Developing and defending a philosophical position is a bit like weaving an intricate piece of fabric. When things go well each strand of the argument adds strength and support to the others, and gradually interesting patterns begin to emerge. But when things go poorly - when one of the strands breaks - it sometimes happens that the entire fabric begins to unravel. A little gap becomes a big gap and soon there is nothing left at all.
This book is about the unraveling of a philosophical position. In some of the chapters, including this one, I'll tell the tale in the first person, since the position that came unraveled was my position, or at least one that I was seriously tempted to endorse. Though it was not mine alone, of course. Several very distinguished philosophers, including Quine, Rorty and Feyerabend, had advanced versions of the view while I was still wearing philosophical knee pants, and a number of well known philosophers continue to advocate the position with considerable passion. The doctrine in question is sometimes called eliminative materialism, though more often it's just called eliminativism. And whatever one thinks of the merits of the view, there can be little doubt that its central thesis is provocative and flamboyant. In its strongest form, what eliminativism claims is that beliefs, desires and many of the other mental states that we allude to in predicting, explaining and describing each other do not exist. Like witches, phlogiston, and caloric fluid, or perhaps like the gods of ancient religions, these mental states are the fictional posits of a badly mistaken theory._
Though a wide variety of arguments have been offered for this rather startling conclusion, all of them share much the same structure. They begin with the Premise_ that beliefs, desires and various other mental states, whose existence the argument will challenge, can be viewed as "posits" of a widely shared commonsense psychological theory -- "folk psychology" as it is often called. Folk psychology, the Premise maintains, underlies our everyday discourse about mental states and processes, and terms like "belief" and "desire" can be viewed as theoretical terms in this folk theory. The Second Premise is that folk psychology is a seriously mistaken theory because some of the central claims it makes about the states and processes that give rise to behavior, or some of the crucial presuppositions of those claims, are false or incoherent. This step in the argument has been defended in many different ways, with different writers focusing on different putative defects. After defending these two Premises, an eliminativist's argument can take one of two routes. The simplest route goes directly from the Premises to the conclusion that beliefs, desires and other posits of folk psychology do not exist. And, of course, if that's right, it follows that no mature science which succeeds in explaining human behavior will invoke the posits of folk psychology. Beliefs, desires and the rest will not be part of the ontology of the science that ultimately gives us a correct account of the workings of the human mind/brain. The second route that an eliminativist's argument can follow reverses the order of these two conclusions. From the Premises it initially concludes that folk psychological posits will not be part of the ontology of any mature science. This, in turn, is taken to support the stronger conclusion that these folk psychological states do not exist.
The Premises of the argument that I've just sketched can be unpacked in many different ways, just about all of which generate controversy. In subsequent chapters I'll take a careful look at several of those controversies. But in this chapter, I propose to put these disputes to one side. For even when the Premises are unpacked in a way that is most favorable to the eliminativists' arguments, and even if we assume, for argument's sake, that these Premises are true, neither of the two conclusions that eliminativists wish to draw follows directly. Some additional premises are necessary. And it is my contention that none of the premises that will do the trick are defensible. If that's right, then obviously eliminativists are in trouble. For even if we grant that their Premises about folk psychology are correct, their ontological conclusions simply do not follow. To support this claim, I'll begin, in Section 2, by elaborating on what I take to be best version of the First Premise, from the eliminativists' point of view, and then assembling, in Section 3, a catalog of the complaints that eliminativists have leveled against folk psychology. The remainder of the chapter will be devoted to setting out my argument that there is no way of getting from the eliminativists' Premises to their conclusions, and exploring the options that are available if my argument is correct.
It is a bit odd that, despite its fundamental importance in eliminativists' arguments, the step linking the Premises to the conclusions has not been the focus of much attention in the literature. In my own writing, at least until recently, it was a step I took quite unselfconsciously. Along with most of the other participants on both sides of the debate, I assumed that the battle would be lost or won by deciding who was right about the virtues and shortcomings of folk psychology. Once it becomes clear how much of folk psychology is denied or abandoned in the mature sciences of the mind/brain, it would be obvious what to say about the extent to which the ontology of folk psychology and of the successful sciences overlap. But gradually over the last several years I have come to realize that this crucial step in eliminativists' arguments is anything but obvious.
My first serious inkling that perhaps all was not well came while I was polishing a paper that I had written with Bill Ramsey and Joey Garon in which we set out one particularly trendy argument for eliminativism. That argument begins with some speculations about the future success of connectionist models of human memory, and notes that the interactions among the states posited by those models are quite different from the interactions among beliefs, as they are construed by commonsense psychology. The argument goes on to conclude that if those connectionist speculations prove to be correct, then the ontology of scientific psychology will not include beliefs. The paper that Ramsey, Garon and I wrote is reprinted as the second chapter in this volume. Though nothing much in this first chapter turns on the details, you might want to give it a quick read before going on, if you haven't done so already. It will give you a feel for what the eliminativist fabric looks like before it begins to unravel.
Just as we were finishing that paper, I had occasion to re-read a characteristically acute essay by William Lycan in which he notes that the conclusion in arguments like ours doesn't follow unless some additional premise is added, and goes on to suggest that the additional premise which is (often tacitly) assumed by most eliminativists is some version of the description theory of reference for theoretical terms. I suspect that Lycan is quite right about what others authors had been assuming, and he is certainly right about me. In Section 4, I'll explain in some detail where the additional premise comes from and how it works. Lycan has never been much tempted by eliminativism, and in the essay that woke me from my dogmatic slumbers, he explains why. Description theories of reference have come in for a great deal of criticism in recent years, and he favors a very different account of reference. Moreover, if that account is correct, then premises detailing untenable features of folk psychology, conjoined with suitable premises about the reference of theoretical terms, will not support the sort of eliminativist conclusions that Ramsey, Garon and I were proposing. Section 5 is devoted to setting out Lycan's argument and exploring some of its implications. In a footnote to our paper, Ramsey, Garon and I offered a hasty rebuttal designed to show that the theory of reference Lycan favors is just as problematic as the description theoretic account that he rejects._ But since we had no better alternative to offer, we hurried on with our own argument, granting that the decision on whether our premises sustained our conclusion would have to be something of a "judgement call."
I wasn't all that happy with this "quick fix," and I resolved that at some point I would try to work out a better theory of reference - one that was more likely to be correct than either Lycan's or the description-theoretic one on which I had been relying. Before I could start on that project, however, there was a prior question to be confronted. If the goal was to produce a correct theory of reference, I would have to get clear on what it is that makes a theory of reference correct or incorrect. What exactly are the facts that a correct theory of reference is supposed to capture? And how can we find out whether a theory has succeeded in capturing those facts? These are the questions I'll take up in Section 6. The discussion there follows a line of thought that I developed in a series of papers which have appeared in the last few years._ But that line kept heading off in a very surprising direction. There are, I think, two quite different stories to be told about what a theory of reference is up to. On one of them, which I'll call the "proto-science" account, the theory of reference is attempting to characterize a word-world mapping that will be useful in one or another empirical discipline, like linguistics, or cognitive psychology, or perhaps the history of science. According to the other story, which I'll call the "folk semantics" account, the theory of reference is attempting to capture the details of a commonsense theory about the link between words and the world. This latter story appears to be favored, albeit tacitly, by most philosophers. However, as I'll argue in Section 7, if this is the view we adopt, then there probably is no correct account of reference for the theoretical terms invoked in a seriously mistaken theory. Reference, in these cases, is simply indeterminate. Moreover, whether or not I am right about the indeterminacy of reference in these cases, the folk semantics story suggests that reference is a quirky and idiosyncratic relation, and that there are lots of alternative relations that we might have used in its stead. There is nothing special about reference that distinguishes it from these alternatives. It just happens to be the member of this cluster that our culture has latched on to. If this is right, then the debate over eliminativism begins to look very odd indeed. For, as argued in Section 7, if reference is not a particularly interesting or important relation, and if the existence or non-existence of the posits of folk psychology turns on whether or not the theoretical terms of folk psychology refer, then it seems to follow that the eliminativists' conclusions themselves can't be all that interesting or important. Even if it turns out that the theoretical terms of folk psychology don't refer, just the opposite conclusion might have followed had we inherited a somewhat different notion of reference.
This was a rather radical conclusion to reach. But I had never let that bother me before. Indeed, I confess that I have a certain fondness for such conclusions. And anyhow, that was where the argument seemed to be leading, provided that we opt for the folk semantics account of the facts that a theory of reference is supposed to capture. Suppose we opt for the other account, the proto-science story? In that case, as noted in Section 8, there is no saying what follows from the eliminativists' Premises, since the relevant sciences have not yet determined which word-world relation will be of use to them. And even if we ignore this problem, the proto-science story leads to some pretty bizarre consequences of its own.
So it looked like I was stuck with the conclusion that an issue I had spent much of the last two decades thinking about was either unresolvable or not very interesting, and that I'd just have to learn to live with that conclusion. As part of the process, I did what many philosophers do these days: I took my show on the road, trying to defend my new view in front of a variety of audiences. Few were persuaded. But it took John Searle (with some help from Frank Jackson and Christopher Gauker) to convince me that there was something very wrong with the argument I was offering. What persuaded me was Searle's insistence that my argument, if sound, was perfectly general. It applied equally to the posits of folk psychology and to the posits of physics. So if my argument were correct, debates about the existence of black holes or the Big Bang should also be unimportant and uninteresting. And that is a crazy conclusion, even for someone with my high tolerance for views that fly in the face of conventional wisdom. Now I was in a real pickle. For while Searle and others had persuaded me that there must be something wrong with my argument, none of my critics had a plausible account of what was wrong with it. Where, exactly, had the argument gone wrong? I am, I admit, still not completely confident that I know the answer to this question. But in Section 9 I'll set out the best analysis I've come up with so far. It's rather a long story, and I won't ruin the suspense by trying to summarize it here. The bottom line, if I'm right, is that the mistake came right at the beginning when we turned to the theory of reference to try to settle whether the eliminativists' Premises supported their conclusions. On the view I'll set out in Section 9, the appeal to reference and the strategy of "semantic ascent" are complete non-starters when it comes to settling ontological questions like those that eliminativists raise.
But if we can't appeal to the theory of reference, what can we do? How do we settle questions about the existence of things spoken of in theories that we no longer take to be correct? One idea, considered briefly in Section 10, is that the notion of a "constitutive" or "conceptually necessary" property will help resolve the issue. That proposal, I'll argue, raises more problems than it solves.
Having reached this point it seemed prudent to look for some other way of determining what we should conclude from the eliminativists' Premises, a way which didn't rely on semantic notions or appeals to conceptual necessity. One idea that seems promising is to look to the history of science in the hope of finding principles of ontological inference that have been used in other cases. If we can locate some candidate principles, perhaps these can then be confirmed by looking at other historical cases, and by testing the principles against our intuitions in hypothetical cases. This would, I think, be an intriguing project. But as I argue in Section 11, there is no guarantee that it would succeed. For it might well be the case, indeed I think it is the case, that there are no principles in this area that are strong enough to specify what ontological conclusions we should draw when confronted with a seriously mistaken theory. Rather, I maintain, these issues are typically settled through a process of social negotiation in which politics, personalities and social factors can all play a role. I'm told that makes me a social constructionist, or at least a fellow traveler. But, as I'll argue in Section 12, the position I'm advocating can also be viewed as a close neighbor to the versions of pragmatism favored by Quine, Rorty and others. And ever since I started out in philosophy, I've thought that's the best neighborhood in town.
