually in an unconscious state of parallel multiple drafts. Only when we briefly open the door is the illusion created that the light is always on. (Every time I look in the refrigerator, the light is on. So someone who doesn’t know better might conclude that it’s always on, continuously. In the same way, every time I reflect on whether I’m conscious or not, I am conscious. So I conclude that (except when I’m sleeping) I have a continuous stream of consciousness.) For a functionalist like Dennett, consciousness arises in any animal species that reaches a certain level of self-monitoring, due to complex social relations and language. It is not something extra, over and above the information-processing that goes on in the brain. 4. Consciousness as Subjective Experience Functionalism has been the dominant paradigm in cognitive science for 50 years. A tremendous amount has been learned about human cognition, or the human mind, in that time. But recently there’s been growing dissatisfaction with functionalism, largely because it doesn’t reflect certain important aspects of how brains actually work. An additional dissatisfaction with functionalism stems from the fact that it seems ill-suited to explain certain aspects of consciousness. There’s a lot of talk these days in the philosophy of mind about “qualia” ( a philosophers’ term for the qualitative properties of conscious experience). Think of the exact smell of popcorn burning in the microwave, for example. We know the smell is caused by certain molecules entering your nose and reacting with receptors there, but the experience of the smell doesn’t seem to be a matter of physical molecules. In your experience, the smell is a vivid, subjective, private quality that’s unique, and can’t be described in words. This particular smell is an example of what philosophers call “qualia”. Sometimes questions about consciousness are phrased as questions about qualia. How are subjective qualia related to the objective physical world? How can a physical thing like the brain produce a subjective, private experience? Dennett simply denies that such things as qualia exist. He denies, in other words, that there is such a thing as conscious experience with particular private ineffable qualitative properties, separable from our judgments about experience (or other dispositions to 13 speech or behavior that our sensory discriminations put into effect). When you smell popcorn burning in the microwave, there is no such thing, he thinks, as a single, unified “moment of conscious experience” separate from the multiple parallel processes by which the brain interprets the sensory input and produces a bodily response. Dennett gives examples to illustrate his point. [slide 24: The Beer Drinker] The experienced beer drinker says that beer is an “acquired taste”. When he first tried beer, he hated it. But now, he loves it. What, exactly, has changed? Is it that beer tastes different to him now than it did then? Or is taste the same, but now he likes what he formerly hated? If there are two separate things, the quale, or how precisely it tastes to him, and his opinion about the taste, then a person should be able to decide which has changed. But can you decide? Assume the beer is chemically exactly the same as before. And his behavior then and now is different: he used to drink little, now he drinks lots, he used to say Yuck, now he says Yum, etc. Somewhere between the identical molecular inputs then & now, and the different outputs then & now, something has changed. But is there an isolatable quality of experience, a bare quale, that hasn’t changed? Dennett thinks not. If Dennett is right, there is no distinction between a stimulus s seeming to be F to a person, and the person’s judging that s is in fact F. There is no bare conscious experience prior to a probe of some type. Other philosophers and neuroscientists disagree with Dennett’s position, and insist that the qualitative aspects of experience have a reality all their own. There is something that it’s like to feel a sharp pain in your gut, or sand between your toes. Each of these conscious states has what philosophers call phenomenal properties, or qualia. When we look into a green light-flash, and then close our eyes, we experience a red after-image. Redness is one of the phenomenal or qualitative properties of the afterimage experience. But if redness is a real property, and we make the assumption that everything real is ultimately physical, what exactly is the redness a property of? The science of human color vision gives us a physical account of how an afterimage is produced. But none of the entities that play a role in the scientific explanation are red. There needn’t be anything red in the person’s external environment, and nothing in his eye or brain is red either. Yet the redness is a real feature of the experience. What physical entity is this experienced redness a property of? It’s not a property of anything in the brain. Nothing, it seems, in the neurobiological account of afterimages explains why the experience of the afterimage feels or seems just exactly the way it does. This is what philosophers have come to call “the hard problem” of consciousness. We want to focus today on this aspect of consciousness as subjective experience. We say that a being is conscious if there is something it’ s like to be that being, to use a phrase made famous by the philosopher Thomas Nagel. A bat, for example, might be conscious. If so, Nagel argued, then there is something it’s like to be the bat. 14 [slide 25. What Is It Like to Be a Bat?] Each conscious mental state, each moment of conscious experience, has a qualitative feel to it. (There’s something it feels like to be conscious.) Consciousness in this sense appears to be a private, inside dimension to life processes. We might even wonder, does everything physical have this inside dimension to it? Some philosophers [Whitehead 1929; Russell 1929; Hegel; Bradley; Royce; Maxwell 1979; Lockwood 1989; Chalmers 1996; Rosenberg 1997; Griffin 1998; Strawson 2000; Stoljar 2001] have concluded that it does. Consciousness in the sense of subjective experience is deeply puzzling. For one thing, it seems that all the information-processing the brain does could just as well be done without subjective experience. For example: any effect of actually feeling how hot the soup is, like waiting awhile before you eat it, could be accomplished by pure informationprocessing triggered by a mechanical sensor for temperature. There wouldn’t need to be conscious experience at all. The same thing—detection of temperature and avoidance of further contact—could be programmed into a machine. This makes it appear, at least, that consciousness is a totally mysterious and causally irrelevant side effect of the information processing that goes on in the brain, a kind of dangling afterthought, an epiphenomenon. [slide 26. Can a Robot be Conscious?] And how does the brain produce subjective experience? There is at the present time no widely accepted scientific answer to this question, though we’ll look at some attempted answers in a moment. The state of our ignorance is such that we can imagine all kinds of strange possibilities for arrangements between the physical and the mental. We have no scientific theory of consciousness to tell us which of these imagined arrangements are possible and which are not. Here is a list of some of the imaginative questions we can still legitimately ask about consciousness, (due to the state of our ignorance), compiled from the work of other people by Steven Pinker at MIT (1997:145-6). [ slide 27: Pinker’s List] 1. If we could ever duplicate the information processing in the human mind as an enormous computer program, would a computer running the program be conscious? 