Tye. Furthermore, he claims that there is something it is like to be Fred, and also something different that it is like to be Fred when he is embarrassed: "What it is like to be the butt of Raucous Roger's jokes is acutely embarrassing. But what it is like to be Friendly Fred is not acutely embarrassing, even though Friendly Fred is the butt of Raucous Roger's jokes" (1995b, 125). In each of these cases, we are to suppose that a back pain and embarrassment are experiences on a par with being someone. But this is certainly a category mistake. There is nothing it is like to be someone if paradigm cases of what it is like to experience something are to be pains. Rather, there is only something it is like to be someone in the sense that there is a collection of phenomenal states, with a certain continuity over time, experienced by an individual. What it is like to be Tye is therefore not at all a property like the pain of backache. What it is like to be Tye is at most a collection of properties, along with dispositions to certain moods, typical responses, typical cares and thoughts, and so on. Furthermore, a causal, and not a representational, explanation is suitable for this difference: again, there are different causal events, one a specific pain event and one a collection of events (most of which are of like complexity and duration as the pain); or one a specific instance of suffering (among other things) and the autonomic changes caused by embarrassment, and the other a collection of events that compose the phenomenal experience of Friendly Fred. To explain these, different causal stories must be told. The basic causal story alone is not going to be wholly sufficient: we need to explain why these are experiences for Tye and for Fred, and we need some account of what makes pain pain, and what makes these other phenomenal experiences what they are. But do we need representations plugged into a belief-desire system to do that?
These difficulties arise because of ambiguities in the notions of fulfilling certain normative constraints and being a representation. We can identify that something has a biological function, for example, and as such it can also ground some representations. In contrast, the notion of representation as a mental event that readily plays a role in a belief-desire system would seem to require the idea of a token that is "standing in" for something else inside a representational system (and hence the representational theory of consciousness is prima facie consistent with the token role theory). In these cases, representation is something more than just satisfying the right kind of normative constraints. All of this matters because, although we may take as a primitive fact that phenomenal experiences like anger are "about" something, we need to understand the exact nature of this about-ness.
This is the fourth puzzle: In what sense of "representation" are phenomenal states representations? Our understanding of basic emotions, and affects like moods, shows that we do not need the notion of representation as a token in a representational system that itself is part of a belief-desire system. Can we use some weaker notion, like intentionality or teleofunction?
end p.168
 
10 A Systems-Based Teleofunctional Theory of Consciousness
Abstract: A theory of the relation between phenomenal experience and body states is introduced that solves the puzzles discussed in Ch. 9. Using a systems-based theory of teleofunctions (or biological functions), phenomenal consciousness is found to supervene on those teleofunctions of which we are aware. The homeostatic intensity hypothesis is introduced to explain differences in the nature and intensity of experiences. These differences are explained by the different teleofunctions that are altered by, and the degree that they are altered by, some body event such as an emotion.
Keywords: biological function, emotion, homeostatic intensity, phenomenal consciousness, phenomenal experience, supervenience, teleofunction
Craig Delancey
Having posed four puzzles for phenomenal consciousness that are made acute by reflecting on the role and experience of the basic emotions and other affects, my goal in this chapter is to solve these puzzles. Some have suggested that consciousness is essentially affective (e.g., Cytowic 1998) but I doubt this (for example, to my own phenomenal experience, it seems plausible that there is such a thing as a color sensation that is in no direct way affective); for that reason, my goal in this chapter is to provide the outlines of a theory of the relation of phenomenal experience to body states. In other words, basic emotions and other affects point us toward neglected aspects of phenomenal consciousness, which in turn must be explained with a more general theory concerned with all phenomenal experiences.
I will proceed by taking a last look at epiphenomenalism and token role theory, and setting both aside. Instead, I suggest that a better theory of the relation of phenomenal experience to body states should describe a strong supervenience relation between the nature of those experiences and the teleofunctions of the body states that give rise to them. This requires a review of the notion of teleofunctions, and of contemporary accounts of function. I argue that a systems-based approach to teleofunctions is the best solution. I then provide both an argument for, and an application of, this approach by showing how it elegantly provides solutions to the four puzzles posed in the last chapter. Finally, I briefly discuss how this approach yields a better understanding of the notion of autonomy.