That brings me to the end of my preview of the current Chapter. The rest of the book consists of essays, some of which have been previously published, that were written while I was struggling with the ideas set out in this Chapter. Once the fabric of the eliminativist argument started to unravel, new holes seemed to pop up everywhere. On a closer look, some of the arguments aimed at showing that folk psychology was not a very promising theory -- arguments that I had once thought quite plausible -- now seemed much less plausible. My current view on these arguments is set out in Chapters 5 & 6. After a while even the first step of the eliminativists' argument, the one that claims there is a folk psychological theory that might turn out to be badly mistaken, began to look much less obvious than it once had. Chapter 3 explores some of the reasons why many philosophers and cognitive psychologists have accepted this assumption, and sets out some of the ways in which it might turn out to be untenable. Chapter 4 focuses on simulation theory, which is the basis of the most recent attack on the First Premise of the eliminativist argument.
Since most of these essays have more than one author, perhaps this is an appropriate place for few words about my collaborators. Throughout my professional career I have been exceptionally fortunate in having the opportunity to interact with many gifted, creative and enthusiastic students. They have always been my best critics and my best inspiration to explore new ideas and to say things more clearly. Many of them have gotten involved in my intellectual projects, or gotten me involved in theirs, and these interactions, more than anything else, are what makes academic life rewarding for me. All of my collaborators in this volume are my former students. And I owe them all a considerable debt. Without them the book would have been much less interesting to read and much less fun to write. At one time I planned to combine the material in this Chapter with Chapter 3, Chapter 5 and parts of Chapter 4, and publish it all together as a single book length study. But in each case my students and I decided it would be best to publish the collaborative work separately. One result of that decision is that there is a bit of overlap in these chapters. Ideas, arguments, and even a few sentences appear more than once.
I've made lots of promises about what I'm going to do in the pages that follow. Now it's time to get to work. But since all work and no play makes for dreary going, let me end this Section with a few mischievous observations on how my project in this Chapter might be construed. For some years now Deconstructionism has been a pretentious and obfuscatory blight on the intellectual landscape. But buried in the heaps of badly written blather produced by people who call themselves "Deconstructionists," there is at least one idea - not original with them - that is worth noting. This is the thesis that in many domains both intellectual activity and every day practice presuppose a significant body of largely tacit theory. Since the tacit theories are typically all but invisible, it is easy to proceed without examining them critically. Yet once these tacit theories are subject to scrutiny, they are often seen to be very tenuous indeed. There is nothing obvious or inevitable about them. And when the weakness of the underlying theories has been exposed, the doctrines and practices that rely on them can be seen to be equally tenuous. If, as I would suggest, this process of uncovering and criticizing tacit assumptions is at the core of Deconstructionism, then eliminativism is pursuing a paradigmatically Deconstructionist program. However, if I am right, the eliminativist deconstruction of commonsense psychological discourse has itself tacitly assumed a dubious package of presuppositions about the ways language and ontology are related. So if the goal of eliminativism is to provide a deconstruction of the mind, one goal of this chapter is to deconstruct that deconstruction.
A central thesis of this Chapter is that even if we grant the Premises in the eliminativists' arguments, there is no plausible way of getting from these Premises to the ontological conclusions that eliminativists want to establish. In this Section I'll set out one version of the eliminativists' First Premise, a version designed to make the job of getting from Premises to conclusions as easy as possible. There are lots of other ways in which this Premise might be unpacked, and in Chapter 3 Ian Ravenscroft and I have tried to explore them in a systematic way. But in this Chapter I propose to ignore these alternatives. There are also lots of reasons to suspect that the version of the Premise I'll set out here might turn out to be false. Some of these reasons will be considered in Chapters 3 and 4. Here, however, I will ignore them. Since I want to focus on the link between Premises and conclusions, I'll just explain what this version of the First Premise claims, and then assume that it is true.
It is an easy job to state the version of the First Premise that I think will give the eliminativists their best shot at drawing the conclusions they want to draw. Explaining it will take a bit more work. What the Premise claims is that our folk psychological capacities are subserved by a theory (that I'll call "folk psychology") which:
i) is largely tacit
ii) is encoded in a declarative linguistic format
iii) asserts (or presupposes) that beliefs, desires
and other intentional states that it invokes have representational (or semantic)
properties, and that these properties play an essential role in individuating
beliefs and desires
(iv) attributes an opulent array of causal powers to beliefs, desires, and other intentional states, some of which are dependent on the representational properties of those intentional states.
Obviously a fair amount of unpacking is in order. Let's start the with the notion of "folk psychological capacities". This is the term that Ravenscroft and I introduced to refer to a cluster of abilities, including: the ability to make predictions (which often turn out to be correct) about what people will do; the ability to attribute beliefs, desires and other intentional states to people in a way that other observers often agree with; and the ability to construct explanations of people's behavior that are couched in intentional terms and that other people often agree with. There are a number of other abilities that might be added to this list,_ but for current purposes this should suffice. One of the few claims that isn't controversial in this area is that normal adults in our culture do indeed have all three of these abilities.
To explain what I mean when I say that our folk psychological capacities are "subserved" by a theory, it will be useful to consider an analogy with another capacity, which might be called our "folk physics capacity". This capacity, too, consists of a cluster of abilities, including: the ability to make predictions (which often turn out to be correct) about the movements of middle sized physical objects (rocks that are dropped, or thrown, or rolled down hill, boxes that are pushed or pulled, swinging pendulums, etc.), and the ability to offer explanations (that other people often agree with) of why the objects behave as they do. How do people go about making these predictions and constructing these explanations? One very plausible answer is that people are relying on a commonsense theory (sometimes called "naive physics" or "folk physics") that includes principles specifying how objects will move under a variety of circumstances, along with other sorts of information that might be useful. The theory might well include terms for forces or for aspects of situations that are not readily observable and whose existence or magnitude must be inferred. Of course, the hypothesis that people rely on a theory to come up with physical predictions and explanations does not constitute a complete explanation of their folk physics capacity, even if the folk theory is specified in detail. We also need some account of how they use the theory -- how they apply it in various situations.
In recent years, cognitive scientists have offered a fair amount of evidence for the hypothesis that people's folk physics capacity is subserved by a commonsense theory. One of the most fascinating findings in this area is that many people seem to base their physical predictions and explanations on a physical theory that is mistaken, and that posits an unobservable internal force which, according to Newtonian (and post-Newtonian) accounts of the world, simply does not exist. McCloskey offers the following summary of this "naive theory of motion":
[The basic theory] makes two fundamental assertions about motion. First, the theory asserts that the act of setting an object in motion imparts to the object an internal force or "impetus" that serves to maintain the motion. Second, the theory assumes that a moving object's impetus gradually dissipates (either spontaneously or as a result of external influences), and as a consequence the object gradually slows down and comes to a stop. For example, according to the ... theory, a person who gives a push to a toy car to set it rolling across the floor imparts an impetus to the car, and it is this impetus that keeps the car moving after it is no longer in contact with the person's hand. However, the impetus is gradually expended, and as a result the toy car slows down and eventually stops._
This basic theory can be elaborated in a variety of ways. One particularly interesting elaboration deals with curvilinear motion.
Many subjects believe that an object constrained to move in a curved path acquires a curvilinear impetus that causes it to follow a curved trajectory for some time after the constraints on its motion are removed._
Evidence that people rely on this theory comes from a variety of experiments. In one set of experiments, subjects (all of whom were undergraduates at a highly selective American University) were presented with problems like the following:
Imagine that someone has a metal ball attached to a string and is twirling it at high speed in a circle above his head. In the diagram [Figure 1a] you are looking down on the ball. The circle shows the path followed by the ball and the arrows show the direction in which it is moving. The line from the center of the circle to the ball is the string. Assume that when the ball is at the point shown in the diagram, the string breaks where it is attached to the ball. Draw the path the ball will follow after the string breaks._
Thirty per cent of the subjects responded with drawings like Figure 1b, indicating that they believed the ball would continue in curvilinear motion after the string broke. Moreover, "most of the subjects who drew curved paths apparently believed that the ball's trajectory would straighten out."_ In another experiment, subjects presented with this and similar problems were interviewed at length about their answers. One subject offered the following explanation for the curved path that he predicted the ball would follow:
"You've got a force going around and [after the string breaks, the ball] will follow the curve that you've set it in until the ball runs out of the force within it that you've created by swinging."_
Figure 1 About Here
In another problem, subjects were shown Figure
2a, and told that it represents a side view of a metal ball swinging back
and forth at the end of a string. They are asked to draw the path the ball
will follow if the string is cut when the ball is in the position shown.
"Several subjects indicated that ... the ball would continue along
the original arc of the pendulum for a short time, and then would either
fall straight down [as in Figure 2b] or would describe a more or less parabolic
trajectory [as in Figure 2c].... One subject who made this sort of response
explained that when the string is cut, the ball has
"the momentum that it has achieved from swinging through this arc and should continue in a circular path for a little while.... then it no longer has the force holding it in the circular path, and it has the force of gravity downward upon it so it's going to start falling in that sort of arc motion because otherwise it would be going straight down."_
In a separate set of studies, Clement found that 88% of a group of entering freshmen engineering students made similar appeals to an impetus-like internal force in answering questions about the motion of a coin tossed straight up. After taking a freshman level course in mechanics, however, only (!) 72% of subjects gave responses that indicated a belief in impetus._
Figure 2 About Here
As McCloskey and Clement point out, these subjects have a lot of distinguished company. The claim that the act of setting an object in motion impresses an internal force in the object that serves to keep the object in motion played a prominent role in physical theories from the 14th century until the time of Newton. It was clearly endorsed by Galileo in his early writings, and it was invoked by Leonardo da Vinci who offered the following description of the motion of an object under conditions similar to those in Figure 1.
Everything movable thrown with fury through the air continues the motion of its mover; if, therefore, the latter move it in a circle and release it in the course of this motion, its movement will be curved._
These findings certainly make a plausible case for the hypothesis that people's folk physics capacity relies on a commonsense theory. And for eliminativists they offer an added attraction, since modern physics has shown that the theory being exploited by many subjects is simply mistaken. Moreover, one of its mistakes has a distinctly ontological flavor. The impetus theory posits the existence of an internal force in most moving objects, a force which obeys a fairly complex set of laws and which explains why the objects move as they do. But, modern physics assures us, that force does not exist. Sensible eliminativists will acknowledge that the facts about folk physics don't by themselves allow us to draw any strong conclusions about the posits of folk psychology. But I think the work on folk physics does make it plausible that the eliminativists' conclusions might be true. For it shows that it is possible that in their everyday dealings with the world -- dealings which are by and large pretty successful -- people rely on commonsense theories which appeal to forces that simply do not exist.
The next bit of jargon that needs explaining in my version of the eliminativists' First Premise is the claim that the theory subserving our folk psychological capacities is "largely tacit." Here another analogy will be helpful, this time an analogy with our linguistic capacities. Native speakers of a language can understand and produce an indefinitely large set of sentences of the language and make a wide array of judgements about the grammatical properties of those sentences. Also, there is an impressive degree of intersubjective agreement on those judgements. According to Chomsky and his many followers, the best explanation for these capacities includes the hypothesis that people have internally represented a generative grammar of their native language, and that the internalized grammar is exploited in various ways in producing, processing and judging sentences._ But, of course, this internalized grammar is stored in a way that makes it largely (perhaps completely) inaccessible to conscious access. People can't simply introspect and tell us the rules of the grammar they have internalized. If they could, the science of linguistics, which tries to specify the grammars that speakers have internalized, would be a lot easier than it is. Rather, Chomsky maintains, the grammars that people use are "tacitly known". The theory subserving our folk psychological capacities may be a bit more accessible to introspection than the grammar of our language. But there is good reason to think that much of the information (or mis-information) that we use in predicting and explaining people's behavior is stored in a way that makes it inaccessible to conscious access._ And that is what my version of the Premise claims.