2. What if we took that program and trained a large number of people, say, the population of China, to hold in mind the data and act out the steps? Would there be one gigantic consciousness hovering over China, separate from the consciousnesses of the billion individuals? If they were implementing the brain state for agonizing pain, would there be some entity that really was in pain, even if every citizen was cheerful and light- hearted? 15 3. Suppose the visual receiving area at the back of your brain were surgically severed from the rest and remained alive in your skull, receiving input from the eyes. By every behavioral measure you are blind. Is there a mute but fully aware visual consciousness sealed off in the back of your head? What if it were removed and kept alive in a dish? 4. Might your experience of red be the same as my experience of green? Sure, you might label grass as “green” and tomatoes as “red”, just as I do, but perhaps you actually see the grass as having the color that I would describe, if I were in your shoes, as red. 5. Could there be zombies? That is, could there be an android rigged up to act as intelligently and as emotionally as you and me, but in which there is “no one home” who is actually feeling or seeing anything? How does each of us know that the others in the room are not zombies? 6. If someone could download the state of my brain and duplicate it in another collection of molecules, would it have my consciousness? If someone destroyed the original, but the duplicate continued to live my life and think my thoughts and feel my feelings, would I have been murdered? 7. What does it feel like to be a bat? Do beetles enjoy sex? Does a worm feel pain when a fisherman puts it on a hook? 8. Imagine that surgeons replace one of your neurons with a microchip that duplicates its input-output functions. You feel and behave exactly as before. Then they replace a second one, and a third one, and so on, until more and more of your brain becomes silicon. Since each microchip does exactly what the neuron did, your behavior and memory never change. Do you even notice the difference? Does it feel like dying? Or, is some other conscious entity moving in with you? These questions all stem from the fact that we have two basic scientific approaches to consciousness: the neurobiological and the computational or functional -- and both of these seem to leave the fact that a mental state is conscious an extra, dangling, unexplained, causally-irrelevant mystery. It seems we could have a completely specified brainstate, and not be sure whether or not the person whose brainstate it is is conscious. (This is the point in questions 6 and 8.) Similarly, it seems we could have a completely specified computational or information-processing state (something like a software program), and wonder whether, if it were run on different media, any of them would be conscious. (This is the point in questions 1, 2, 4, 5 & 7.) Consciousness is still not successfully integrated into either one of our two major scientific approaches to the mind. 5. The Neurobiological Approach to Consciousness: Is there a neural correlate of consciousness? Neuroscientists often assume that a mental state (like having a thought) simply is a brainstate. The idea is this: if the neural systems of your brain are in a particular state of activation, then, by definition, you are having a particular conscious experience. To 16 some people this just seems obvious. Injuries to the brain interfere with mental functioning. Different parts of the brain are active during different kinds of mental activity. There’s nothing magical besides the brain inside the head, so it must be that the mind is simply the brain. Medically speaking, learning what happens where in the brain is immensely important. It wasn’t very long ago that doctors trying to relieve seriously disabling and otherwise intractable epilepsy removed the medial temporal lobe on both sides of a man’s brain, not realizing that by doing this they destroyed his ability to lay down new personal memories for the entire remainder of his life! This patient can learn new skills, but he cannot remember his experiences from one moment to the next. He lives now almost completely confined to the present moment. His life as a meaningful, ongoing narrative came to an end on the fateful day of his surgery back in 1953 [H.M. 1953. Bilateral temporal lobectomy. Hippocampus. Anterograde amnesia. Brenda Milner. Procedural vs. declarative memory. Cf. Clive Wearing.] So neuroscientific knowledge is important. Each hard-won increment in our understanding of the neuroscience of the brain enables therapeutic interventions that save lives and raise the quality of life for many people. There could hardly be more important work. It used to be that localizing the brain lesion that led to a patient’s cognitive and behavioral deficits had to await an autopsy after death. But now neuro-imaging techniques permit the neurologist to analyze lesions in a 3-D reconstruction of the living patient’s brain, displayed on a computer screen, sometimes while the patient performs certain cognitive tasks. [slide 28. PET Scan] [slide 29. MRI Scan] While the therapeutic value of these neuroimaging techniques is uncontroversial, the implications of the knowledge they generate for our understanding of consciousness is a matter of debate. One philosopher put it this way, in an article complaining that too many tax dollars were going into neuroimaging studies at the time. [slide 30: Fodor’s Question] I t i sn ’ t , af t er al l , ser i ou sl y i n dou bt t h at t al k i n g ( or r i di n g a bi cycl e) depen ds on t h i n gs thatgooninthebrainsomewhereorother. Ifthemindhappensinspaceatall,it happenssomewherenorthoftheneck. Butwhatexactlyturnsonknowinghowfar north? Itbelongstounderstandinghowtheengineinyourcarworksthatthe functioning of its carburetor is to aerate the gas; that’ s part of the story about how the engine’spartscontributetoitsrunningright. Butwhy(unlessyou’rethinkingof 17 having it taken out) does it matter where in the engine the carburetor is? What part of how your engine works have you failed to understand if you don’ t know that? (“Let yourbrainalone.” JerryFodor,LondonReviewofBooks,9/30/99) Where in the brain something happens, Fodor suggests, doesn’ t tell us a lot about conscious mentality. [On a personal note: I remember that years ago when I first heard about the new neuroimaging techniques that were being brought into use at the time, I thought that using PET scans and MRIs to learn how we acquire language or do calculus problems was about as likely to succeed as trying to understand American military policy bystudyingwhereinthePentagonthelightswereon.] Weneedtounderstandwhat functions the brain performs and how, Fodor argues, but where these things happen doesn’t tell usmuch. Fodor’ s analogy is off the mark, in important ways. Neuroimaging techniques can help ustestfunctionalhypotheses. (Theytellus,notjustwherethelightsareonandwhen,so to speak, but whose phone calls are going to whom, and under what circumstances.) Still, it’ s not clear