A Last Look at Epiphenomenalism and Token Role Theory
I have argued that our phenomenal experience of affects indicates that an inseparable experiential link exists between the function or role of an affect and its phenomenal qualities. This is some reason to reject epiphenomenalism and token role theory; however, it is not conclusive since phenomenology
end p.169
 
can err. But there are other reasons to reject either alternative. I take it that the standard objections to epiphenomenalism, familiar though they are, remain sufficient reasons to reject the view. First among these is the simple observation that if we make consciousness something that can play no functional role, then we seem to have placed it beyond even the possibility of scientific theory. This does not show that epiphenomenalism is wrong, but it does suggest that one should not resign to accept it unless there is no alternative. After all, when we know so little about the mind, and we are learning so much every year, we should delay indefinitely accepting any theory that, without anything like a substantive proof, counsels us to give up any hopes of having a natural account. Another reason to give up on epiphenomenalism is that there are reasons to believe that phenomenal conscious states do play (that is, are strongly related to) some role. Our experience of affects is that they are motivational. But there is also the simple matter that we can talk about our phenomenal experiences. This would suggest that they are playing some causal role—they are causing or helping to cause talk about them. This is not an airtight argument: we also talk about numbers, and yet 2 does not play a causal role in our world. However, we have some theories about the kinds of things that numbers are. They are logical constructions, which refer to whatever is shared by two cats, two dogs, two trees, and so on—for any two instances of items falling under some concept. We have no such account of phenomenal consciousness. It is a state, of some kind, in which a conscious agent can be. It is not a logical construction. For these and other reasons, I will not consider epiphenomenalism a viable approach.
These concerns do not apply to token role theory. My first puzzle was that our experience of emotions is inconsistent with both token role theory and epiphenomenalism. But, recognizing that first-person reports are not infallible, we might suppose that not only are inverted spectra possible, but that my experience of sorrow can be your experience of joy (although we both act the same at funerals or at parties). On such a view, the phenomenal experience is acting as a representation that could be replaced by any other phenomenal experience as long as the representational role is the same.52 This approach is objectionable because it fails to explain anything. First, it does not explain experiential distinctiveness or intensity. Second, we might still want to explain why it seems that our emotional experience feels inseparable from its motivational role. After all, the nature of experience is what is in question in the consciousness debate. Although first-person reports should never be taken as reliable, it is in large part the nature of the first-person report that we are trying to explain. Third, and much more important, separating the token that represents the experience from the experience it stands for threatens to eliminate the experience. For if my experience of joy could be switched with my experience of intense sadness, and the external roles be the same, then why isn't there anything more to the experience of joy than just the information that one is smiling? This objection is more subtle than it at first sounds. The token role theorist's first
end p.170
 
answer is presumably that there are other behaviors that joy causes besides smiling, but that all of these, not just smiling, constitute joy. But then, why suppose that all of these can send a completely different message—in particular, the message that intense sadness sends—to wherever consciousness happens? That is, even from a neurocomputational perspective, why assume that the token is separable from the experience, or at least from the thing reported? There may be something of the fallacy of what Daniel Dennett calls "the Cartesian theater" (1991) at play here: the view that there is a kind of theater in the brain somewhere, and in it consciousness happens. We imagine the various activities of sadness outside the theater, sending in some arbitrary message, which in the theater is by a kind of convention understood to mean "I am sad." The problem here is that both taking representation as the explanation of the experience and separating the representation from the experience seem to defeat the very purpose of the explanation. It would seem that no reference to the experience need be made at all. All that is being accounted for is the gross behavior of the affect and that we are able to report that we are aware of it. Paul Churchland is right to criticize the view that there is some strange essential token, the quale, that is needed to explain all this (1989b, 27-28); but it would be equally strange to suppose that all of this is experience not of the functional role, but rather of some token being sent to a brain theater—a token which can be replaced by any other.
Weak and Strong Supervenience
If we could account for the seeming relation between experience and role and motivation in these cases, we would have a theory that explained more. We must begin by explaining the kind of stronger relation we need to identify. My rejection of epiphenomenalism and of token role theory present a starting point in this endeavor. Token role theory outlined a kind of dependency and co-variance relation: that qualia play a role in a person's actions, but that these qualia could be moved around. This relation is very similar to weak supervenience.53 We can understand weak supervenience to mean that if some set of properties54 S weakly supervenes on another set of properties B, then:
In any one world, for any objects x and y, if x has the same properties of S that y has, then x has the same properties of B that y has.