Much of the debate in cognitive science over the last three decades has turned on the format in which information of various sorts is represented in the mind. Early on, it was widely assumed that most of the information stored in the mind is represented in linguistic or quasi-linguistic form. Some theorists argued that the natural language a person spoke (or something close to it) would be a suitable medium for storing most of what the person knows, while others maintained that natural languages would not do, and that one or more species wide "languages of thought" had to be posited._ But non-language like competitors were soon suggested, including quasi-pictorial representation, holographic representation, various sorts of mental models and, most recently, various sorts of connectionist representations._ Since eliminativists want to argue that folk psychology is false, it had better be the case that folk psychology is represented in a way that admits of such assessments. And, while various sorts of representation might arguably fit the bill, linguistic representations are the least problematic. So on my version of the First Premise, it's claimed that folk psychology is stored in a linguistic or quasi-linguistic format. Not just any linguistic representation will do, however. For there are lots of linguistic constructions -- imperatives, for example, and questions -- which can't be straightforwardly evaluated as true or false. Thus my version of the tacit theory Premise assumes that significant parts of our tacit folk psychological theory is stored declaratively.
According to both eliminativists and their staunchest critics, folk psychology takes beliefs, desires and other intentional states to be representational states. My belief that Reno is further west than Los Angeles represents the world as being a certain way, and so my belief is true if and only if the world is that way. That state of affairs is the truth condition of my belief. Desires represent the world as the person with the desire would like it to be. My desire to have sushi for dinner is fulfilled if and only if I do have sushi for dinner. That state of affairs is the fulfillment condition of my desire. Jargon abounds in this area. Truth and fulfillment conditions are sometimes collectively referred to as "conditions of satisfaction." Sometimes they are called the "content" of the beliefs and desires that have them. Having a satisfaction condition is sometimes called a "semantic property," or an "intentional property."
Many philosophers contend that semantic properties play an essential role in folk psychology's scheme for individuating instances (or "tokens") of propositional attitudes and classifying them into types. If we ask when a belief that a person has at a given time is identical with a belief she has at some later time, folk psychology's answer, according to these philosophers, is that the beliefs are identical if and only if they have the same content. Similarly, if we ask when two different people have the same belief (or "believe the same thing") folk psychology's answer, according to these philosophers, is that they have the same belief if and only if they have beliefs (or "belief tokens") with the same content. Though this account of how folk psychology individuates beliefs is not without it's critics, the version of the First Premise that I'm developing takes the account to be correct.
This is far from a complete account of the folk psychological scheme for individuating beliefs, however. To tell a more detailed story, we would have to specify the circumstances under which folk psychology counts two belief tokens as having the same content. And on this point controversy abounds. Some of the arguments surveyed in Section 3 assume that folk psychology relies on one set of principles for content identity, while other arguments assume that folk psychology relies on quite different principles. In my account of the First Premise, I propose to leave the matter unsettled. So the Premise, as I construe it, is compatible with any reasonable account of how contents are to be individuated.
Much of our ordinary folk psychological discourse can be construed as making causal claims in which intentional states figure prominently. We often say things like: "His mother's tone of voice led the child to believe that she was angry at him." and "The child looked under the chair because he believed his kitten was hiding there." Though terms like "led" and "because" might be interpreted in a variety of ways, it has become commonplace in philosophy to view them as making causal claims. The tone of voice was the cause (or at least a cause) of the belief, and the belief caused the child to look under the chair. The causal interpretation of such commonsense psychological claims was not always widely accepted by philosophers. I suspect that the two papers that did most to convince philosophers that this was the right interpretation were Brandt & Kim (1963) and Davidson (1963) -- they were certainly the two that convinced me. There have always been dissenters, however, and in recent years their ranks have grown as some philosophers have tried to fend off the challenge of eliminativism by denying that commonsense psychology attributes causal properties to beliefs and desires._ This is an important debate, and the matter is far from settled. But since my current aim is to set out the version of the eliminativists' First Premise that is most likely to support their conclusions, I will simply assume that the dissenters are wrong, and that ordinary folk psychological discourse does indeed make robust causal claims. This assumption does not, by itself, tell us anything about the tacit theory that subserves our commonsense discourse. But one very natural way to explain the fact that we often make causal claims involving intentional states is to suppose that the underlying tacit theory includes nomological generalizations specifying the causes and effects of intentional states, and that many of these generalizations are couched in terms of the content of intentional states. On my version, this is what the First Premise in the eliminativists' argument claims.
Horgan and Graham (1990) suggest some vivid terminology that can be pressed into service to characterize competing accounts of folk psychology. Accounts that portray folk psychology as making lots of substantive claims about the nature of intentional states, including lots of nomological generalizations, are opulent accounts, while accounts portraying folk psychology as making relatively few such claims are austere. Austere accounts have some appeal to those who would challenge the Second Premise in the eliminativists' argument, since the fewer claims a theory makes, the less likely it is to be wrong. But the First Premise of the argument, as I propose to construe it, opts for an account of folk psychology that is at the opulent end of the spectrum. Folk psychology, the First Premise insists, includes many putative laws and makes many claims about intentional states.
In the previous Section my goal was to explain the version of the First Premise that I think will give eliminativists their best shot at building an argument from their Premises to their conclusions. In this Section, my goal is to do the same for the Second Premise. My approach, however, will be quite different. It takes no great effort to state the best version of the Second Premise, and since it is so straightforward, there is no need to explain what it means. What it claims is that folk psychology is a seriously defective theory because many of the claims that folk psychology makes or presupposes are false. What does need some explaining is why eliminativists endorse this claim, and why many opponents of eliminativism take it very seriously and feel they have to refute it. To explain that, I will provide a brief tour of what I take to be the most influential arguments aimed at showing that folk psychology is bad psychology. In this Chapter I don't propose to dwell on the details, or to say much about the counter-arguments that have been urged by the friends of folk psychology, though in later Chapters I'll return to a few of these arguments and discuss them at greater length. This Section is intended mostly for readers who are new to the debate, and for those who would like a reminder of how some of the arguments go. The rest of you can scoot ahead to Section 4.
For heuristic purposes, the arguments I'll sketch can be divided into two categories. Those in the first category focus on what might be called the structural and nomological commitments of folk psychology - the claims it (allegedly) makes about the structure of the psychological states underlying behavior, and about how these states causally interact with one another and with other sorts of states. Arguments in this category typically try to marshall evidence indicating that structures or processes of the sort folk psychology appeals to are not likely to be found in the systems that actually produce behavior. Since the available evidence is often quite fragmentary and inconclusive, these arguments commonly indulge in more than a bit of futurology or science fiction. Arguments in the second category focus on the fact that folk psychology attributes semantic properties to many of the states it invokes, and that it exploits these semantic properties in some of its generalizations. These arguments try to show that, for one reason or another, semantic properties are ill suited to the role folk psychology would have them play. Though there is some appeal to science fiction in these arguments too, much of the work is done by metaphysical or methodological principles. There is, I should stress, no sharp divide between these two categories of arguments, and nothing much turns on how an argument is classified. I find that dividing things up in this way is a useful strategy for surveying the literature; I'm sure there are other taxonomies that would do equally well.
Except where explicitly noted, all the arguments I'll sketch work best if folk psychology is assumed to have all the characteristics specified in the previous Section. But in many cases these are not enough. To get the arguments going, it must be assumed that folk psychology has properties in addition to those already specified. I'll note these additional assumptions as we go along. Not surprisingly, all of them are controversial. And one standard strategy used by opponents of eliminativism is to deny that folk psychology has the additional property that the argument at hand needs. Though I'll note some of these controversies, I don't plan to take sides, since it's my view that even if the additional claims about folk psychology are granted, there is still no plausible way of getting to the ontological conclusions that the eliminativists want.
In his seminal paper, "Empiricism and the Philosophy of Mind," Sellars offers a mythical reconstruction of the birth of folk psychological theory._ In the myth, a great genius, Jones, puts forward a theory to explain the fact that people behave intelligently even "when no detectable verbal output is present."_ The explanation Jones offers posits internal events that he calls thoughts. And his model for these internal episodes is "overt verbal behavior itself." Thus, for Jones, thoughts are a kind of "inner speech," and the posited processes that produce intelligent behavior are "inner discourse." Since thinking is modeled on overt discourse, it inherits many of the semantical properties that apply to stretches of public language. Individual thoughts have meaning, and sequences of thoughts may be logically related in various ways. Indeed, when things go well, a sequence of thoughts can have the structure of a sound deductive or inductive argument, and these covert logically sound arguments play a central role in Jones' explanation of intelligent behavior._
It is an impressive testimonial to the importance of Sellars' work that a number of leading figures on both sides of the eliminativism debate take this account very seriously. These philosophers agree that folk psychology characterizes thinking and believing on the model of inner speech, and that it seeks to explain intelligent behavior by appealing to the logical cogency of covert quasi-linguistic episodes. They disagree totally, however, on the plausibility of this folk psychological theory.
Those who think that Jones and the folk who followed him have got it terribly wrong make much of the fact that non-linguistic creatures, like monkeys, cats and dogs, can behave in strikingly intelligent ways. They also stress that infants achieve some remarkably intelligent feats long before they have begun to talk. If Jones' hypothesis about the covert processes subserving intelligent behavior is plausible for human adults then, these eliminativists urge, it would be equally plausible for infants and animals. Yet surely, they insist, this is simply too absurd to take seriously. It is preposterous to suppose that science will discover a language in which the family dog thinks, and it's even more preposterous to suppose that our future science will explain Fido's clever behavior by appeal to his covert construction of deductive and inductive arguments._
This sort of reductio ad absurdum only works, of course, if it is conceded that the putative implications of folk psychological theory really are absurd. And it is certainly not the case that everyone is prepared to make that concession. Indeed, Jerry Fodor, who has long been one of the staunchest defenders of folk psychology, has advanced some very sophisticated arguments aimed at showing that some of the best of contemporary cognitive psychology presupposes a "language of thought"._ Moreover, since most of the psychological theories on which Fodor bases his arguments are equally applicable to people and their pets, Fodor concludes that some of the best animal psychology also presupposes a language of thought. As for children, here again Fodor and some of his followers are more than prepared to bite the bullet. Children certainly do lots of intelligent things before they learn to talk. Indeed, learning to talk is itself one of their more impressive accomplishments. But, Fodor insists, since the best explanations we have for how children succeed in learning to talk presuppose that they already have a language of thought, it must be the case that the language of thought is innate.
A second cluster of eliminativist arguments that rely heavily on Sellars' portrait of folk psychology looks to neuroscience to impugn the folk. If Jones were right, then there would have to be lots of sentence-like states bouncing around in the brain when people do intelligent things. But contemporary neuroscience seems to have little use for states modeled on sentences. And, gazing into their crystal balls, the advocates of this argument do not foresee the neuroscience of the future having any more commerce with brain writing. Patricia Churchland, for example, concedes that "there is some sentence crunching, almost certainly."_ But she goes on to endorse Hooker's prophesy that "language will surely be seen as a surface abstraction of much richer, more generalized processes in the cortex, a convenient condensation fed to the tongue and hand for social purposes."_ In Tim van Gelder's projection of the course of neuroscience there is even less place for "sentence crunching." He sees an "intimate association" between various non-symbolic "distributed" representational systems and "the actual machinery underlying human cognition." This intimate association, van Gelder maintains,
stands in plain contrast with the biological remoteness of symbolic representations. Though CTM [the Classical Theory of Mind] demands a language of thought, and CTM advocates insist that the expressions of this language are realized in the neural substrate, and consequently predict the eventual discovery of `symbols amongst the neurons,' neuroscience has never yet stumbled across syntactically structured representation in the brain. This discrepancy only becomes more embarrassing to CTM as the sum of neuroscientific knowledge increases, and provides at least a prima facie argument in favor of any biologically motivated alternative._
So if it is granted that intelligent behavior must ultimately be explained by what goes on in the brain, then if these futurologists are right, Jones and the folk were wrong, and so too is folk psychology. Of course, those who distrust the neurophilosophers' crystal ball are not much impressed by the argument.
