According to Chalmers, a theory of consciousness should, at the very least, "give the conditions under which physical processes give rise to consciousness, it should specify just what sort of experience is associated. And we would like the theory to explain how it arises, so that the emergence of consciousness seems intelligible rather than magical. In the end, we would like the theory to enable us to see consciousness as an integral part of the natural world. Currently it may be hard to see what such a theory would be like, but without such a theory we could not be said to fully understand consciousness."(p5) To what extent does Chalmers make progress, judged by his own criteria? He touches on a few of the scientific findings which might reveal physical processes which "give rise to" consciousness, but that is not what his book is about. His real concern is to suggest what an appropriate theory of consciousness would be like. This is a book by a philosopher addressed primarily to philosophers. Chalmers writes well and many of the chapters will be accessible to the general reader. However, non-philosophers will find the technical nature of some of the chapters heavy going (and might be better served by his more succinct presentation of his views in "Facing up to the problem of consciousness", in the Journal of Consciousness Studies, 2, 200-219, 1995). Given the fluid state of consciousness studies, progress towards the "shape" of an appropriate theory, fully argued, will nevertheless be of interest to many.
In the 20th Century, philosophy of mind has been predominantly reductionist, arguing that consciousness will eventually be shown to be nothing more than a physical state or function of the brain. By contrast, Chalmers joins a small, but growing band of theorists who maintain that consciousness can only be understood within a nonreductionist science of the mind. His central argument against reductionism hinges on the claim that consciousness is "naturally supervenient" but not "logically supervenient" on physical states. That is, it actually supervenes on brain activity, but it is conceivable that it might not. This, he argues, makes consciousness different from all other properties, including "emergent" biological properties such as "life." If we know all the microproperties of a physical system and their relationships, we necessarily know the macroproperties - for example, according to Chalmers, given that the physical facts are as they are, it is inconceivable that life could be other than it is. In this sense, a higher order property such as "life" is nothing more than its combined, physical constituents. By contrast, he can conceive of a zombie which is physically indistinguishable from himself, but which lacks consciousness. Consequently, consciousness cannot be a physical property. Nor can it be reduced to physical properties. There are, he claims, no other nonphysical properties in this sense.
This tension between what is "natural" (or actual) and what is "logical" (or logically possible) runs throughout the book, sometimes uncomfortably. Many scientists, for example, will be dubious about the actual knowledge of macroproperties that is provided by knowledge of microproperties. Morphogenesis (the creation of life forms) for example, does not seem to be very well accounted for by gene theory (at present), let alone quantum mechanical events. So even physical wholes might, in some sense, exhibit properties not reducible to the sum of the parts. Chalmers' response to this would be that in such cases we have insufficient knowledge of the parts; if we had complete knowledge of genes, subatomic particles, and so on, it is logically inconceivable that macroproperties such as morphogenesis would not be understood. This insistence on complete knowledge provides Chalmers with an escape from any possible counterexample, as he can always claim these to be cases where our knowledge of microproperties is incomplete. However, in actuality, our ability to account for all complex forms and functions in terms of the activities of physical microproperties seems to be more a matter of faith, than logical necessity. Functions, for example, often need to be understood not just in terms of their constituent processes but also in terms of the environments which embed them. And the reducibility of psychological beliefs, desires and so on to physical microproperties seems counterintuitive; how, for example, could even a complete knowledge of quantum mechanical events provide an understanding of why one needs to catch a 94 bus?
A simple, definitive argument against the reducibility of phenomenal properties to physical ones would nevertheless be very welcome, and this is Chalmers' more central concern. Unfortunately, the force of Chalmers' argument rests entirely on what he finds to be conceivable - and its consequent weakness, that different theorists might claim to be able to conceive of entirely different things. Zombies are not just functionally indistinguishable but also physically indistinguishable from humans. A reductionist might therefore claim that he cannot conceive of such a creature lacking consciousness. If consciousness just is a state of the brain, then it inconceivable that a creature with a brain in that state could lack it. Faced with this stand-off, the only recourse would seem to be to take a vote (about what is conceivable) - hardly a satisfactory way to resolve the issue. Chalmers nevertheless goes on to assemble a range of standard arguments against reductionism from the literature, elaborates these where necessary, and gives a critique of the counterarguments. His mastery of philosophical terrain is impressive, and taken together, his arguments pose a serious challenge for materialist theories of mind.