that even if we had complete knowledge of which neurons were firing, at what rate, at each location in the brain at every moment of our conscious experience, that that would amount to anything more than a bare correlation. It wouldn’ t explain how orwhyneuralactivityinthebrainleadstoconsciousexperience. Whyshouldbunchesof neurons exchanging chemicals back and forth have such a strange effect? Itwouldn’teventellusthatneuralactivityleadstoconsciousexperience. Conscious experience and neural activity can be correlated, but maybe it’ s just the reverse: maybe consciousexperienceleadstoneuralactivity. Ormaybebothareeffectsofacommon cause. In a series of influential articles in the 1990’s [1990, 1995, 1998] Francis Crick and Christof Koch argued that neuroscience is the right way to develop a scientific understanding of the mind, and that it was time for neuroscientists to stop avoiding the difficult topic of consciousness. They suggested a research program to address the question of consciousness directly, by looking for differences between neuronal processes in the brain that are accompanied by conscious experience and those that are not. [slide 31: Necker Cube] When we look at a Necker cube, we see it first one way, then another. This is called “rivalry”. We don’t see both appearances at once, or combine them both into one. We experience an alternation back and forth between the two appearances. This means that our conscious experience changes while the visual input stays exactly the same. What explains this difference? Call one way it looks to us A and the other way it looks to us B. Can we look for a difference in the brain between the situation in which the physical stimulus is there and the interpretation of the cube as looking A is consciously experienced, and the situation in which the same physical stimulus is there and the interpretation of the cube as looking A is not consciously experienced? And if we do 18 locate such a difference, what’s the relationship between the objective physical facts in the brain and the subjective facts about how it appears to us? The first research using this approach was done with macaque monkeys (who have visual systems anatomically similar to ours), by Logothetis and his colleagues. The monkeys were trained to report which of two pictures they were seeing by pressing a lever. Trained monkeys were then put in an experimental set-up where different displays were shown to each eye. They reported binocular rivalry, just as humans do in these situations. [slide 32. Binocular Rivalry ] [slide 33. Bistable Perception] Next the researchers made recordings from various areas of the monkeys’ brains, using electrodes in single cells. They were looking for the cells whose firing rate correlated with, not the unchanging visual input, but the changing conscious perception, as reported by the monkey’s behavior. [slide 34. Single-Cell Recording] They found that neural activity in the early stages of the visual pathway—primary visual cortex (V1), and V2—was better correlated, on the whole, with the unchanging input. The activity level in these neurons didn’t change when the monkey’s perception changed. Further along the visual pathway, some of the cells responded to what the monkey reported seeing. Finally in the inferior temporal cortex (IT) almost all the cells changed their response according to what the monkey reported seeing. So if the monkey pressed the lever to indicate a flip, most of the cells that were firing stopped firing and a different set of cells started. It looked as though activity in this area corresponded to what the monkey was consciously seeing, not to the physical stimulus. (Logothetis & Schall 1989; Leopold & Logothetis 1996, 1999) [slide 35. Macaque Results] Does this mean that the NCC for the monkeys’ conscious vision lies in IT? More recently, other researchers have done similar experiments with humans. Using fMRI, EEG and single unit recording, they’ve identified changes in cortical activity that are precisely correlated with the changes in conscious perception (Alais & Blake 2004; Lumer et al. 1998; Brown & Norcia 1997; Kreiman, Fried & Koch 2002). Let’s look more closely at the idea of a neural correlate of consciousness. Most neuroscientists assume that for every conscious state we experience, there is some minimal neural substrate that is necessarily sufficient for its occurrence. The relation between the minimal neural substrate and conscious experience is assumed to be a lawlike relation of either causation or identity. Most vision scientists, for example, assume 19 that there exists somewhere in the stream of visual processing a set of neurons whose activities form the immediate substrate of conscious visual perception. That would mean that the occurrence of a particular pattern of activity in this set of neurons is sufficient for the occurrence of a particular conscious perceptual state. If this neural activity happens, the conscious experience will occur, no matter what’s going on elsewhere in the brain. So, for example, if you could set up conditions to properly stimulate the NCC in the absence of both retinas, or in the absence of any visual input whatsoever, the correlated conscious vision would still occur. A growing number of investigators believe that the first step toward a science of consciousness is to discover such a neural correlate of consciousness. [slide 36. NCC] We can state the NCC assumption this way: For every conscious experience E, there is a neural correlate of consciousness (NCC) such that (i) the NCC is the minimal neural substrate whose activation is sufficient for the occurrence of E, and (ii) there is a match (isomorphism) between features of the NCC and features of E. We need to understand 4 concepts here: a) the minimal neural substrate b) sufficiency c) necessity, and d) isomorphism. a) the minimal neural substrate A neural system N is the minimal neural substrate for a conscious experience E if the states of N suffice for the corresponding states of consciousness, and no proper part of N is such that its states suffice for the corresponding states of consciousness. So the NCC for a particular conscious experience is the smallest set of neural structures whose activity suffices for that conscious experience. b) sufficiency. NCC--> E Not every neural state that is correlated with conscious experience is a sufficient condition for conscious experience. We can see this by looking at the example of blindsight. People with damage to primary visual cortex sometimes exhibit a phenomenon called “blindsight”. The patient D.B., for example, had a tumor removed from area V1 on one side of his brain, leaving him blind on the opposite side of his visual field. If he looks straight ahead and an object is placed on his blind side, he cannot see it. (“hemianopia”) [slide 37. Hemianopia Caused By Damage to V1] 20 In one experiment he was presented with a circle filled with stripes in his normal field and asked whether the stripes were more horizontal or vertical. He had no trouble answering correctly. Then he was shown the same thing in his blind field. He said “I can’t see anything at all.” But when he was asked to guess on repeated trials which way the stripes were oriented, he was correct 90- 95% of the time. (Weiskrantz 1986, 1997) [slide 38: Blindsight] [slide 39: Testing for Blindsight] So there’s a correlation between activity in V1 and conscious seeing: when you’re missing some