More simply, weak supervenience states that in any world, if the supervenience relation holds, then anything that has the supervening properties has the base properties. So, if in some world the experience of seeing red weakly supervenes on a particular kind of neural brain state, then in this world whenever a person will have a particular experience of red that neural brain state occurs. The supervenience relation is weak because it specifies that things hold this way only in a single world, in a single complete
end p.171
 
Supervenience
Supervenience is a useful philosophical notion that has become a standard tool of the contemporary consciousness debate. The details can become quite complicated, but the fundamental idea is very simple. Supervenience allows us to formulate a relation of co-variance and dependence (Kim 1993d). We say that some set of properties or events supervenes upon another set of properties or events if there can be no change in the former without some change in the later. For example, the market capitalization of General Electric (the total value of all the shares of stock) supervenes on the stock market: there can be no change in the price of all of GE without there being some change in the market (I allow that identity is trivially a supervenience relation). Supervenience is thus a useful notion because often when we develop theories we aim to outline the way some things are dependent upon others, and the ways in which they change together (as a scientist might put it, we aim to discover the dependent variables in a system, and the nature of their dependency).
The notion is open to a great deal of refinement. The principal way we can sharpen the notion of supervenience is to specify the kind of possibility involved, and also the scope of this possibility. This really is just clarifying what we mean by "can" when we say that there can be no change in the supervening properties without some change in what they supervene upon. The kinds of possibility include things like logical possibility or physical possibility. As it is usually understood, something is logically possible if there is not a contradiction in its appropriate description. Thus, it is logically possible that Adam is five meters tall, even though neither he nor anyone else is five meters tall; but it is not logical possible that Adam has a mass of more than and less than 100 kilograms. The notion of natural or physical possibility is stronger. We would say that a situation is naturally possible if we mean that the occurrence of that situation does not contradict the laws of physics and whatever other sciences we believe properly describe the natural world. Now, it may be that in the natural history of our universe the naturally possible situation in question never occurs. But we can mean, by saying that it is (naturally) possible, that given a specific different history, with the same natural laws, the thing would have occurred. For example, if the universe had had the same natural laws but much more, or much less, mass, things might have worked out very differently.
way that things could be. In another world, that same brain state could result in a green experience.55
There are two issues here that need to be considered. First, token role theory can be even weaker than this, since some token role theorists entertain the possibility that your red experience here in this world is my green experience here in this world (although we both call it "red"). For my purposes, it is sufficient to criticize the stronger version of weak supervenience given above. Second, for other versions of token role theory, depending on
end p.172
 
what goes into the base properties, the relationship may be superficially stronger than weak supervenience. This is because there is something that stands as the representational token for the experience and this may be included in the things upon which the phenomenal experience supervenes. From the perspective of someone who denies weak supervenience, however, this approach will be practically the same, since the base properties will not include this token, but rather a larger set of things that constitute the functional role of the corresponding body state. That is, if the base properties of anger are taken to be the rise in blood pressure, the change in heart rate, the brain changes specific to anger, and so on, then the token that the token role theorist takes to represent anger—but which by their own supposition can be replaced by another, different token—can be excluded from the base properties because it is arbitrarily correlated and hence replaceable. The result would be, at its strongest, something like weak supervenience.
Given this characterization, my goal is to outline a relationship of strong(er) supervenience. Some properties S strongly supervene on some other properties B if:
For any two worlds, and for any objects x and y, if x in one world has the same properties of S that y in the other world has, then x has in its world the same properties of B that y has in its world.