Auntie says that it is crude and preposterous and unbiological to suppose that people have sentences in their heads. Auntie always talks like that when she hasn't got any arguments._
Before moving on, we should pause to note that eliminativists who endorse the argument just sketched can't also endorse the version of the First Premise that I set out in Section 2. For that version of the Premise stipulates that folk psychology itself is stored in quasi-linguistic form. So if neuroscience does indeed establish that there are no sentences in the head, the First Premise will be undermined. I don't think this is a devastating objection to eliminativists who advocate the Argument from Neuroscience, since it is open to them to adopt some other account of how folk psychology is stored in the brain. That's a delicate business, however, and not just any account will do. It has to be an account on which it makes sense to say that folk psychology makes claims or asserts propositions. For if folk psychology does not make claims, then it can't make false claims, and that would undermine the Second Premise. But, as we'll see in Chapter 3, there are several options available that might be adopted by eliminativists who are convinced that neuroscience will establish that there are no "symbols amongst the neurons."
It is conceded on all sides that natural languages are highly structured systems, and that well formed sentences in natural languages must comport with complex syntactic and semantic principles. So if Sellars is right - if folk psychology really does take natural language as the model on which to base its conception of thoughts and other propositional attitudes - then the commonsense conception of propositional attitudes will view them as highly structured as well. But many defenders of folk psychology would deny that the commonsense conception of propositional attitudes requires them to have syntactic structure._ And, on the other side of the fence, some of the structural arguments aimed at showing that folk psychology is mistaken do not assume that folk psychology claims propositional attitudes must have internal quasi-linguistic structure.
The argument set out in Chapter 2 that Ramsey, Garon and I developed is an example. That argument says nothing about the internal structure of beliefs. Indeed, as far as that argument is concerned, folk psychology might perfectly well claim that beliefs have no internal structure. The only special assumption about beliefs the argument requires is that folk psychology views beliefs as "propositionally modular" -- they are semantically interpretable states that can be causally implicated in some cognitive episodes and causally inert in others. If this is right, then not all of our beliefs need be causally implicated in each inference we make. Having tried to make this assumption plausible, the argument then indulges in a bit of science fiction. There is a family of connectionist models capable of storing a set of propositions in a widely distributed way. But these models have no functional parts that can be identified with the storage of individual propositions. Rather, as one critic of our paper put it, "there is a real sense in which all the information encoded in the network's connectivity matrix is causally implicated in any processing in which the network engages."_ Thus when one of these radically holistic models does its thing, it makes no sense to ask which of its encoded propositions were causally active and which were inert. Now if it turns out that models like this provide the best psychological account of human belief or propositional memory, then folk psychology, which rejects this radical causal holism, will have made a pretty serious mistake._
Another argument in which connectionist models play a central role is due to Martin Davies._ According to Davies, folk psychology assumes that there is "a single inner state which is active whenever a cognitive episode involving a given concept occurs and which can be uniquely associated with the concept concerned."_ But many advocates of connectionism take it to be a virtue of connectionist models that their strategy for conceptual representation can be "context sensitive." So, for example, in the sort of models that Smolensky favors, the representation of coffee when the coffee is in a cup will be somewhat different from the representation of coffee when the coffee is in a can._ And, Davies argues, if this sort of model turns out to be the right account of what goes on in human cognition, then folk psychology's conceptual modularity assumption is mistaken.
Yet another non-Sellarsian structural argument, one that does not appeal to connectionism, exploits some intriguing findings in cognitive social psychology. Those results suggest that, in some cases at least, people's sincere reports about their own mental states and processes do not match up very well with the mental states and processes that are actually responsible for their behavior. Rather, what people report in these cases seems to be driven by socially shared theories about how their behavior is to be explained. A bold hypothesis that has been proposed to explain the experimental results posits two largely independent cognitive sub-systems. One of them "mediates behavior (especially unregulated behavior), is largely unconscious, and is, perhaps, the older of the two systems in evolutionary terms. The other, perhaps newer, system is largely conscious, and its function is to attempt to verbalize, explain, and communicate what is occurring in the unconscious system," on the basis of "theories about the self and the situation."_ To turn this empirical speculation into an argument against folk psychology, we need an additional premise which claims that folk psychology embraces a more unified picture of the mind. According to this premise, folk psychology assumes that beliefs and desires play a central role in the processes guiding our non-verbal behavior and in the processes leading to our verbal behavior. Folk psychology claims that the very same pair of states which lead us to walk toward the refrigerator also lead us to explain our behavior by saying, "I want a beer." and "I think there is a beer in the refrigerator." If our verbal behavior and our non-verbal behavior really are subserved by independent systems, then this putative presumption of folk psychology is just wrong._
The version of the First Premise set out in Section 2 includes the claim that many of the nomological generalizations of folk psychology are couched in terms of content. It also claimed that content plays an essential role in folk psychology's scheme for individuating belief and desire tokens. A pair of belief tokens or a pair of desire tokens are (type-) identical if and only if they have the same content. But as we noted, this is far from a complete account of how folk psychology individuates propositional attitudes. To flesh it out, we have to say more about how folk psychology determines sameness or difference of content. The first of the semantic arguments that I'll sketch assumes that folk psychology exploits a "wide" account of content identity; the second assumes that folk psychology individuates contents holistically.
About twenty years ago, Hilary Putnam and Saul Kripke put forward a famous argument which allegedly showed that "meanings just ain't in the head."_ The now familiar story on which the argument is based asks that we imagine a planet in some distant corner of the universe which is all but identical to our own. On this planet each of us has a Twin - a molecule for molecule replica. The only difference between Earth and Twin Earth is that the stuff in their lakes and rivers is not H2O, but another clear, tasteless liquid, XYZ. Now, it is claimed, when Twin Stich says "Water is wet" folk psychology takes the content of the belief he is expressing to be different from the content of the belief that I express when I utter the same sequence of words. My Twin's belief is true if and only if XYZ is wet, while mine is true if and only if H2O is wet. If this is right, then as folk psychology sees it, the contents of beliefs and other propositional attitudes do not supervene on the non-relational physical properties of the believer, since ex hypothesis my Twin and I are identical in all our non-relational physical properties._ In the colorful though occasionally misleading jargon that has become commonplace in the literature, folk psychology's account of content identity is wide.
To spring their trap, the critics of folk psychology must introduce one additional premise. This one is not a claim about the commitments of folk psychology or folk semantics. Rather, it is a metaphysical thesis (or perhaps it's a methodological thesis -- I confess that I'm less than clear about where methodology ends and metaphysics begins). What it claims is that the only properties that may legitimately be invoked in scientific explanations of behavior, and thus the only properties that may be legitimately invoked in scientific psychological theories, are properties that supervene on the non-relational physical properties of the subject._ If this is right, then whatever explanation a scientific psychological theory offers for the behavior of an organism will apply as well to the behavior of the organism's physical replicas. There is, to put it mildly, considerable controversy surrounding this thesis. Some writers, myself included (but I was younger and much more naive at the time), have claimed that it is intuitively obvious that scientific psychology should treat organisms that behave in the same way to be psychologically identical. And since Putnamian Twins behave identically in all possible settings, psychology should treat them as psychologically identical. Others have tried to defend the thesis by deducing it from other, perhaps less controversial, metaphysical doctrines._ And still others have claimed that it is simply false._ But if it is not clear whether the thesis is defensible, it is clear that if the thesis is accepted then the argument has all it needs to show that folk psychology is in trouble. For, according to the version of the First Premise set out in Section 2, folk psychology includes lots of nomological generalizations that are couched in terms of the content of intentional states. But the Twin Earth argument (putatively) demonstrates that content does not supervene on the non-relational physical properties of an organism. And the metaphysical principle insists that the properties invoked in the generalizations of scientific psychology must supervene. So if scientific psychology has it right, then folk psychology must have it wrong.
A second semantic argument begins with the contention that folk psychology takes the content of a propositional attitude to be dependent in part on the network of other propositional attitudes that a person has. Thus, if the doxastic networks surrounding a pair of belief tokens are sufficiently different, the tokens will differ in content. One example that is supposed to illustrate this phenomenon focuses on the case of an elderly woman, Mrs. T, who gradually loses beliefs as the result of some degenerative disease._ Before the onset of the disease, she believes that McKinley was assassinated, and she has a whole slew of related beliefs of just the sort one would expect. But as the disease progresses, she loses the belief that McKinley was a U.S. President; then she loses the belief that assassinated people are dead; then she loses most of her beliefs about the differences between the living and the dead - she no longer has any idea what death is. Even at this advanced stage of her disease, she is still capable of answering the question: "What happened to McKinley?" by saying "McKinley was assassinated." However, it is claimed, folk psychology does not count the belief that underlies this answer as having the same content as the belief she had before her illness began. She no longer believes that McKinley was assassinated. The change in the doxastic surround has altered, perhaps even destroyed, the content of the belief that remains.
One way to parlay examples like this into an argument against folk psychology is to add a premise which claims that on the account of psychological state individuation that will be embraced by scientific psychology, the psychological state that causes Mrs. T to say "McKinley was assassinated" need not have changed at all as her disease progressed. To support the premise, a bit of science fiction is required. Imagine it is the case that people store information the way certain computers models do. They have long lists of syntactically complex symbolic structures stored in memory. Imagine further that the causal interactions of these symbol structures are akin to the causal interactions of their analogs in computer memories. These interactions depend only on the "shape" of the individual symbols and on the syntactic properties of the structures into which they are assembled. This story may be wildly mistaken, of course. But it is hardly unfamiliar. On the view of many observers, it is just the sort of account that is presupposed by most computational models in cognitive psychology. As Fodor has observed, these models are "really a kind of logical syntax (only psychologized)."_ For our purposes, the essential fact about models of this sort is that their symbol structures are individuated without any appeal to the other structures stored in memory. The same symbol structure might at one time be surrounded by thousands of related structures and at another time be surrounded by only a few related structures, or by none at all. Thus in models of this sort, symbol structures are not individuated in ways that are sensitive to their "doxastic surround."
O.K. Now we have all the pieces needed to assemble the argument for the eliminativists' Premise. If we have drawn the right conclusion from the Mrs. T case, then the content of a belief is sensitive to its doxastic surround. When the surround is very different, the content is different. And, since folk psychology assumes that beliefs are individuated by their contents, if the surround is very different, then the belief itself is different. But if the computational paradigm that we've just sketched is on the right track, then the psychological states that actually cause behavior are not individuated in a way that is sensitive to their surround. Since beliefs are individuated in a way that is sensitive to their surround, and the actual causes of behavior are not, beliefs are not among the causes of behavior. Applying this argument to the Mrs. T. case may make the point a bit more vivid. Since folk psychological belief individuation is surround-sensitive, none of the beliefs that Mrs. T has when her illness is far advanced can be the same as the beliefs she had before she became ill. But since the individuation of computational symbol structures is not surround-sensitive, Mrs. T may well have some of the same symbol structures in memory before and after her illness. If one of these structures causes her utterances of "McKinley was assassinated" both before and after her illness, then neither of these utterance is caused by a belief. Since folk psychology claims that utterances like these are caused by beliefs, folk psychology is wrong.