But what is his alternative, nonreductionist, theory? In Chalmers' ontology, there are two kinds of property, physical properties and phenomenal ones, and there are two kinds of law, physical laws and psychophysical laws or "bridging principles" which govern the way phenomenal properties "naturally supervene" on physical ones. Nothing else, he claims, is required for a complete account of consciousness and mind. Although Chalmers presents this as a novel, nonreductionist position, this much is fairly standard. A nonreductionist might claim more, for example, the existence of mental laws. But it would be hard for a nonreductionist to claim less than the existence of physical and phenomenal properties along with some way of relating the two. Positing the existence of "psychophysical laws" might raise some eyebrows in philosophy, but the search for such laws is as old as experimental psychology (witness the "Weber-Fechner law" and "Stevens' power law" which relate the way changes in physical stimuli are translated into perceived changes). Muller, in 1896, also gives an instructive formal description of psychophysical laws (which is rather lost in Chalmers' presentation in note 12, on page 385).
Not all of Chalmers' ontology, however, is so straightforward. Sometimes for example Chalmers describes his position as a "double-aspect" theory - which poses the question "double-aspects of what?" According to Chalmers, phenomenal properties and their physical correlates in the brain will be structurally coherent, in the sense that they will encode the same information. On these grounds Chalmers justifiably describes his position as a "double-aspect theory of information." This much is plausible (I presented an identical "dual-aspect theory of information" in "Is human information processing conscious?" in The Behavioral and Brain Sciences, in 1991). However, at other times, to avoid positing some transcendental ground for physical and phenomenal properties Chalmers describes his position as "naturalistic dualism" - in which consciousness becomes "basic" in the same sense that energy is basic in physics. This raises the question, "If phenomenal and physical properties are equally basic, distinct, and not grounded in something more fundamental, then what is it that relates them to each other so precisely?" Alternatively, if phenomenal properties "supervene" on physical ones (as he argues throughout the book), then why regard the phenomenal properties as "basic"? It is up to Chalmers to decide which position he wishes to defend (the rest us will have to wait and see).
Further problems arise from Chalmers' psychofunctionalism - perhaps the most innovative (and controversial) aspect of his theory. Although Chalmers refers in much of his work to the need for "psychophysical" bridging laws, it eventually becomes clear that these are psychofunctional rather than psychophysical. That is, he believes there to be an invariant relationship between physical and phenomenal properties, but it is the way a system functions, and not its physical embodiment that determines what that system experiences. According to Chalmers a machine that functions in a way that is indistinguishable from humans has experiences that are indistinguishable from humans (a version of "strong AI"). This would be true whether the system is made out of silicon chips, beer cans, or the population of China - provided only that in their detailed activity, these systems instantiate the same causal relationships, ie., function in the same way. Chalmers' combination of a nonreductive phenomenology with standard functionalism is, to my knowledge, novel within philosophy of mind (standard functionalism claims that consciousness is nothing more than brain functioning) - and, given that the phenomenology of consciousness has proved to be the stumbling block of functionalism, it is not surprising that Chalmers' position has attracted considerable interest.
Chalmers develops this position from two thought experiments, which he describes as "fading qualia" and "dancing qualia." In these he considers the familiar scenario in which the neurons of the brain are gradually replaced by silicon chips which exactly replace the functioning of the neurons they replace. As the replacements progress, do the qualia gradually fade? Or, if one were able to switch between one's normal brain and a replacement silicon brain (with exactly the same functions) would the qualia dance? According to Chalmers if one replaced the functions exactly one could not notice the difference either externally in terms of behaviour, or internally in terms of what one experiences. One would, after all, have to report the same things - otherwise the functioning of the silicon systems would not be the same as the neural systems they replace. Hence, functioning of certain sorts is necessarily accompanied by experiences of certain sorts (there is no way to distinguish any difference).
This argument was initially put in the special issue of the Journal of Consciousness Studies (mentioned above), and in my own commentary (in the same issue) I suggested that Chalmers had presented the options in the silicon replacement experiments in an unnecessarily restrictive way. In actual practice, one could envisage replacing the absent functioning of some damaged brain circuitry with silicon or other hardware in an otherwise normally functioning human being - and leave open the question of whether repairing the functioning (defined in information processing terms) would also replace any missing experience. For example, it might be possible to devise a cortical implant for blindsight. Patients with blindsight cannot see stimuli presented to their blind hemifield, but when forced to guess, they can identify them. A cortical implant which restored the ability of visual circuitry to communicate the results of pattern recognition to the systems used for reporting might restore the ability to identify stimuli with ease. But whether it restored the ability to have a visual experience of the stimuli remains an open question. If so, functioning can be dissociated from experience experimentally, at least in principle.