This is stronger because it says that the supervenience relation must hold in every world. Strong supervenience is the kind of relation that a naturalist about phenomenal experience will be seeking if she expects to link, in any way stronger than just token role, the experience to the behavior it can cause or of which it can be a part. Note that strong supervenience as defined here is consistent both with phenomenal experience being identical to some brain states, and with the view that some natural law links the two (under some notions of natural law, strong supervenience will entail that if they are not identical, then such a law links the two). Strong supervenience is also consistent with our experience of the basic emotions and other affects. If our experience of anger is inextricably linked to a (motivation to) attack, then we should expect the relation to be one of strong supervenience of the experience upon the teleofunctional role of the corresponding body states: you cannot switch these kinds of experiences around, while keeping the body states and their role the same. There can be no zombie world, and no world where the experience of depression and joy are switched, because these things are necessarily linked to the roles they play. I am assuming, in these cases, that the notion of possibility and necessity is one of natural nomic possibility: that what is consistent across these worlds are natural laws of the kind that one hopes to uncover in a natural science. Finally, I should note that I believe that the occurrence of the kinds of systematic teleofunctions that I describe below are sufficient for the occurrence of the relevant phenomenal experiences, and this is stronger than the relationship
end p.173
 
of strong supervenience (which only outlines necessary conditions); however, to solve the puzzles posed in the last chapter this standard notion of supervenience will suffice.56
Getting the Function of Consciousness
Our experience is consistent with the claim that at least some aspects of affective experience strongly supervene upon the role that the affect is playing: any change in these phenomenal experiences should be explained by reference to a change in the functional role of the body state from which it arises. To make sense of the notion of an affective state playing a role, we need to identify what it is for that state to have a teleofunction (hereafter, I use this term to avoid any confusion with the use of "function" in its simpler, mathematical sense). The representational theory of phenomenal consciousness nicely captures our sense that our phenomenal experience is about something (if only because it strongly supervenes upon a physical state that is intentional): my experience of a sunset is of, or about, the actual sunset; my experience of pain is of, or about, my body state. But in linking phenomenal experience to representations, the problem of specifying teleofunction remains—although it can be rephrased as a problem of representational intentionality: How is it that my representations manage to be about something in the teleological sense? In a more general teleofunctional account, the question becomes how some capability can have a function which is specified seemingly in reference to future possible occurrences. This problem is now a familiar one, and the goal of solving it is the goal of providing an acceptable naturalist account of a kind of teleological explanation.
To briefly review the issue: to say that the teleofunction of fear, for example, is to motivate flight from some kinds of threat, is—on the surface of the claim—to refer to future events as an explanation of a present action or capability. For, on normal usage of such teleofunctional ascriptions, this is the teleofunction of the neural circuits underlying fear even before they ever so operate. Similarly, it seems correct to say that the teleofunction of sperm is to inseminate an egg, even if the sperm is had by an individual who has no children nor ever will have children. And, even more complicated a matter is that the future event in question need not even be possible in any strong sense: the function of the sperm of the last man remains to inseminate an egg, even if this were now historically impossible. Thus, the primary problem with teleofunctional explanations or ascriptions is that very weakly possible future events are used to explain present capabilities; but our best scientific explanations refer only to past events to explain all events or states. The challenge is to provide a theory of teleofunctions which coheres with our best naturalist approaches. Two approaches have become standard: etiological historical accounts, and systems-based accounts.
end p.174
 
Etiological Accounts of Teleology
Perhaps the most obvious sense in which teleofunctions can arise is in the design of objects (Kitcher 1993). A hammer can be manufactured with its form specifically so that it can hammer nails; and it seems therefore that hammering nails is uncontroversially its teleofunction. However, some artifacts seem to come to have their teleofunctional form through trial and error on the part of the manufacturer (Bigelow and Pargetter 1987, 185-186); and, of course, almost all biological systems are not designed. However, there may be a sense in which we can think of biological systems (and perhaps also trial-and-error inventions) as being "designed": they have been "designed" by evolution.
The most developed theories of teleofunctions have been etiological ones: approaches that attempt to explain teleofunctions by reference to a history of the thing with the function. An early and clear statement of an etiological account is found in Larry Wright, who, expanding on ideas in the work of Charles Taylor (1964), argues that "teleological explanations are causal in the very broad sense that they explain what produces or brings about the behavior in question; they offer an etiology" (1976, 25). Wright's teleological explanation of functions is of the form (81):
The function of X is Z iff:
(i)  
Z is a consequence (result) of X's being there,
(ii)  
X is there because it does (results in) Z.
Thus, the function of the heart is to pump blood because (i) the pumping of blood occurs because of the heart; and (ii) the heart is there because it does pump blood.