With a bit of fiddling, this argument can be recast along lines quite parallel to the argument in 3.3.1: Folk psychology couches many of its nomological generalizations in terms of content, and content is surround sensitive. The nomological generalizations in computational models of cognition are couched in terms of the syntax of mental symbol structures, and syntax is not surround sensitive. So if the computational models have it right, then folk psychology must have it wrong.
The arguments just sketched rely on some heavy duty assumptions about the account of psychological state individuation that will be embraced by scientific psychology. For the arguments to work, scientific psychology has to buy into what I have elsewhere called the "Syntactic Theory of the Mind."_ Other writers have attempted to argue from meaning holism to the conclusion that folk psychology is mistaken without assuming anything as controversial as the Syntactic Theory of the Mind. But these arguments typically rely on a much more virulent version of holism. On the version of holism that is assumed in the Mrs. T example, content identity requires a similar network in the doxastic surround. On the more extreme version of holism, a pair of belief tokens are identical in content only if they are embedded in identical doxastic surrounds._ If that is right, then no two people will have beliefs that are identical in content, nor will two time slices of the same person, provided the person is awake and the time slices are separated by a minute or two. But this, the argument continues, would make appeal to content useless in the generalizations of scientific psychology, since the goal of scientific psychology is to find nomological generalizations that apply to many people, or many organisms._ So, while the generalizations of folk psychology appeal to content, the generalizations of scientific psychology will have to be stated in content free terms. And, once again, if scientific psychology is right, then folk psychology is wrong.
A third semantic argument begins with the premise that the belief tokens that folk psychology classifies as having the same content can be extremely heterogeneous. There are a variety of psychological dimensions on which people can differ enormously and still be classified by folk psychology as having beliefs that share the same content. Perceptual capacities provide one cluster of illustrations of this phenomenon. Some people have sharp vision, others see poorly, and still others are blind. Yet there are circumstances in which folk psychology would attribute the belief that the traffic light has just turned green to all three sorts of people. Even someone like Helen Keller, whose perceptual deficits are quite staggering, might perfectly well believe that the traffic light has just turned green (if she is told that this is the case by someone she trusts, for example, or if she knows that the car she's riding in has been stopped at a red light, and she feels it begin to accelerate.) Cognitive skills provide another cluster of illustrations. Some people are swift and agile in reasoning, they see lots of logical connections, and they are quick to draw valid conclusions. Others are much slower and much more prone to logical errors. And there may well be people who, as the result of illness or brain damage, are simply incapable of drawing one or another kind of basic logical inference. Yet, under appropriate circumstances, folk psychology will attribute a belief with the same content to the clever, the retarded and the brain damaged._ Indeed, the premise maintains, folk psychology is often quite comfortable in attributing belief tokens with the same content to both animals and people. The dog and his master may both believe that the stick has been thrown down the hill. In the right setting, folk psychology may even sanction the attribution of beliefs with familiar contents to fish, or to bees. So for many propositions, p, it looks like the class of mental states that folk psychology will count as having the content that p will be very heterogeneous indeed. The neurological states subserving these beliefs will differ drastically both physically and functionally.
To complete the argument, we need yet another premise speculating about what a mature science of the mind/brain will look like. What it claims is that the heterogeneous content-categories invoked in the generalizations of folk psychology, categories that group together belief tokens in Einstein's head, in a brain damaged person's head, and in a dog's head, will be too inclusive to be of any use in that future science. Sophisticated sciences categorize states in terms of their causal powers, and from that point of view the dog's belief and Einstein's are too different to be grouped together. The nomological generalizations of the cognitive science of the future will invoke much less heterogeneous categories. At best, the premise maintains, folk psychology, with its coarse grained content-based generalizations, is going to miss most of the interesting generalizations. More likely, it's going to turn out that there are no true content-based generalizations to be found. If that's right, then once again folk psychology will be in conflict with the future science of the mind.
One of the most widely discussed family of arguments aimed at showing that folk psychology is a defective theory begins by arguing that the intentional properties of beliefs and desires, and the other characteristically "mental" properties invoked by folk psychology, are neither identical with nor reducible to physical properties. The are various routes to this conclusion. Perhaps the most plausible turns on claims about the possibility of multiple physical realizations of beliefs. Humans can think that 2 + 2 = 4. But surely, the argument insists, there is no reason to be chauvinistic. If Martians exist, they too may be able to think that 2 + 2 = 4, even if their brains are made of "green slime" whose chemistry is quite different from ours. Computers of the future, like HAL in the movie 2001, will also be able to think that 2 + 2 = 4. And it is plausible to suppose that there are indefinitely many physically different ways to build such a computer. Thus there is a vast, open ended class of physical systems that could think that 2 + 2 = 4. Since these systems have different physical properties, no physical property can be identified with thinking that 2 + 2 = 4. Since the class is vast and open ended, no reduction of intentional properties to physical properties is possible. This argument has been challenged in various ways._ One strong motive for challenging the argument is that if the conclusion is correct, it can be used in a second argument aimed at showing that intentional properties are causally impotent.
Here is an abridged version of that second argument, modeled on a more detailed exposition due to Robert Van Gulick._
1. Token Physicalism: Every intentional event token (i.e. every event-token having intentional properties) is identical with some physical event-token (i.e some event-token having physical properties).
2. The causal powers of a physical event-token are completely determined by its physical properties.
3. The Non-Reducibility of the Intentional: Intentional properties are neither identical to nor reducible to physical properties.
4. Thus intentional properties are not causally potent; they are causally irrelevant.
If this is right, and if, as we assumed in the previous section, folk psychology includes an opulent collection of nomological generalization that are couched in terms of content, then folk psychology must be very seriously mistaken.
For many friends of folk psychology, the last semantic argument that I'll mention is the most worrisome. In some ways it is also the most puzzling. The premise about content with which this argument begins is the claim that content (and related intentional notions) can't be "naturalized" - there is "no place for intentional categories in the physicalistic view of the world," and they "will prove permanently recalcitrant to integration in the natural order."_ But, the argument continues, if semantic content can't be naturalized, then it's not real - it doesn't exist at all. Here is how Fodor puts the point:
It's hard to see ... how one can be a Realist about intentionality without also being, to some extent or other, a Reductionist. If the semantic and the intentional are real properties of things, it must be in virtue of their identity with (or maybe of their supervenience on?) properties that are themselves neither intentional nor semantic. If aboutness is real, it must be really something else._
If it's true that intentional or semantic properties can't be naturalized, Fodor would insist, it follows that they can't be "reduced" to non-intentional, non-semantic properties. And if that's right, then they are not "real properties of things" at all.
To conclude the argument, we need only recall that, according to the version of the First Premise set out in Section 2, folk psychology assumes that propositional attitudes do have semantic content, it views sameness of content as a necessary element in individuating propositional attitudes, and it couches many of its generalizations in terms of content. Plainly, if content is not a real property of things, then folk psychology has got all this pretty badly garbled.
Before accepting that conclusion, however, there are at least three clusters of questions that need answering. First, what exactly would be required to "naturalize" content? In the brief passage I've quoted from Fodor, he mentions reduction, supervenience and property identity. In another paper he specifies that naturalizing content requires providing non-intentional necessary and sufficient conditions for the application of intentional predicates. And in still other places he says that sufficient conditions will do. Until we get a lot clearer on what "naturalizing" comes to, the remainder of the argument is going to be all but impossible to evaluate. A second set of questions focuses on the relation between naturalizing and being real. Why, one would like to know, would the fact that intentional properties can't be naturalized entail that they are not real properties of things? Is everything real reducible to (or supervenient upon, or definable in terms of) the physical (or the non-intentional, or the natural)? And if the answer is yes, what's the argument that makes this answer plausible? A third group of questions takes aim at the premise with which the argument begins. What reason do we have to think that content can't be naturalized? Is it simply that no one has figured out how to do it (whatever exactly it turns out to be)? If so, then it may well be that content has lots of company, since there are lots of things for which no one knows how to provide a reduction (or definition) or an account of how it supervenes on the physical. Are all of them unreal?
Good questions, these, but not easy ones. To address them seriously would require a whole paper. And, as it happens, Stephen Laurence and I have written just such a paper. It is reprinted as Chapter 5.
This completes my brief survey of arguments aimed at showing that folk psychology is a seriously defective theory. It isn't intended to be an exhaustive survey; there are a number of other arguments to be found in the literature. I've focused mostly on arguments that at one time or another I myself have been tempted to endorse. Nor do I claim to have presented the most subtle or persuasive version of each argument. Where brevity and subtlety conflict, I often opted for brevity. Still, I hope that some of these arguments strike you as plausible. For argument's sake, I'm going to assume that at least some of them are sound, and thus that some of the central claims made or presupposed by folk psychology are indeed mistaken. That assumption sets the stage for the questions I want to ask next: If folk psychology is a seriously mistaken theory, what conclusions ought we to draw about the existence or the scientific utility of intentional states? Should we conclude that a mature science of the mind/brain will not invoke beliefs and desires? Should we conclude that beliefs and desires are like witches and phlogiston -- they do not exist at all? 4. From Premises About Folk Psychology to Conclusions About the Existence of Beliefs and Desires: Lewis's Strategy
Eliminativists, of course, answer both of these questions in the affirmative, and as I noted earlier, by and large their opponents seem to agree. Most authors on both sides of the debate think that the battle over the virtues and shortcomings of folk psychology will be the decisive one. If folk psychology really is a seriously mistaken theory, then the eliminativists will have won._ But why is this conditional so widely accepted? Why, exactly, does the falsity of folk psychology lead to the eliminativists' conclusions? There is remarkably little sustained discussion of this question in the literature on eliminativism. So, rather than trying to extract an answer from the literature, I propose to take a different tack. In this Section I'll try as best I can to reconstruct the considerations that once persuaded me that if folk psychology is very wrong, then there are no such things as beliefs. Having said why I once accepted this crucial step in the eliminativists' argument, I'll go on, in Section 5, to recount how I came to doubt it, and I'll explain why I no longer find the line of argument set out in this Section to be very persuasive.
It's my guess that the cluster of views I'll recount in this Section has been widely, though often tacitly, accepted by lots of authors on both sides of the debate._ These views are the tacit theory that is most often lurking in the background when people impugn or defend the virtues of folk psychology and then draw ontological conclusions. I may, of course, be quite wrong about this. It may be that others saw (or thought they saw) some quite different link from premises about the shortcomings of folk psychology to conclusions about the non-existence of beliefs, desires and the like. I know of only two other likely candidates, however. I'll discuss one of them in Section 10 and the other in Sections 11 and 12.
For me the essential element linking the falsehood of folk psychology to the non-existence of the states it invokes was a theory about the meaning and reference of theoretical terms. Versions of the theory have been suggested by a number of distinguished philosophers including F. P. Ramsey and Rudolf Carnap, but the version that most influenced me was the one put forward by David Lewis. In a series of elegant and important papers Lewis developed an account according to which a theory typically provides an "implicit functional definition"_ of the terms it introduces. His "general hypothesis about the meanings of theoretical terms" is that "they are definable functionally, by reference to causal roles."_
To make his hypothesis plausible, Lewis offers the following illustration:
We are assembled in the drawing room of the country house; the detective reconstructs the crime. That is, he proposes a theory designed to be the best explanation of the phenomena we have observed: the death of Mr. Body, the blood on the wallpaper, the silence of the dog in the night, the clock seventeen minutes fast, and so on. He launches into his story:
X, Y and Z conspired to murder Mr. Body. Seventeen years ago, in the gold fields of Uganda, X was Body's partner... Last week, Y and Z conferred in a bar in Reading... Tuesday night at 11:17, Y went to the attic and set a time bomb... Seventeen minutes later, X met Z in the billiard room and gave him the lead pipe... Just when the bomb went off in the attic, X fired three shots through the French windows...