of your neural activity in V1 you’re missing some of your conscious vision. And it’s a robust correlation -- it shows up reliably across patients with this kind of brain damage. But we can’t conclude from this that neurons in V1 are the NCC that Crick & Koch and other neuroscientists are looking for. Neural activity in V1 may be a necessary condition for conscious vision, but it’s not a sufficient condition. Here’s an analogy. Activity in the power cord of an LCD projector is positively correlated with there being a ppt presentation up on the screen. If the power cord’s defective, you don’t get a presentation. But the power cord activity is not a sufficient condition for the ppt presentation. Lots of other stuff has to be working properly, too. So the NCC is the smallest set of neural structures whose activity is sufficient all by itself for a particular conscious experience. (Language about sufficient conditions expresses a logical relation. Neuroscientists differ as to whether the underlying ontological relation is one of causation or identity.) So in answer to our question about the monkey experiments: No, activity in inferior temporal cortex is not, all by itself, the neural correlate for the monkeys’ visual experiences. c) necessity. The third element is this: if discovering the NCC is to be the first step in developing a scientific explanation of consciousness, then the relationship between the NCC and consciousness has to hold in a lawlike way (as a natural law), and not merely by accident. d) isomorphism And finally, it is not enough that neural activity in the structures that form the NCC be sufficient, as a matter of natural necessity, for conscious experience. It is also widely assumed by neuroscientists that there must be a mapping (under some description) from features of the conscious experience to features of the minimal neural substrate. For example, if a certain pattern of activity in the NCC is sufficient for the occurrence of E, and E is a visual experience of 2 surfaces with a brightness difference between them, then the NCC must exhibit patterns of activity corresponding to the 2 surfaces and a pattern of activity corresponding to the perceived difference in brightness. 21 A map is a good example of an isomorphism. The San Bernardino National Forest and a map of the San Bernardino National Forest differ in almost every respect (size, weight, color, history, location in space, temperature, molecular structure, etc.). But there is an isomorphism between them, meaning that certain relations between elements of the map reflect in a systematic way certain relations between elements of the San Bernardino National Forest. We saw an example of isomorphism in the spatial maps formed in the hippocampus of the mouse after training on the water maze. Notice that isomorphism isn’t enough to make a scientific explanation all by itself. Isomorphism works together with sufficiency and nomic necessity. You can have an isomorphism that’s totally accidental (like the resemblance between a pattern that forms in the clouds and President Bush’s face) and it won’t be explanatory. But for a reductionistic explanation, the link between the NCC and conscious experience needs to involve more than logical sufficiency and lawlikeness. If the NCC is going to be the key to a reductionistic scientific explanation of conscious experience, there has to be sufficiency, lawlikeness, and some explanatory mechanism or isomorphism between them. When we say “Water is H20” it’s not just an arbitrary correlation (even one that holds universally and by necessity) between being water and being H2O. A physical chemist’s understanding of the atomic structure of hydrogen & oxygen, and of the structure of the water molecule explains why water has the properties it does. Its atomic structure gives us the mechanisms at the atomic level that lead to the emergence of water’s molecular properties. So the isomorphism requirement is met, generally speaking, when we discover a mechanism at one level of the functional hierarchy that leads to the emergence of a novel property at the next higher level. In the case of consciousness, if we restrict ourselves to the level of neurobiology, no generally accepted explanatory isomorphisms have yet been found. [The most convincing example of a possible explanatory isomorphism I’ve seen is the hypothesis of an isomorphism between our phenomenal color space (how our experiences of color are related to each other) and its neuronal basis, modelled in terms of vector coding. [slide 40: Pat on vector space of opponent cell coding] This is an interesting example of an attempted reductionistic explanation, and if we had time, it would be worth investigating further. We might consider it as a possible topic for some future conference.] Instead, what we find about the relation between the properties of our conscious experiences and the properties of the patterns of neural activity in the brain that are correlated with them is a distinct absence of any explanatory isomorphism. A neural representation of a large visual object won’t be larger than a neural representation of a smaller one, and a neural representation of red won’t be red. The pattern of neuronal activity in the brain that underlies our experience of a continuously filled-in page of text 22 won’t be a continuously filled-in neural pattern. In general, what is represented offers little or no information about the way it is represented in the brain. Let’s look at an example. [slide 41. Change Blindness. Ronald Rensink (2000)] When our eyes are open we seem to experience a uniformly detailed and complete scene in front of us. Since our experience is as of a complete scene, the assumption has been that the brain must somehow integrate its successive inputs from the retinas into one big uniformly detailed and complete representation of the world in front of us, that stays stable and complete across all our body movements, head movements, eye movements and blinks. For example, we know that when we blink, there’s no input to the eyes for a short period of time. But we don’t experience little black-out periods, even though we’re constantly blinking. The assumption has been that the brain “fills in” where there are gaps in visual input (like blinks, saccades, and the blind spot), constructing a complete picture of the scene in the brain. But several recent experiments suggest that this is not the case. Our brain does not create our feeling of seeing a complete scene by actually creating a completely detailed representation of the scene to be viewed by a little man in our brain. Our brain creates the conscious experience of a complete and continuous scene by representing that the scene is complete and continuous, not by forming a complete and continuous inner representation of it. Here’s how we know this. Beginning in the 1980’s, eye trackers were used to detect a person’s eye movements (called saccades) as they looked at a visual display. Changes were then made to the display during their saccades. The changes were large and obvious ones, that you couldn’t miss under normal circumstances. Still, when made during the eye movements, they went unnoticed. If the assumption were true that our brain makes a rich and detailed inner representation of the scene that can be used to compare details from one moment to the next, it’s hard to see how such big changes could go unnoticed. You can get the same effects without