This is a useful analysis. It has come under some major criticism because it allows us to count as teleofunctions some things which seem quite obviously not to be such (Boorse 1976). But, setting these aside for a moment, the trick is to explain clause (ii): in what sense is the heart said to be there because it pumps blood? The foremost answer to this question in an etiological account has been to turn to evolutionary theory; variations on this approach have been developed by Peter Godfrey-Smith (1994), Paul Griffiths (1993), Philip Kitcher (1993), R. G. Millikan (1984, 1993), K. Neander (1991), and others. The basic idea is that the relevant item or ability is in the individual because it was selected for the teleofunction in the genotype, and that means we can rightly identify this as the teleofunction of the item. A bit of a story is usually necessary to flesh this out. A heart's teleofunction is to pump blood because the reason that the heart exists is because it pumps blood; if it failed to pump blood the organism would die and not reproduce. And in some sense, it will be right to say that the heart evolved because of selection pressures. But such an explanation is going to be complex and indirect: there is not going to be an organism with a heart competing with a close relative that is the same but for lacking a heart. Instead, over
end p.175
 
time, selection pressures could select for features that ultimately constitute, in descendants in some future population, a heart. Evolution, the idea goes, selects (and here we can read "selects" very weakly) for certain traits; this selection pressure explains why those traits carry on; and this explains the sense of "because" in clause (ii) (although it may be a very indirect explanation!).
A number of problems plague this kind of etiological account. For example, it is unclear how we can account for the fact that capabilities change their teleofunction over time; that is, what historical pressures must be referred to in order to identify the present teleofunction of a structure, if the teleofunction appears to be different now than at some earlier time? This problem has given rise to the recent history theories of teleofunctions (Godfrey-Smith 1994): one might propose that it is not the history of the origin of the item or ability, but rather it is the most recent history of it that matters. But I need not discuss the fine details of the various etiological accounts to show why they are not appropriate to be used to identify the supervenience base of phenomenal experience. They are inappropriate because they aim to explain teleofunctions in a weak sense inappropriate for any subsequent application of the right kind of supervenience claim.
Let me begin with an example. Suppose that our theory of the teleofunction of color recognition was an etiological-historical account that refers to selection of a trait in order to identify the teleofunction; and suppose that phenomenal experience were to supervene on this teleofunction. Roughly, mental state S represents color E if it is a proper function of S to correlate with seen occurrences of kind E. This is the proper teleofunction of S if S were ever selected for its correlating with instances of kind E (again: the story will be more complex, since it may never come down to competition over this particular capability; but the point will hold regardless of how we later bootstrap the issue). Suppose also that experience of color E would arise from the occurrence of this representation (such an account is consistent with Lycan 1996). Thus, my experience of blue is somehow going to be dependent upon the underlying representational machinery having been selected for representing blue (or at least color, or at the very least some capacity of seeing things) in the environment. Now, consider the following scenario. We know that evolution works in a ramshackle way. Suppose that an organism is born and because of a random mutation it can see a particular shade of blue that is bluer than is perceivable by the other members of its species—we'll call it deeper-blue. In this individual, this capability has not yet been selected for or against, and so it has merely appeared via dumb luck. But it seems common sense that there will be a phenomenal experience accompanying the exercise of this capability. It could even happen that the ability is passed on in a subpopulation but never selected for or against. This would happen, for example, if although there may be deeper-blue things in the environment, none of them was ever of significant use or harm. Is it then impossible to have a deeper-blue experience? If that were
end p.176
 
so, how would we ever get any conscious experience? Unlike the twin arguments against teleofunctions (in which we are asked to imagine having an exact twin with a different history, or even no history), this scenario describes something like what actually must have happened: surely a capacity must occur in some individual of a species before it can be selected for or against. Thus, on an evolution-based teleofunctional theory of the supervenience base for phenomenal experience, an organism can have a kind of zombie capability (such as the capability to see deeper blue without having a phenomenal experience of deeper blue) because the capability has as yet no proper teleofunction.
There are several potential answers to this objection. One is that the ability to see deeper blue gets a teleofunction and thus a representational role in a derived way. Although I accept that there can be derived teleofunctions, I believe that derived teleofunction is the weak point of the etiological teleofunctional accounts, since it is unclear when such a claim is legitimate; but I will set this concern aside here. Another answer is that teleofunctions (and the ability to derive some kinds of teleofunction), and therefore consciousness, emerge gradually in a continuum (see Lycan 1987, 44). But neither of these answers is satisfactory, since the real perplexity is not the method of the bootstrapping, but that the bootstrapping happens at all. We have supposed that several generations of a subpopulation had the deeper-blue trait, and it was not selected for. Suppose now that deep-blue-colored predators appear on the scene, and so this subpopulation now has selection pressures in its favor as its deeper-blue—blind rivals are eaten up. The first group of generations having this trait and the latter group would have no internal differences that explain why in the latter but not the former there are the phenomenal experiences in question; rather, it would be a fact about the history and environment of the organism that determined this. Thus, phenomenal experience would spring out of contingent external environmental factors.