And so it goes: a long story. Let us pretend that it is a single long conjunctive sentence.
The story contains three names, `X', `Y' and `Z'. The detective uses these new terms without explanation, as though we knew what they meant. But we do not. We never used them before, at least not in the senses they bear in the present context. All we know about their meaning is what we gradually gather from the story itself. Call these theoretical terms (T-terms for short) because they are introduced by a theory. Call the rest of the terms in the story O-terms. These are all the other terms except the T-terms. They are the old, original terms we understood before the theory was proposed...._
In telling his story, the detective set forth three roles by X, Y and Z. He must have specified the meanings of the three T-terms `X', `Y' and `Z' thereby; for they had meanings afterwards, they had none before, and nothing else was done to give them meanings. They were introduced by an implicit functional definition, being reserved to name the occupants of the three roles...._
Suppose that after we have heard the detective's story, we learn that it is true of three people: Plum, Peacocke and Mustard. If we put the name `Plum' in place of `X', `Peacocke' in place of `Y', and `Mustard' in place of `Z' throughout, we get a true story about the doings of those three people. We will say that Plum, Peacocke and Mustard together realize (or are a realization of) the detective's theory._
...I claim [that] the T-terms are definable as naming the first, second and third components of the unique triple that realizes the story, ... [and thus] the T-terms can be treated like definite descriptions._
On Lewis's view, the moral to be drawn from this example applies quite generally. Theoretical terms are "defined as the occupants of the causal roles specified by the theory...; as the entities, whatever those may be, that bear certain causal relations to one another and to the referents of the O-terms."_ For Lewis, we have specified the sense of a term when we have specified its denotation in all possible worlds. And "[i]n any possible world, [T-terms] ... name the components of whatever uniquely realizes [the theory] in that world...."_
There are two further features of Lewis's account that deserve special emphasis. The first is the strategy Lewis urges for dealing with terms introduced by mistaken theories. Theoretical terms, Lewis tells us, are implicitly defined by the causal patterns specified in the theory that introduces the terms. Thus if no set of entities exhibit the specified causal patterns -- if the theory has no realization -- then the theory is mistaken. When this happens, Lewis maintains, the theoretical terms themselves will lack a denotation - they will refer to nothing.
[I]f we learnt that no triple realized the [detective's] story, or even came close, we would have to conclude that the story was false. We would also have to deny that the names `X', `Y' and `Z' named anything; for they were introduced as names for the occupants of roles that turned out to be unoccupied._
Now, as Lewis goes on to note, this is a rather extreme doctrine. For it entails that if a theory makes even a small mistake, then all of its theoretical terms will be denotationless. To make the view more palatable, Lewis offers "A complication:"
[W]hat if the theorizing detective has made one little mistake? He should have said that Y went to the attic at 11:37, not 11:17. The story as told is unrealized, true of no one. But another story is realized, indeed uniquely realized: the story we get by deleting or correcting the little mistake. We can say that the story as told is nearly realized, has a unique near realization. (The notion of a unique near realization is hard to analyze, but easy to understand.) In this case the T-terms ought to name the components of the near realization. More generally: they should name the components of the nearest realization of the theory, provided there is a unique nearest realization and it is near enough. Only if the story comes nowhere near to being realized, or if there are two equally near nearest realizations, should we resort to treating the T-terms like improper descriptions._
So it is only in those cases where the theory is very wrong that the theoretical terms refer to nothing. When the theory is only a little bit wrong the theoretical terms will still denote (provided there isn't more than one equally near "nearest realization").
Lewis does not say much about the boundary between these two sorts of cases. He makes no serious effort to specify how wrong a theory has to be before its theoretical terms fail to denote because it has no unique nearest realization that is "near enough". In the sentence following the passage just quoted he tells us that "scientific theories are often nearly realized but rarely realized, and that theoretical reduction is usually blended with revision of the reduced theory."_ This suggests that he thinks the boundary between theories that have near enough nearest realizations and those that do not is a blurry one, and perhaps that it has little theoretical significance. But, of course, in the context of arguments for and against eliminativism, the boundary is of enormous significance, since if we accept Lewis's account, it is the boundary that separates those false theories whose theoretical posits exist from those whose posits do not exist. If the boundary is a blurry one, then it may well turn out that even after all the scientific facts are in both about the mind/brain and about what folk psychology claims, there will be no way of determining whether or not the eliminativists' thesis is correct. The truth or falsehood of eliminativism may simply be indeterminate._ But even if it turns out that there is no sharp boundary between false theories whose posits exist and false theories whose posits do not exist, Lewis's account might still provide an essential step in arguments aimed at establishing that the entities invoked in one or another false theory do not exist. For the fact that a boundary is vague or indeterminate is fully compatible with there being clear cases on both sides of the divide. There is, after all, no sharp or principled divide between bald and hairy people, yet some people are clearly bald, while others are unmistakably hairy. So if Lewis is right about theoretical terms, then to show that the posits of a theory do not exist, it will suffice to show that the theory is very wrong and not just a little bit mistaken.
The second point about Lewis's account that merits special emphasis is that if we "think of commonsense psychology as a term-introducing scientific theory, though one invented before there was any such institution as professional science,"_ then everything he has claimed about theoretical terms in general can be applied straightforwardly to the theoretical terms of commonsense psychology. Moreover, on Lewis's view, this is the right way to think of commonsense psychology.
Imagine our ancestors first speaking only of external things, stimuli, and responses - and perhaps producing what we, but not they, may call Ausserungen of mental states - until some genius invented the theory of mental states, with its newly introduced T-terms, to explain the regularities among stimuli and responses. But that did not happen. Our commonsense psychology was never a newly invented term-introducing scientific theory - not even of prehistoric folk-science. The story that mental states were introduced as theoretical terms is a myth.
It is, in fact Sellars' myth.... And though it is a myth, it may be a good myth or a bad one. It is a good myth if our names of mental states do in fact mean just what they would mean if the myth were true. I adopt the working hypothesis that it is a good myth._
Putting together the two points that I've been emphasizing gives us just the link that the eliminativist needs to go from claims about the shortcomings of commonsense psychology to the conclusion that the posits of commonsense psychology do not exist.
If the names of mental states are like theoretical terms, they name nothing unless the theory ... is more or less true._
Indeed, if we suppose that commonsense psychology is an integrated theory containing or entailing claims about causal relations among various different kinds of psychological states, then, if Lewis's account is correct, problems in one part of commonsense psychology put our entire mental ontology in jeopardy.
[O]n my version of causal definability, the mental terms stand or fall together. If common-sense psychology fails, all of them alike are denotationless._
So, on Lewis's view, if folk psychology turns out to be seriously mistaken, it's not just beliefs and desires that will have to be dropped from our ontology. Pains, pleasures and other conscious states will have to go as well. Lewis himself sees little chance that things will work out this way. He identifies commonsense psychology with the psychological "platitudes which are common knowledge among us - everyone knows them, everyone knows that everyone else knows them, and so on."_ And he has no doubt at all that most of these platitudes will turn out to be correct. But we have been assuming that folk psychology is more than a collection of commonly accepted platitudes. My version of the eliminativists' First Premise takes folk psychology to be a largely tacit, opulent, internally represented theory. And if we grant that some substantial subset of the arguments set out in Section 3 are sound, then by any reasonable standard, folk psychology will turn out to be pretty badly mistaken. So it appears that Lewis's account of the meaning and reference of theoretical terms provides eliminativists with an attractive way of getting from the Premises of their argument to the conclusions._
In my own case, I blush to admit, Lewis's account was so attractive that until recently I was barely aware of how heavily I was relying on it. It had become one of the unnoticed and unquestioned assumptions on which the more controversial and fussed over parts of my philosophical view were built. But, as our deconstructionist friends are fond of noting, when the foundations are hidden so too are the cracks.
All this began to change, for me, while Ramsey, Garon and I were at work on the paper that is reprinted as Chapter 2. At that time, as luck would have it, I was asked to review Bill Lycan's book, Judgement and Justification. And while re-reading the essays in that volume I was brought up short by several brief passages that were not at all central to Lycan's projects. In those passages Lycan notes that most eliminativists seem to presuppose something like Lewis's theory about the meaning and reference of theoretical terms, and emphasizes that this is not an assumption that one gets for free. Accounts like the one Lewis develops are not the only game in town. Indeed, in the recent philosophy of language literature, they are not even the most popular game in town. On the very different account of reference that Lycan favors, premises detailing untenable features of the commonsense conception of beliefs and desires simply do not support the sort of ontological conclusions that eliminativists are wont to draw. Here are a couple of passages in which Lycan sets out his view:
I incline away from Lewis's Carnapian ... cluster theory of the reference of theoretical terms, and toward Putnam's causal-historical theory. As in Putnam's examples of `water,'`tiger,' and so on, I think the ordinary word `belief' (qua theoretical term of folk psychology) points dimly toward a natural kind that we have not fully grasped and that only mature psychology will reveal. I expect that `belief' will turn out to refer to some kind of information-bearing inner state of a sentient being, ... but the kind of state it refers to may have only a few of the properties usually attributed to beliefs by common sense._
I am entirely willing to give up fairly large chunks of our commonsensical ... theory of belief or of desire (or of almost anything else) and decide that we were just wrong about a lot of things, without drawing the inference that we are no longer talking about belief or desire._
There are a several points suggested in these passages that I want to emphasize and endorse. First, Lycan is certainly right to note that description based accounts of reference of the sort offered by Carnap and Lewis are not the only option available._ Putnam, Kripke, Devitt and others have suggested a quite different family of theories about reference which many philosophers find more plausible. According to these causal-historical accounts, the reference of a term is determined by the appropriate sort of causal chain connecting users of the term with previous users from whom they acquired the term, and ultimately proceeding back to an event or series of events in which the term is introduced to refer to a certain object or kind. Since serious alternatives to description theories have been proposed and defended with considerable ingenuity, eliminativists surely cannot legitimately do what I did for many years. They cannot simply take some version of the description theory for granted. If they are going to rely on it to get from their Premises to their conclusions, they will have to defend it against the competition. It is Lycan, I think, who deserves the credit for starting the deconstruction of the eliminativists' deconstruction by stressing that eliminativists themselves are relying on a barely acknowledged theory of reference that might well turn out to be unacceptable.
Second, Lycan is quite right in claiming that causal-historical theories will not sustain an argument from premises about the falsehood of folk psychology to conclusions about the non-existence of the mental states that folk psychology invokes. Indeed, it is one of the selling points of causal historical theories that they do a much better job than description theories at handling what Devitt and Sterelny call "the problem of error."_ One way to explain the problem is to imagine a community of ancient star gazers who have what we now know to be wildly mistaken views about the objects visible in the night sky. They think that most of them are holes in a black and otherwise solid celestial dome through which we can see the light in the heavenly region that surrounds the dome. But even though their theory about the stars is about as mistaken as it is possible to get, it seems to make perfectly good sense to say that when these ancients spoke about the objects in the night sky, they were talking about stars. It is plausible to suppose that there was a term in their language that referred to stars, and they sometimes used that term to make profoundly mistaken claims about stars. Now, as advocates of causal-historical accounts of reference often note, cases like this pose a serious problem for description based accounts of reference. For on description based accounts it seems to follow that the ancients in our little tale aren't talking about stars at all. Since their theory is so seriously mistaken, description theories will entail that the astronomical terms used by the ancients refer to nothing -- they are, as Lewis might say, "denotationless". Causal-historical accounts, by contrast, do not entail this sort of counter-intuitive conclusion. If one of the ancients' astronomical terms was introduced in settings that provided the right sort of causal links to stars, and if the history of transmission of that term was of the right sort, then their term refers to stars no matter how badly informed they may be about what stars really are. Thus, as Lycan rightly notes, causal-historical accounts of reference will not enable eliminativists get from their Premises to their conclusions.