eye trackers using what’s called the flicker method. Rensink, O’Regan and Clark (1997) showed an original image alternating with a modified image (each shown for 240 msec), with blank gray screens (shown for 80 msec) in between. This creates gaps in input similar to blinks or saccades. Then they measured the number of presentations until the subject noticed the change. Typically subjects take many alternations before they detect the changes, even though the changes are large ones that would be noticed directly if presented without the blanks in between. [try some of the demos] Here’s an explanation of what happens. When we’re not blinking or moving our eyes (we’re fixated on the scene), motion detectors in the visual system pick up changes in visual input and direct our attention to that location in the visual field. But when we move our eyes this causes a massive blur of activity that swamps the change detection 23 mechanisms, leaving only memory to detect changes. And contrary to the idea of a complete inner representation of the scene in the brain, trans-saccadic memory is extremely poor. With every saccade, most of what we see is thrown away. There are different theories about just what, and how much, is retained across blinks and eye movements. Simons & Levin (1997) suggest that during each visual fixation we extract the meaning or gist of the visual scene. Then, when we move our eyes, we get new visual input, but if the gist remains the same, our perceptual system assumes that the details are the same. So we don’t notice changes. We get a feeling of continuity and completeness because the brain retains only the gist of the scene and uses a default assumption that details remain the same. O’Regan and Noe (2001) have a slightly different view. They suggest that what remains between saccades is not a picture of the world, but the information needed for further visual exploration. The visual system has no need to construct a complete and detailed representation of the world, they suggest, because the world itself can serve as our external memory. Our sense of the completeness and detail of the visual scene is based on our brain’s sensorimotor programs for visual exploration. Whatever the final explanation for change blindness ends up being, the fact that we don’t notice even large changes if these occur between fixations suggests that our brains probably do not construct complete and detailed inner representations of the visual scene in front of us. This is just one of many illustrations of the general fact that there is a significant absence of any explanatory isomorphism between the pattern of neural activity that constitutes the vehicle of a representation and the content of that representation. (Philosophers call this the vehicle/content distinction.) There is a mismatch between the content of our experience (the feeling of a completely filled in scene) and the neurological process that causes or constitutes it. We can describe this in philosophical terms by noting that the content of a mental representation is often propositional. Some pattern of neural activity in the brain represents that there’s a full page of script, or that there’s a complete visual scene, and it does this without having to be continuous, or completely filled in, or isomorphic with the content of the experience in any way. The brain represents that something is the case independently of any intrinsic properties of the pattern of neural activity that is doing the representing. [This very feature of the brain’s ability to represent things is what permits the range, open-endedness and complexity of the thoughts we can have, and points to the existence of some language-like code in at least some subsystems of the brain. The relationship between this language-like code and its neural basis will be systematic but totally arbitrary from the neuronal point of view.] This is what has convinced many researchers and philosophers that there will never be a reduction of mental states to the intrinsic neurobiological properties of patterns of neural activity in the brain. I am myself one of the many philosophers who finds the quest for the NCC, as originally presented, to be a misconceived project. But I’d like to end this section with an argument sometimes offered in favor of the NCC project, because I think it’s an argument that has some merit. 24 The argument goes something like this. The NCC, if there is one, will not be a simple structure, like a single set of neurons in the monkey’s inferior temporal cortex. If there is a NCC, it will no doubt be a very complex reality, involving certain levels of activity in certain kinds of neurons in certain cortical layers in particular anatomical structures of the brain with re-entrant connections to certain other parts of the brain, whose activity is synchronized in certain ways, etc., etc. So when a skeptic about neural reductionism says “How can any pattern of neural activity possibly explain conscious experience?” this might be a little like asking the similar question: “How can the activity of simple nonliving things like molecules possibly explain life?” We no longer take this question about life seriously. Now that we understand the chemical nature of genes, the great subtlety, sophistication and variety of protein molecules, the elaborate nature of the control mechanisms that turn genes on and off, and the complicated way that proteins interact with and modify other proteins, the question loses its meaning. We realize that these complex processes of metabolism, replication, gene expression,and so on are life. In a similar way, the neuroscientist looking for a reductionistic, physicalist explanation of consciousness may reasonably believe that once we have the full science of the NCC we’ll see that that complex pattern of activity and connectivity just is consciousness. 6. Combining the Functionalist and Neurobiological Approaches One way to accommodate the requirement for an explanatory isomorphism between the NCC and conscious experience, without committing vehicle/content confusions, would be to suggest that the isomorphism holds only at the level of informational content. In other words, one could admit that there is no reductionistic explanation of how intrinsic electrical or chemical properties of neurons constitute or cause conscious experience, but then claim that it’s similarity of informational content that gives the known correlations their explanatory force. Crick and Koch may have had something like this in mind in their NCC articles. They said, for example, that “whenever some information is represented in the NCC it is represented in consciousness.” (2003:35) So the idea would be something like this. You don’t just look for a set of neurons whose firing pattern covaries in a systematic way with the subject’s report of a certain conscious experience . You look for one that covaries in that way and has the same informational content as the conscious experience. If you discover this, you might think, you’ve really hit pay dirt – you’ve discovered the place in the brain where the conscious experience happens, and you’ve got an explanation why the experience is of what it’s of—why it’s an experience of seeing that the door in front of you is open, for example. But if this is what the program calls for, then the explanation of conscious experience will be a functional explanation, not a narrowly neurobiological explanation. This is