Chalmers' response to this was that the two outcomes are not functionally equivalent. In one outcome subjects report that they can see the stimulus as well as identify it and in the other, that they can't see the stimulus. My response would be that what is at stake is what subjects can actually see, not what they can report. In the human brain, the visual system is dissociable, at least in part, from the systems serving language and speech (as blindsight itself demonstrates). The aim of the "cortical implant for blindsight experiment" is to determine whether the input-output functioning of the visual system (discrimination, pattern recognition, etc) can be restored without restoring the accompanying visual experience. Subjects' discrimination ability can be assessed (by external observers) from their behaviour, but only the subjects themselves can tell us about what they experience. Leaving subjects free to tell us one way or the other is therefore central to the experiment. In Chalmers' version of the thought experiment the visual system and the language/speech system are treated as non-dissociable (ie as one system). Consequently, "functional equivalence" requires identical verbal reports by definition. Set up this way, the claim that functioning cannnot be dissociated from experience becomes unfalsifiable - which, for science, removes the value of the experiment.
Whatever one may think about the "fading/dancing qualia" arguments, the view that "bodies don't matter" for what we experience is highly counterintuitive. On Chalmers' account, not just machines made of silicon chips might experience in the way that humans do, but so would virtual minds (instantiated in the symbol manipulations of programmes) and even systems consisting of symbols written on bits of paper by the population of China, provided only that the causal relationships governing the creation of those symbols, simulate those of the human mind's symbol manipulations accurately. Processes within the human brain normally thought of as unconscious would also have to be conscious in Chalmers' system (by virtue of their functioning) - in which case the conscious/nonconscious distinction loses its meaning. The theoretical cost of this position to consciousness studies is considerable. If the conscious/nonconscious distinction cannot be made, how could one investigate the conditions for consciousness in the human brain - which rely on contrasts between neural conditions adequate or not adequate for conscious experience? How could one make sense of the extensive experimental literature on the differences between preconscious, conscious and unconscious processing? And what of psychodynamic theory - is all talk of a personal or transpersonal unconscious just muddled thinking?
Note that Chalmers is forced into this uncompromising position by his fading/dancing qualia argument (whatever functions is conscious by virtue of its functioning). Given this, all brain functions must be conscious. Consequently, he maintains that those functions which do not seem to enter into our consciousness must be autonomously conscious (they are conscious to themselves). This leads to the extravagant claim that there are as many distinct consciousnesses cohabiting in the human brain as there are distinct functions.
Nor does Chalmers see any reason to draw the line at brains or systems which simulate the functioning of the brain. If consciousness of given sorts is invariably associated with functioning of given sorts then all forms of functioning are associated with experiences, irrespective of their embodiment. This "panpsychofunctionalism" (my term for this) is quite different from panpsychism (the view that all material forms are accompanied by forms of experience). If true, then not only do thermostats experience in ways that relate to their function (sensing hot and cold), but so does rain falling in that it functions to make the earth wet - and even rainbows experience something relating to their production of beautiful sensations in the human mind. The central difficulty for Chalmers is that functioning is observer-relative. Chalmers' defence is that the structure of physical systems does, to some extent, constrain their potential functioning. But this really misses the point. Does the chinese statue on my desk really experience something different when it functions as an incense holder rather than as a decorative object? And if it does both simultaneously, does it really simultaneously have both experiences? As John Searle has commented in his recent New York Times review of Chalmers' book, "It is rather as if someone got the result 2+2=7 and said "Well maybe 2 plus 2 does equal 7."
In many ways this book provides an impressive display of expertise and argument and it has been well received by many (although not all) philosophers of mind. But it gives rather confused directions, in my view, to some future theory of consciousness. I find it hard to take issue with its antireductionist stance, or its view that consciousness is somehow "basic", or that "information" may have a central role to play in any theory of how phenomenology relates to the brain (I have argued for the same positions myself). But "double-aspect theory" isn't interchangeable with "naturalistic dualism", conscious processes do need to be distinguished from nonconscious ones in the brain, thought experiments with unfalsifiable conclusions need to be distinguished from actual experiments, and I very much doubt that the stiff whisky I have just swallowed is aware of its function in sustaining my ability to write this review. Philosophers, quite rightly, feel free to contemplate possible worlds as well as actual ones. But once one has explored where the logic of an argument leads, one still has to consider whether to temper logic with common-sense.