These considerations are a clear indication that eliminativists have some work to do. But they do not suffice to show that the eliminativists' argument can't ultimately be made to work. While the "problem of error" argument certainly casts some doubt on description theories of reference, it is not, by itself, enough to show that description theories are untenable. Nor does it show that some causal-historical theory of the sort that Lycan favors is correct. For those theories seem to have problems of their own, and in just the opposite direction. If description theories sometimes make it too hard to refer, causal-historical theories sometimes make it too easy. To see the point, consider some of the parade cases of ontological elimination that eliminativists are fond of citing. There are no witches and there is no such thing as phlogiston. So when our forebears used the words `witch' and `phlogiston' they were referring to nothing. But if Lycan is right about reference, it's hard to see how this claim could be sustained. If the term `witch' was introduced and transmitted in the right way, then it actually refers to certain women who behaved in strange or socially unacceptable ways. Of course witches "may have only a few of the properties usually attributed to [them] by common sense." And the term may "dimly point toward a natural kind that we have not fully grasped." If we follow Lycan's lead, we should be "entirely willing to give up fairly large chunks of our commonsensical ... theory ... and decide that we were just wrong about a lot of things," without drawing the inference `witch' is a term that fails to refer. And, of course, the same can be said, mutatis mutandis for phlogiston. These conclusions are no less counter-intuitive than the conclusion that the ancient star gazers were not talking about the stars. So Lycan and other opponents of eliminativism can't simply assume that some version of the causal-historical theory of reference is correct, and thus that terms like `belief' and `desire' refer no matter how wrong we ultimately discover folk psychology to be.
Where does all of this leave us? Here's one assessment of the situation: The most promising way for eliminativists to get from their Premises to their conclusions is to invoke a theory of reference. And one well worked out theory (Lewis's) will do just fine. But there are problems with that theory and with other versions of the description theory as well. They seem to entail some quite counter-intuitive claims. Moreover, there is another widely accepted family of theories about reference on the market, causal-historical theories, and theories in that family will not enable eliminativists to get from their Premises to their conclusions. Causal-historical theories have problems of their own, however. They too seem to entail some quite counter-intuitive claims. So it looks like both eliminativists and their opponents would be well advised to turn their attention to the theory of reference. They have to determine which sort of account of reference is right, and find some plausible way of explaining away the objections to that account. Or perhaps the objections indicate that neither description theories nor causal-historical theories are right, and that some new, better theory_ of reference is needed. But in any event, the theory of reference has now moved to center stage. Before we can determine whether the eliminativists' Premises support their conclusions, we're going to have to settle which account of reference is correct.
At this juncture many philosophers who were trained to do philosophy the way I was would be tempted to roll up their sleeves and jump into the fray -- constructing new theories of reference (or fine tuning old ones), exploring the consequences of these theories, developing arguments, looking for counter-examples. That's what we do for a living; it's how the game is played. So, having persuaded myself that the assessment offered at the end of the previous section was correct, I set to work looking for a better theory of reference. But after lots of work and very little progress, I gradually became convinced that this is the wrong way to proceed. What changed my mind was not that it was hard to come up with alternative theories of reference. Quite the opposite, it was actually quite easy. I have a notebook full of them. But as my list of alternatives grew, I found that I got less and less clear about how I was supposed to evaluate these alternatives. How could I tell which one was the right one?
At first I assumed that the problem I was confronting was just another case of a familiar epistemic problem that has to be confronted by theory builders in almost every domain. Very few areas of inquiry have anything even close to established decision procedures for determining whether or not a theory is correct. There may be no obvious or generally accepted set of procedures for evaluating alternative theories of reference, but much the same can be said for theories in physics or biology or archeology as well. Gradually, however, I became convinced that the problem I was having in evaluating theories of reference was deeper than this. For in trying to assess theories of reference, the familiar epistemic problem is exacerbated by a quite basic methodological problem (or perhaps it's really a metaphysical problem -- I've already confessed that the distinction is one I have trouble drawing). Other domains may not be entirely free from this methodological problem, but in the theory of reference it arises in a particularly acute form. In just about every area of inquiry, it's hard to determine whether a given theory is successful in capturing or explaining the facts that it is intended to capture or explain. That's what I've been calling the "epistemic" problem. But in most parts of physics or biology or archeology it is pretty clear what the theory is expected to do. Though there may be a bit of squabbling about it from time to time, there is typically considerable agreement about the sort of facts that a theory is expected to describe or explain. In the theory of reference, by contrast, it is far from clear what sorts of facts the theory is supposed to account for. Indeed, it is my suspicion that, while the issue is only rarely a topic on which they have explicitly formulated views, different writers have quite different expectations. And, no doubt, some of the disagreement about which theories of reference are most promising can be traced to this underlying, largely tacit, disagreement about the job that a theory of reference is expected to do. Other writers have no coherent views at all about what a theory of reference is expected to do. It is not surprising that they have a particularly hard time figuring out which theory is best. My problem, in attempting to evaluate my growing collection of alternative theories, was that, without being clearly aware of it, I fell squarely into this latter category. I didn't know what a theory of reference was supposed to do.
When all of this finally came into focus, I decided to put my collection of theories on the back burner for a while and attend, instead, to the methodological question that had to be clarified before I could make any progress in assessing alternative accounts of reference. So I started thinking about the sorts of projects people might have in mind when they debate the virtues of various theories of reference. The decision to concentrate on the methodological question turned out to be a pivotal step in the intellectual adventure (or misadventure) that I'm recounting in this Chapter. For it ultimately led me to the conclusion that my entire conception of the eliminativism debate was radically mistaken. But here I am getting way ahead of myself. To explain how I reached that conclusion, I'll start by sketching the most plausible answers I have come up with to the question about what a theory of reference is supposed to do.
One family of answers begins with the observation that appeal to people's spontaneous judgements, or their intuitions as philosophers often call them, seems to play a central role in many debates about the virtues of competing theories of reference. Sometimes the intuitions invoked concern actual cases of language use. But often they are intuitions about hypothetical or imaginary cases, some of which can be more than a bit bizarre. We have already seen several examples of the way in which intuitions are invoked in these debates. In explaining "the problem of error" argument I imagined a community of ancient star gazers who had a seriously mistaken theory about the objects visible in the night sky. Description theories, I noted, typically entail that terms embedded in seriously false theories don't refer to anything, and thus if description theories are right then the ancients aren't actually referring to the stars. But this, the critics of description theories insist, is "counter-intuitive". Our intuitions tell us that the ancients were referring to the stars, and this conflict with our intuitions poses a problem for description theories. Similarly, the intuition that `witch' does not refer to anything played a crucial role in the objection I offered against Lycan's views on reference.
In his famous monograph, Naming and Necessity, Saul Kripke sets out a number of cases designed to bring out our intuitions about the reference of proper names. With a bit of familiar elaboration, one of these cases might be put as follows: Suppose that in biblical times there really was a man who survived for three days and three nights in the belly of a great fish. After escaping from the fish, he was killed by bandits, and his memory has vanished without a trace. Suppose also that at roughly the same time there was another man who did not endure any such exciting adventure, but about whom, for some reason, people told increasingly tall tales. Over the years, one of these tales evolved into the Biblical story of Jonah, which has been passed down from generation to generation. Now if this is really what happened in history, to which of these men do readers of the Bible refer to when they ask questions or make claims about Jonah? Description theories will typically claim they refer to the first man, since most of the claims they would make using the name `Jonah' are true about him. But, many philosophers maintain, our intuitions support the opposite judgement. If the facts are as stipulated in the story, our intuitions tell us that when a reader of the Bible asks questions in which the name `Jonah' is used, the name refers to the second man, and thus most of what the Bible says about Jonah is false. Examples like this convinced many philosophers that description theories of the reference of proper names were wrong, and that causal-historical theories were more plausible._
Clearly, appeal to intuitions plays an important role in debates about the theory of reference. But why? Why should these intuitions be at all relevant to questions about the correctness or incorrectness of a theory of reference? There are a variety of answers that might be offered here. The most straightforward of them is that intuitions are relevant because capturing the relevant intuitions is what a theory of reference is supposed to do -- producing a theory that entails the intuitions is one of the goals of the theory of reference. This answer can be elaborated in two different ways. To pull them apart, it will help to consider the analogy between the theory of reference and grammatical theory.
In grammar, too, intuitions play a central role. They are far and away the most important source of data for the descriptive grammarian. One attempt to explain the role of intuitions in grammatical theory urges that a correct grammatical theory just is an idealized theory of grammatical intuitions. The goal of grammatical theorizing, on this account, is to produce the simplest and most elegant theory that captures most of a native speaker's grammatical intuitions._ An alternative account of the role of intuitions in grammatical theory urges that the real goal of the theory is not to capture intuitions but rather to characterize the grammatical principles used by the psychological mechanism that gives rise to the intuitions. This view typically assumes that there is a distinct underlying psychological mechanism that subserves grammatical processing, and that this same mechanism plays a central role in the production of grammatical intuitions. The grammar mechanism interacts with other components of the mind, including perceptual processors, attention mechanisms, short term memory, etc., and together these mechanisms produce the linguistic intuitions that people report. Thus the intuitions are a good source of evidence about the grammatical principles used by the mechanism. But they are not an infallible guide, since memory limitations, failures of attention and other factors may produce various sorts of "performance errors" including intuitions that fail to reflect a speaker's underlying "grammatical competence."_
Now one way to elaborate on the idea that the goal of a theory of reference is to capture the relevant reference intuitions is to view the theory of reference as analogous to the grammatical theory on the first account of grammar sketched above. On this view, a theory of reference just is an idealized theory of reference intuitions. The theorist's goal is to produce the simplest and most elegant set of principles that captures most of the intuitions that people have about reference._
The other way to develop the idea is to suppose there is a systematic body of information or a set of principles stored in the mind which plays a central role in producing reference intuitions. The goal of a theory of reference, on this view, is to give an accurate account of those mentally represented principles, or of the word - world mapping that they specify. In order to produce intuitions, the mentally represented principles must interact with other components of the mind including those responsible for attention, inference and short term memory. Thus the intuitions that people offer will not always be an accurate reflection of the principles, though they will be a rich source of data for a theorist to use in trying to determine what those principles actually are. On this account, theories of reference have much the same status as theories about people's internally represented "folk physics" discussed in Section 2. In both cases, the goal is to describe an internally stored body of information, and in both cases people's spontaneous judgements or intuitions provides a rich, though occasionally misleading, source of data for the theorist to use. Theories about folk physics and theories of reference, on this account, are both descriptive psychological theories. The theories are correct if they accurately describe the principles of an internally represented commonsense theory. So on this view, it seems natural to think of the theory of reference as an account of another sort of folk theory which might be called "folk semantics."_
The analogy between theories of reference and theories about folk physics and the analogy between theories of reference and grammatical theories are both useful in explaining the view that the goal of the theory of reference is to correctly describe our tacit "folk semantics". But there is an important distinction between these two analogies, and elaborating on that distinction will set the stage for a quite different account of the job of the theory of reference. To see the distinction I want to draw, it is crucial to keep in mind that in both grammar and folk physics there are two theories to keep track of. A researcher who is interested in characterizing the folk physics used by a group of subjects wants to describe a theory that is represented in the minds of her subjects. Her description of that theory is itself a theory -- a theory about what's represented in her subject's heads. And, of course, her theory about what's in her subject's heads might be wrong. It is also possible that the subject's theory might be wrong. Indeed, the work on folk physics recounted in Section 2 indicates that the physical theory inside the heads of many subjects is wrong. Now let's consider grammar. Here too, there are two theories to keep track of. The descriptive grammarian is attempting to specify a set of principles that are represented in the minds of speakers of the dialect the grammarian is concerned with. The specification that the grammarian offers is thus a theory about the principles her subjects are using. And, just as in the case of folk physics, the grammarian's theory may be wrong. She may mis-characterize the principles in her subject's heads. Moreover, as in the case of folk physics, the principles inside the subject's heads can themselves be regarded as a theory, since (we have been assuming) one of the things they do is entail lots of claims about the grammatical properties of sentences in the speaker's dialect.