This problem applies not just to the claim that phenomenal experiences are representations, but also to any theory that they strongly supervene on teleofunctions where the teleofunctions are similarly specified. The recent history version of the etiological account has the same problem, as does the propensity version (Bigelow and Pargetter 1987), and the relational theory (Walsh 1996). Also, this problem may arise for some of the other versions of teleofunctions. For example, some teleofunctions might be specified in a historical account of learning. Thus Fred Dretske (1981), for example, exploits not evolution but rather the notion of a training period to explain some mental teleofunctions. A representation can get a functional role by having a training period in which certain correct associations are learned; then future associations are right or wrong inasmuch as they live up to the training period's cases. But if we were to use Dretske's approach to explain teleofunctions, we would have something like a bunch of brain reactions happening throughout an organism's life, but which become phenomenal
end p.177
 
conscious only after the end of some arbitrary training period. And similar kinds of problems arise for many explanations of causal co-variance under optimal conditions, depending upon how we explain optimal conditions.
The problem with these accounts lies with the nature of etiological teleofunctional explanation and related notions. The etiological theories primarily offer an account of how it can be legitimate to describe something as having a function. That is, they are not ontological57 claims, but rather analysis of when it is appropriate to claim a thing plays a teleofunctional role. But, if a teleofunction is going to be posited as the supervenience base for (some aspect of) a phenomenal experience, then the theory of teleofunctions is providing not just an analysis or explanation, but also is identifying kinds for another theory. Taking etiological theories in this way leads to a very weird kind of externalism, as I showed above. (But since an etiological account of functional explanation alone does not entail an ontological claim, the fact that in the thought experiment we cannot call the ability to see deeper blue a teleofunction is not going to be a problem; we might just accept that the ability to see deeper blue does not have the kind of function that is etiologically specified until it is selected for or until it can serve to assist survival, since this will be how we define the relevant teleofunctions.) I think that some weird consequences are inevitable in a teleofunctional account of consciousness, but they can be less extravagant. The primary notion of teleofunction that needs to be clarified is that one specified in terms of the actual internal conditions of the organism; that is, the relevant teleofunctions need to be understood as occurrent, individual organic conditions.
Systematic Teleofunctions
There is an alternative approach to explaining some teleofunctions; instead of appealing to an etiology, it is possible to appeal to the organizational properties of the thing with the function. This is the systems-based approach. This approach has been less developed, perhaps because the view is more complex and so the demands of explaining it are great; but important defenses of the view are found in Gerd Sommerhoff (1950), Ernest Nagel (1977), and Gerhard Schlosser (1998). The approach can exploit the insights of the evolutionary account of teleofunctions, but remains local and quite natural in its effects.
The basic idea of the systems-based approach to explaining teleofunctions is that goal-directed behavior is a property of whole systems. We can think of an organism as a complex system that has certain capabilities which we identify as teleofunctions because they maintain certain features of the system. Prima facie, most organisms are adaptive and self-organizing complex systems. The notion of a self-organizing system is not, unfortunately, very easily defined. But for my purposes here, we do not need it to be. What matters instead is that, undeniably, such systems do exist, and that affective organisms are surely examples of them. This is most obvious
end p.178
 
when we consider homeostasis. The process of evolution produces systems which act in their environments in a way that maintains a stable range of values on many parameters of the organism: body temperature, water levels, physical integrity (e.g., some avoid being broken in two), and many others. These homeostatic functions are in part demonstrable in a simple empirical way. Given the opportunity to drink water, many organisms will do so in a way that maintains a range of the quantity of water in their bodies. Many organisms maintain a constant narrow range of body temperature, and many also, when given the opportunity to walk along a temperature gradient, will seek a place that allows them easy thermoregulation. And so on. Of course, all organisms depend upon their environment for stability on some dimensions. Thus, some organisms will eat too much if food is available in excess of what would normally be the case in their environment. Similarly, no organism that I know of can directly detect and avoid x-rays. Organisms also have limited abilities to respond; they can travel only a particular distance, or thermoregulate in a particular range of external temperatures, and so on. But within certain parameters, which should be expected to be close to those in which they evolved, organisms demonstrate homeostasis for much of their lives. There is no doubt an issue about what counts as a proper or normal set of such ranges for an organism, and one might worry that this will start the account back onto a historical and externalist track. But this can be avoided by the observations that the relevant parameters will be maintained only in a particular range which we can empirically identify (a rat heated to 10,000 degrees centigrade is not going to thermoregulate, or maintain its internal water levels, and so on); thus, we can empirically identify the ranges of environmental conditions in which the organism will succeed in various kinds of homeostasis. Conversely, we can know as an experimentally demonstrable fact that an organism will die if it does not stay in these ranges.