because a pattern of neural activity in the brain has no informational content whatsoever all by itself. It gets its informational content from the role it plays in some larger arrangement that includes it. Information is a relation. To see this, let’s look at the thermostat example again. [slide 42. Thermostat Again] 25 Intrinsic electrochemical properties of neurons and patterns of neural activity in the brain are like the coefficients of expansion of the two metals in the bimetallic strip. They only carry information when they’re in a certain arrangement with other things. They contain no information all by themselves. They also cause and explain the behavior of the larger system of which they’re a part only when things are arranged in that very special way. (It’s for this reason that neuroscience is broadly functional in its orientation. Remember how the reductionistic approach we looked at in the maze learning example combined functional decomposition and neural localization.) [One of the interesting details about the Logothetis experiments is that the receptive-field properties of monitored single neurons depended on what the animal as a whole was doing. In these experiments the monkey’s head was restrained and it was trained to maintain fixation on a certain spot. But even in these restrained conditions it was found (1989, 1996) that the response properties of neurons thought to be the NCC “were influenced by the perceptual requirements of the task” (1996:551). Some of the cells that responded preferentially to the direction of motion or the orientation of a grating when the monkey’s task was to discriminate these features, showed no such preferences when their receptive fields were mapped conventionally during a fixation task. Other studies on alert monkeys have shown that attention and the relevance of a stimulus for the performance of a behavioral task can have a significant effect on the responses of individual neurons (Treue & Maunsell 1996; Moran & Desimone 1985; Haenny et al. 1988). The lesson we should draw from this is one insisted on by Francisco Varela: there is no way to establish the receptive field contents of individual neurons or neural networks independently of the sensorimotor context of the animal as a whole (1984; 1991).] As a way to summarize and reflect on what we’ve seen so far, I’d like us to consider the question of consciousness in split brain patients. [slides 43-52 Split Brain Patients] [slide 53. Which Method Is Appropriate? (1) Is the RH conscious, on functionalist grounds? (2) Is the RH not conscious, on functionalist grounds? (Can we ever be sure that an unreported event is unconscious? How would we know whether it’s unconscious or conscious but not remembered?) [In some cases, apparently unconscious events may be momentarily conscious, but so quickly or vaguely so that we can’t recall them even a few seconds later. (Sperling, iconic memory.) William James understood this problem, and suggested that there may be no unconscious psychological processes at all! (James 1890. Baars p. 6)] 26 (3) Should we look for the NCC in the LH, and then see if a similar neural pattern of activity occurs in the RH? If we succeed in this, can we determine whether the RH is conscious on neurobiological grounds?] 7. Meaning and the Importance of the Self [slide 54. In one of their recent articles, Crick & Koch say the following: “An important problem neglected by neuroscientists is the problem of meaning. Neuroscientists are apt to assume that if they can see that a neuron’s firing [in a monkey’s brain, for example] is roughly correlated with some aspect of the visual scene, such as an oriented line, then that firing must be part of the neural correlate of the seen line. They assume that because they, as outside observers, are conscious of the correlation, the firing must be part of the NCC. [But] this by no means follows.” ((2003:48)] The “problem of meaning” that Crick & Koch refer to here is an essential aspect of the puzzle of consciousness. Whenever there is conscious experience, there is a subject of the experience, who is conscious of something as being a certain way. Conscious experience is always of something (either in the external environment, our body, or our memory). But there’s also always a subjective pole, so to speak. We’re always also conscious (at least to some degree) of ourself having the experience. Conscious experience presents something as appearing a certain way to us. How does this happen? Suppose you have a computer running a program that monitors the inventory at a supermarket. Given a string of 0s & 1s as input (a 6-oz. can of Campbell’s tomato soup has just been scanned by the optical scanner at a checkout stand), the computer will go through a series of computations and emit an output (the inventory count of that item in stock has been adjusted downward by one can). Thus, the input string of 0s & 1s represents a can of Campbell’s tomato soup being sold, and the output string of 0s & 1s represents the amount of Campbell’s tomato soup still in stock. When the manager checks the computer for a report on the available stock of that item, the computer “reports” that the present stock is such and such, and it does so because “it has been told” (by the checkout scanners) that 25 cans have been sold so far today. The relation between input & output is a physical, causal relation. It makes no difference to the software program the computer is running what the strings of 0s and 1s mean. If the input string had meant the direction and speed of wind at the local airport, the computer would have gone through exactly the same physical computational process and produced the same output string. What the input and output strings stand for is irrelevant to the computation. [Jaegwon Kim 1996] So there’s no meaning for the computer. When the computer is hooked up into the supermarket situation in the right way, its input carries context-relative information about its environment. The input string carries information about the can of soup being sold because it was caused by the can of soup being sold, in much the same way as the curvature of the bimetallic strip in the thermostat carries information about the air 27 temperature in the surrounding room, because it is caused by the air temperature in the surrounding room. But the information has no meaning for the computer itself. Let’s assume that our brains produce representations of external objects and events, like the spatial maps in the hippocampus of the mouse’s brain. The question is, Who reads the maps? How does the brain produce the subject of conscious experience, the experienER for whom or to whom experience presents something as having some property? Input and output strings have no meaning for the computer, but conscious experience always has meaning for the organism having it. Why is this? What makes the difference? How do brainstates, unlike computer states, give rise to an experiencing subject for whom the information the brainstate carries has some particular meaning? The answer to this question has 3 parts: information, meaning, and self-representation. We’ve already talked about information. Our brains contain a large collection of specialized information-processors, controlling everything from heart rate to voluntary activity. We can see how patterns of neural activity in various parts of the brain can carry information about the environmental variables that