The primary difficulty confronting the systems-based approach is that the parameters that we identify as demonstrating the effects of teleofunctions need to be, to some degree, "orthogonal to" (i.e., independent of) each other (E. Nagel 1977, 273). This is because all kinds of rather trivial systems can be said to maintain state along dependent parameters. Ernest Nagel uses the example of a ball set in motion in a spherical bowl (274). Is its gyring trajectory toward rest teleofunctional? His answer is that this is a different kind of process since the laws of physics make all the relevant parameters tightly related. This may then seem a bit odd: we identify as having teleofunctions those systems that have components which operate not as a simple, but rather as a complex, application of laws of nature. As I shall repeat below, however, I don't think that this is a problem. Teleofunctional systems, and hence autonomous systems, come in degrees, and a ball moving in a bowl is a limit case of a system demonstrating negligible self-organization and autonomy. We should not shy away from the possibility that there is no clear division between complex autonomous systems, such as the homeostatic systems in organisms, and simpler physical ones.
end p.179
 
And, for our purposes here, it is sufficient to note that the kinds of organisms that are our concern (those that have affects) are of more than sufficient complexity to meet this criterion.
Recently, Schlosser (1998; see also Christensen 1996) has offered a more general teleofunctional account which is systems based, and which provides a definition of teleofunctions that utilizes many of the insights of Larry Wright (1976) and others working to explore an etiological account. Schlosser argues that a "functional state or trait X is any state or trait of a complex system, which by generating another state or trait F (the function of X) is under certain circumstances necessary for its self-production" (305). The basic idea is that X and F occur in a complex system, and in a sense they are necessary for the generation of one another. Schlosser is concerned that the notion of "complex system" is very vague; in fact, it is playing a role like Ernest Nagel's notion of the "orthogonal" parameters: if X and F produce each other in a tight, trivial loop, then we are disinclined to think of F as a teleofunction. But if the relation between X and F is complex, then the teleofunctional ascription is more compelling. Again, we can set aside this issue with the observations that there is a continuum in nature of various degrees of complex systems, and that the biological systems that demonstrate basic emotions and other affects are undoubtedly complex. Another feature which is equally problematic is the time that such functions endure. Schlosser recognizes that the cycles of production must be more than one loop; and that we will think of a system as a homeostatic and therefore interestingly self-organizing system only if it is homeostatic for some amount of time. But this, too, is a matter-of-degree issue which, though requiring further analysis, poses no refutation of the view.
Thus, Schlosser offers the following account (315):
F cis a function of X c (t) iff: for a certain period of time t 0<t <t + x + y <t 0+ T
1.  
X(t) is directly causally necessary to establish F(t+x) (under certain circumstances c 1 );
2.  
F(t+x) is indirectly causally necessary to establish X(t+x+y) under certain circumstances c 2 );
3.  
the causal relations between X(t), F(t+x), X(t+x+y) are complex.
The time indexes are required to clarify that the causal relation between these states is correctly ordered; the reproduced states are instances of types.
Thus, consider a homeostatic process like the maintenance of internal water levels. In a primate, central nervous system osmoreceptors (kind X) detect cellular dehydration (teleofunction F), which can motivate the organism to drink water (see Rolls 1999, 205ff). We can say that the osmoreceptors have as a teleofunction the detection of cellular dehydration in order to stimulate rehydrating behaviors; these behaviors in turn are indirectly causally necessary for the reproduction of such osmoreceptors, since
end p.180
 
without sufficient hydration the organism would simply not be able to produce and maintain this neural subsystem.