cause them. Information is simply the nonrandom covariance between the properties of two communicating systems, like the nonrandom correlation between the presence of some feature in a stimulus and the firing of certain cortical neurons. Information is everywhere. But we can’t speak of meaning, or cognitive content, or representation unless the system that contains an informational state can use it in some way. [slide 55. A Definition of Meaning A pattern of neural activity in a subsystem of the brain can be said to have cognitive content, or to be a representation, when that pattern of activity is causally correlated with the presence of a particular environmental feature and the presence of the state modifies the behavior of the organism in ways specifically adaptive to that environmental feature. ] The simplest kind of representational state is embedded in a single, fixed adaptive behavior. A well-known example of this kind of representation exists in the frog brain. There are specific neurons in the visual cortex of a frog that are excited only by small moving objects in its visual field, and that produce “fly-catching” movements of the frog’s tongue. In this situation we can say something about what the informational state of the neurons means for the frog. It carries information about or indicates anything that causes it, under any description. (a fly, the set of cells making up the fly, the set of molecules making up the fly. A fly-looking moving object like a BB. The molecules the BB is made of, the atoms the BB is made of. Etc.] But it only means, or represents what it has the function of indicating. To the frog, the stimulus that sets off its fly-catching tongue movement has a meaning that we would express with a word like “food” or “prey”. (And because it has this more restricted meaning for the frog, it can be mistaken, as it is when the frog goes into this representational state in the presence of a BB rather than a fly.) 28 [slide 56. Meaning for a Frog] So, a brainstate, when it’s in a particular environmental context can carry information, and the behavioral repertoire of the organism whose brain it is determines the meaning that that information can have for it. But even when we know a certain pattern of neural activity somewhere in a creature’s brain carries information about something in the environment, and we know that the information is modifying the behavior of the organism in some way, we can’t know exactly what meaning that information-carrying brainstate has for the creature. [slide 57. What Does What the Pigeon Sees Mean to the Pigeon?] The meaning that a neural state can have for an organism is limited by the uses the organism can make of it. The pigeon’s behavior shows that it can discriminate the door with a prime number on it, but not that it discriminates it as a door with a prime number on it. The third element is self-representation. Consciousness as we know it is self-referential. The brain will not produce conscious awareness unless the nervous system also generates a representation of self—a representation that establishes a “point of view”. The neurologist Antonio Damasio has emphasized that the neurobiological mechanisms for visual awareness, for example, are essentially interconnected with the mechanisms for representing oneself as a thing that has experiences, feels, remembers and plans; as a thing occupying space and enduring through time. Damasio’s ideas come from decades of observing the ways in which consciousness is related to self-representation, and how that in turn is related to body-representation. Body representation, which systematically integrates environmental stimulation and body-state information, provides the scaffolding for self-representation, and self-representation is the anchor point for conscious awareness. We have already seen how important memory is for conscious awareness. Someone who is not forming autobiographical memories of their current experience may be judged to be conscious by others who are observing their speech and behavior, but they will not be judged to be conscious by themselves. There is no way for us to distinguish in ourselves between being unconscious and being conscious without memory. Similarly, a person who has deficits related to self-representation, or an infant whose self-representational capacities and “theory of mind” are in their early stages, will not be conscious in the fully human sense of the word. There is a growing body of research on deficits related to self-representation and on the development of self-representation and “theory of mind” in infancy and childhood. I believe this research will become increasingly important in the future scientific study of consciousness, so it’s especially exciting that we will have the opportunity to hear about developmental cognitive neuroscience from two of our speakers today. The development of self-consciousness and a full theory of mind is probably unique to humans (though elements of both are present in some non-human primates), and these are, I believe, a key to future progress in the neuroscience of consciousness. 29 8. Broader Implications Consciousness fits uneasily into our scientific conception of the natural world. On the most common scientific conception of nature, the natural world is the physical world. But given our experience of consciousness and our common sense understanding of it, it’s not easy to see how consciousness can be something physical. It seems that to understand the place of consciousness in the natural order, we must either revise our conception of consciousness, or revise our conception of nature to include the non- physical. As long as we’re thinking of “consciousness” just as information processing --the ability to discriminate stimuli, or to monitor internal states, or to control behavior -consciousness can be explained in computational terms. I have tried to give you some idea of how this would go, using simple examples like thermostats and the frog’s flycatching neural circuits. The task is to explain certain behavioral or cognitive functions that play a role in the production of behavior. To explain the performance of such a function, one need only specify a mechanism that plays the relevant role. And there is good reason to believe that neural or computational mechanisms can play those roles, as we saw in the water maze example. The hard problem of consciousness is the problem of experience. Human beings have subjective experience. Conscious experience has both an objective and a subjective pole. It is an experience of something in the world as being a certain way, accompanied by an experience of myself as taking it in that way. This aspect of consciousness, its internal aspect so to speak, does not seem to be required by anything we know about either the neurobiology or the computational states of the brain. We can’t really say at this point, why the processes that go on in the brain don’t just go on “in the dark”, without consciousness. This is the central mystery, the hard problem, of consciousness. It’s possible that consciousness is a physical phenomenon, and that there is a reductionist explanation of how physical states produce consciousness, but that human beings will never find it. Maybe we’ve run up against the cognitive limitations of our species. Electromagnetism is a perfectly natural physical phenomenon, even though oysters can’t understand it. Oysters just don’t have the