One very significant advantage of this approach is that it is closer to how we typically ascribe biological teleofunctions. If we were to discover a new kind of mammal, for example, we would not be stumped to understand it, waiting until the evolutionary biologists placed it in evolutionary history and then told us why this or that organ or behavior evolved. We could just watch it in its environment, and working with some basic naturalist presuppositions (and we have many of them: all animals must eat something to gain energy sources, energy is necessary for life, high temperatures impede the biochemical processes of life, and so on), we could develop theories of what it was up to. This is how ethologists have always had to work, after all.
The Schlosser account is in part a kind of microetiological one, in which the reproduction of capabilities is the criterion used to identify their teleofunction. One might therefore argue that it is also strange that phenomenal experience would supervene on a system with a history. But the criterion is local in both space and time, and I consider this much less costly an explanation as a result. And the two explanations should actually not diverge very much. The etiological explanation of teleofunctions should be expected to yield a set of functions that is nearly coextensive with the systems-based account of teleofunction. They can diverge primarily when newer functions appear that would not yet have been selected for.
The hypothesis that I propose is that phenomenal experience strongly supervenes upon the systems-based teleofunctions of which we are aware—I will call these a-teleofunctions. My point in picking out these specific teleofunctions is a very simple one: we are not hard-wired to be able to be directly aware of the functioning of our livers, for example, and so although there are teleofunctions that it serves, these are not a-teleofunctions, and we have no phenomenal experience of them. Other teleofunctions are potentially a-teleofunctions but sometimes have effects that are too tiny or fast to generate awareness.
I have placed special stress upon homeostatic processes. I believe that homeostasis yields the primary teleofunctions that constitute phenomenal experience. This is a view endorsed, for example, by Jaak Panksepp, who argues that "sensations generate pleasure or displeasure in direct relation to their influence on the homeostatic equilibrium of the body" (1998, 164). Panksepp gives an analysis of how the phenomenal experience of thirst can become so intense it is overwhelming (165ff), arguing that this is illustrative of how feeling is related to homeostatic drives. One objection to this view is that there are teleofunctions which are not homeostatic on the usual conception of this term; sexual behavior is a typical example, and we do clearly have phenomenal experiences associated with sexual drives and sexual activity. We could offer a compelling unified account of how things like sexual behavior can be indirectly homeostatic, where endogenous factors alter the system and set it on a course of behavior that can be thought of
end p.181
 
as seeking a return to another state. Thus, lust would be like an endogenously produced thirst, and sexual behavior like drinking. But I think that we can pass over what may be a contentious issue, and just observe that homeostasis provides the clearest, and perhaps the most important, cases of the kind of systems-based teleofunctions upon which, when we are aware of the process, phenomenal experience strongly supervenes.
The Puzzles Revisited
I can now show how this approach solves the four puzzles presented in the last chapter.
Puzzle 1: Strong Supervenience and the Role of Phenomenal States
I have hypothesized that phenomenal experience strongly supervenes upon the teleofunctions of various processes of which we are aware. Thus, the phenomenal experience of a conscious instance of fear is inseparable from the bodily changes that we experience as part of this emotion: the motivation of flee, the change in heart rate, the shaking excitement of adrenaline. The phenomenal experience of joy, instead, will be completely different, and necessarily so; to change the phenomenal experience of joy with that of fear, for example, would be to change the a-teleofunctions (and hence all the relevant body states) from joy to fear. There is no room here for inverted emotions. Experiences are not tokens in a representational system that can be replaced by other tokens—they are not, like words, wholly arbitrary—but rather, the experience is inseparable from the a-teleofunctional events.
It is worth noting that this steers a plausible path between two extremes that have plagued theories of phenomenal experience. One extreme was the simple brute assertion that phenomenal experiences just are brain states, without any reference to their teleofunctional role. Another extreme is a functional account so broad that it allows for a variety of multiple realizability arguments. Many philosophers have supposed that psychological properties were multiply realizable, and hence for some of these philosophers this means that phenomenal experience also will be multiply realizable. The intuition that compelled many was that psychological properties are teleofunctional, and presumably a Martian or a robot can have capabilities that fulfill some of the same teleofunctions as are present in humans. Hence, be it made of wholly different biological matter or of silicon chips, these functionally similar things can feel pain. A simplisticidentity theory or some versions of strong supervenience are seen to be incompatible with this, since they would identify or tightly link pain, for example, with something like one kind of neural event not had by Martians or robots.
end p.182