right cognitive equipment to understand that kind of thing. Likewise, since human beings have a particular kind of perceptual system and particular kinds of computational capacities, it could be that consciousness is a straightforwardly physical phenomenon that we simply don’t have the right cognitive equipment to understand. (Colin McGinn, Steven Pinker) This is certainly a logical possibility. But it would be premature to adopt this position as anything more than a bare possibility at this point in time. We are in the very early stages of the neuroscience of consciousness. We may not even have the phenomenon of consciousness identified properly. One of the interesting things about studying the history of science is that we see that over and over again, people’s sense that they have a 30 perfectly good understanding of what some familiar word refers to, turns out to be mistaken. Here’s an example. In the early stages of a scientific investigation, a thing’s category membership is determined largely by similarities in easy-to-observe properties. For many centuries in the pre-modern era the category “fire” included a wide range of phenomena, all of which involved the giving off of heat or light. But as physics and chemistry progressed in the modern period, more theoretically-informed properties were used to determine category membership, and the phenomena had to be regrouped. [slide 58. The Definition of “Fire” ] We’re in the same situation today with consciousness. We don’t have a scientific theory of consciousness, so we don’t have the proper theoretically-informed properties for determining category membership. There are familiar mental phenomena about which we cannot even say for certain at this point whether they’re conscious or not, like the postsurgery experience I described, or the RH information-processing of split-brain patients. This is a clear symptom of the absence of theory, and of the fact that we’re at the very beginning of the science of consciousness. It’s very possible that consciousness is a physical phenomenon, and that we will one day have a reductionistic scientific explanation of it. Still, there are things about consciousness that do seem to make it a special case, something especially difficult to reduce to physical properties. Consciousness seems to resist reductionist explanation in a way that other phenomena do not. Maybe consciousness is not a physical phenomenon. It could be a natural, but non-physical, phenomenon—something that is completely natural but cannot be given the type of reductionistic explanation usually pursued in the sciences. Some very great philosophers and scientists, both past and present, have held this view. Prof. Hoffman’s theory, for example, is a non-physicalist theory of consciousness. Non-physicalists about consciousness sometimes repeat an observation that goes back to Bertrand Russell and Whitehead in the early years of the 20th c. Russell pointed out in The Analysis of Matter(1927) that physics characterizes physical entities and properties by their relations to one another. For example, a quark is characterized by its relations to other physical entities, and a property such as mass is characterized by an associated dispositional role, such as the tendency to resist acceleration. Physics says nothing about the intrinsic nature of these entities and properties. Normally, where we have relations and dispositions, we expect some underlying intrinsic properties that are the basis for the dispositions and relations. (An intrinsic property is a property an entity has independently of its relations to other entities.) Physics is silent about the intrinsic nature of a quark, or about the intrinsic properties that play the role associated with mass. The properties that figure in the fundamental theories of physics are all dispositional or relational properties. So maybe every real individual entity in the universe has an inside to it, as well as its external relations described by the laws of physics. If this were true, then the intrinsic, qualitative features of subjective experience would be adumbrated in 31 the simpler intrinsic properties of all other entities in the universe, and consciousness would be the key to a new metaphysics of nature. Two philosophers, Chalmers & Jackson, have reintroduced Russell’s observation into the contemporary discussion of consciousness. They argue that by the very nature of physical explanation, physical accounts explain only structure and function, where the relevant structures are spatiotemporal structures, and the relevant functions are causal relations in the production of a system’s behavior. Explaining structures and functions cannot, as a matter of principle, suffice to explain the intrinsic, qualitative features of subjective experience. So subjective experience is not physical. (Chalmers 2002:248) [slide 59. The Explanation Argument 1. Physical accounts explain at most structure and function. 2. Explaining structure and function does not suffice to explain consciousness. 3. No physical account can explain consciousness. (1,2) 4. What cannot be physically explained is not physical. Therefore, Consciousness is not physical. (3,4) Consciousness as subjective experience has so far resisted reductionistic scientific explanation in a way that other phenomena have not. There are several alternative conclusions we might draw from this fact. Maybe consciousness will be reduced to physical properties and laws in the future, and the reason we can’t envision exactly how it will go at this point in history is that we’re just at the beginning of this scientific project. People who hold this view feel they have history on their side. A second possibility is that consciousness is physical and a reductionistic explanation is possible in principle, but humans don’t have the cognitive equipment they need to do it, so we won’t ever know what the reductionistic explanation of consciousness is, even though one exists. A scientific explanation of consciousness will elude us by reason of our own cognitive limitations. A third conclusion we might draw is the conclusion Prof. Hoffman draws in the paper he’ll be presenting today, that physicalism is simply mistaken, that consciousness is fundamental and physical properties and laws are derived from consciousness. The fact that information processing in humans is accompanied by subjective experience is not necessitated or entailed by anything in our current neurobiology or cognitive science. It is thus far unexplained. If this remains the case, it may be mean that consciousness involves something ontologically novel in the universe. The philosopher Saul Kripke expressed the thought this way: After choosing all the physical laws (plus physical constants and initial/boundary conditions) for our universe, God had to make further decisions, having to do with consciousness. In other words there are ontologically fundamental features of our universe over and above the features characterized by physics. Just as we take spacetime, charge and mass to be fundamental 32 features that cannot be further explained, perhaps consciousness is a fundamental feature in the same way. Let me set out some possible ways of understanding the relationship of consciousness to the physical in a schematic way. It could be that consciousness emerges from the physical when physical entities are arranged in the right kinds of ways. This is emergence in the weak sense. 33