Now we can raise the question that brings out the important distinction between grammar and folk physics: Can the grammatical theory inside the speaker's head be wrong in the way that folk physics can? Is it possible, for example, for a speaker's internalized grammar to entail that a sentence in the speaker's dialect is grammatical when it isn't? The answer to this question turns on the sort of answer we accept to a cluster of further questions about the nature of grammatical properties themselves, questions like: What is it for a sentence in a dialect to be grammatical? In virtue of what does a sequence of phonemes count as a grammatical sentence? There are various answers that might be explored here. The one that seems to be favored by Chomsky and some other leading figures in linguistics is that a sequence of phonemes is grammatical in a dialect if and only if it is classified as grammatical by the grammar inside the heads of the speakers. It is the grammar itself that determines whether or not a phoneme sequence is grammatical. If this is right, then it follows that the answer to the questions at the beginning of this paragraph is no. The grammatical principles inside a speaker's head can't be wrong in the way that folk physics can be wrong. For what makes it the case that a phoneme sequence is grammatical is that it is classified as grammatical by the rules or principles inside the speaker's head. Of course it is possible that a member of some community might have a set of grammatical rules inside his head that is slightly different (or very different) from the rules inside the heads of other members of the community. If this happened, it might well be the case that the non-conforming grammar classified as grammatical some phoneme sequences that the grammars inside other heads classified as ungrammatical. But, on the account of grammatical properties that I am attributing to Chomsky, this would not count as an error on the part of the non-conforming speaker or his grammar. Rather, it would be the case that the non-conforming speaker spoke a different dialect. Perhaps he is the only speaker of the dialect, in which case it is best described as an "idiolect." However, the claims entailed by the speaker's internalized grammar cannot possibly be wrong about sentences in his own idiolect, since the grammatical properties in that idiolect are determined by the internalized grammar of the idiolect._
Let's return, now, to the theory of reference and the analogies with grammar and with folk physics. Obviously, if the job of a theory of reference is to describe an internalized folk semantics, then any particular account that a theorist gives may be mistaken. The theorist may misdescribe the folk semantic theory inside people's heads. But what about the folk semantic theory itself. Can it be mistaken? Here, as before, the answer turns on what we say about some further questions -- this time questions about the nature of semantic properties (or relations), properties like reference: What is it for a term in a language to refer to an object? In virtue of what does a term count as referring to an object? If we push the analogy with the Chomskian account of grammatical properties, the answer is that it is the internalized folk semantic theory which specifies the conditions a term must meet if it is to refer to an object. The reference relation just is whatever the internalized folk semantics says it is. And if that's right, then of course, the internalized theory can't be mistaken. If two people have different internalized folk semantic theories, then the notions of reference that they are using are simply different. When they use the term `refer' they are talking about different relations. Occasionally such people may appear to be disagreeing about what refers to what in a particular case. But actually they are not disagreeing at all. Rather, their situation is much the same as the situation of people who speak different idiolects who appear to be disagreeing about the grammaticality of a particular sentence. They are not disagreeing at all. Assuming their judgements accurately reflect their internally represented theories, they are both right.
Suppose, however, that the right analogy to push is not the one between theory of reference and grammar but rather the one between theory of reference and folk physics. In that case, there will be no guarantee that our internalized folk semantic theory is correct. Folk semantics, on this view, is just a collection of commonsense beliefs about reference and what determines reference, and the real facts about reference, like the real facts about physics, are as they are quite independently of what our folk theory may say about them. So if folk semantics is like folk physics, then our attempt to describe the commonsense theory of reference may be an interesting bit of psychology, but there is no reason to suppose that it will tell us much about reference. If we want to learn about the laws governing motion we don't study what ordinary folks think, rather we study the science of physics; and if we want to learn about the nature of various diseases, we don't study folk theories of disease,_ we study medicine and pathology. So if folk semantics is like folk physics, then if we want to learn about reference, we shouldn't study the folk conception of reference. Rather we should study the science that tells us about reference. But, it would seem, at just this point the analogy between folk semantics and folk physics hits a snag. For while there is a well developed science of physics for the person interested in the laws of motion to study, and a well developed science of pathology for the person interested in disease to study, it is not clear that there is any science at all whose business it is to investigate the real nature of reference.
It might be thought that this shows that the analogy between folk semantics and folk physics is indefensible, and the analogy with grammar is the only tenable one. But I am inclined to think that this conclusion might be too hasty. For while it is true that the notion of reference plays no role in any well developed science, it might well be the case that one or another reference-like word-world mapping relation will prove to be of considerable importance in various domains of empirical investigation. Perhaps linguistics will be able to make good use of such a relation; or perhaps parts of cognitive psychology or evolutionary biology will find a need for such a relation. Or, turning to very different domains of inquiry, perhaps anthropology or history and sociology, particularly the history and sociology of science, will find an explanatory use for a word-world mapping that's not too different from the intuitive notion of reference. There is no way of knowing a priori whether any of these possibilities will pan out. The only way to find out is to elaborate various word-world mapping relations and then try to put them to use in one or another scientific or historical project. The job of hunting for scientifically useful word-world relations is, of course, not distinct from actually doing the science in question. Rather it is best thought of as an activity that might be pursued in the early stages of the development of the science, a sort of proto-science, if you will. So the alternative to viewing the theory of reference as attempting to characterize a relation specified by folk semantics is to view it as a kind of proto-science which tries to find or construct word-world relations that will be useful in some explanatory project or other.
Perhaps I am being too cautious here. I have met a few philosophers who think that there already are up and running sciences that invoke the notion of reference. Linguistics is the candidate most often mentioned, though various parts of cognitive psychology are sometimes mentioned as well._ I am inclined to be more than a bit skeptical about the claim that any of these areas of inquiry make genuinely explanatory use of a reference-like word-world relation. But for present purposes there is no need to enter into a debate on the matter. For if it is true that linguistics or parts of cognitive psychology, or some other discipline already exploits a notion of reference, then the project for the theory of reference can be viewed as providing an explicit description or explication of the reference relation that these disciplines are using. This sort of project is quite familiar in the philosophy of science. For even in well developed sciences it is sometimes the case that researchers will use a concept quite productively without providing a fully explicit or philosophically satisfying account of the concept. In these cases, philosophers of science sometimes step in and try to make the notion in question more explicit. In recent years there have been illuminating studies of fitness, space-time, grammaticality and a host of other notions._ Part of this work can be viewed as straightforward conceptual description, where the concepts being characterized are those of working scientists. Often however, philosophers of science discover that the concepts that they are studying and the theories in which they play a role are uncomfortably vague or obscure in various ways. When this happens, it is not at all uncommon for philosophers of science to propose improvements in the concepts and theories they are describing. Typically, it is no easy matter to say where description stops and construction begins, and for most purposes it hardly matters. Nor is there any clear boundary between this activity and the sort of proto-science described above. In both activities the goal is to provide explicit accounts of kinds or properties or relations that will be (or already are) useful in some scientific project. The proposal I'm offering in this section is that building a theory of reference might be taken to be an activity of just this sort.
If this proposal is accepted, then the appeal to intuitions that plays such a large role in the philosophical literature on reference will have to be viewed with considerable skepticism. For our intuitions are, at best, an indicator of our current commonsense conception of the reference relation, and of the theory in which that conception is embedded. This may be a reasonable place to start in trying to construct a scientifically useful account of reference -- we have to start somewhere, after all. But on the proto-science view, the fact that a proposed account of the nature of reference flies in the face of our intuitions provides little reason to reject it, just as the fact that the layperson finds a proposed account of the nature of space or disease or water to be counter-intuitive provides little reason to reject that account.
Finally, before turning to other matters we should note that there is no a priori reason to suppose that the proto-scientific project of characterizing scientifically useful word-world relations will yield a unique result. For it may turn out that the word-world mapping useful in linguistics is different from the one useful in cognitive psychology, and that both of these are different from the one useful in anthropology or in the history and sociology of science. Or it may turn out that two or more word-world relations are explanatorily useful within the same science in much the same way that the pre-modern notion of speed divided into the Newtonian notions of velocity and acceleration._
Let's pause for a moment to take stock of where we are, how we got there, and where we might go from here. In Sections 1-3, I sketched the overall structure of the eliminativists' argument, and explained how I thought the Premises might be most charitably construed. In Section 4, we saw that one way in which the eliminativists' Premises might lead to their conclusions would be to invoke an additional premise -- a description-theoretic account of reference. But in Section 5, we saw that this additional premise is not one that the eliminativists get for free. There are other accounts of reference to be reckoned with, and which account is correct is a hotly contested issue. Moreover, as we saw in Section 6, it is far from clear what would settle the dispute between advocates of various theories of reference, since it is not clear what a correct theory of reference is supposed to do. Two possibilities were proposed. On one the theory of reference aims at describing the principles underlying our intuitions about what refers to what in various cases. On the other, the theory of reference aims to specify a word-world relation that will be useful in some scientific project. That's where we are, and how we got there. Where should we go from here?
It looks like the most reasonable next step is to use our two accounts of what a theory of reference is supposed to do in order to assess alternative theories of reference. The theory of reference that comes out looking most promising can then be plugged into the eliminativists' argument and, if we're lucky, it will tell us whether or not the conclusions actually do follow from the Premises. When I reached this point in my own thinking, that's just what I decided to do. It might seem that the existence of two quite different accounts of what a theory of reference is supposed to do would have posed a problem for this strategy. But actually it didn't, at least not for me. For on the proto-science account, there is no saying what reference is until we have made some progress at building a science in which a reference-like word-world mapping plays a role. And as I have already noted, I don't think there is anything around that fills that bill. Nor do I have any idea how to construct such a science. So, from a practical point of view, the only way to make progress is to concentrate on the account that views a theory of reference as an attempt to describe the intuitive reference relation, the one specified by folk semantics. That's the way I decided to proceed.
My approach was as direct and head-on as it could be. The question that concerned me was what the tacit theory that guides our folk semantic intuitions entails about the reference of terms that are embedded in theories which we now take to be seriously mistaken. To answer the question, I set out constructing examples designed to elicit the relevant intuitions. Some of the examples were comic book versions of actual events in the history of science:
Here's what Democritus said about atoms (or what Mendel said about genes): ....
Here's what modern science says about atoms (or genes): ....
If we assume that modern science is correct, was Democritus actually referring to the same things that modern science is? (or: ... was Mendel actually referring to the same things that modern science is?)
Others were completely fictional:
The Tuoba people in the Highlands of Papua New Guinea have the following theory about the cause of a certain set of symptoms (or about what sometimes causes the moon to go dark in the clear evening sky, or ....): ....
Modern medicine (or astronomy, or ...) explains these symptoms (or these eclipses, or...) as follows: ....
If we assume that modern medicine (or astronomy, or ...) is correct, is it the case that the Tuoba word `---' actually refers to what contemporary scientists call `***'? (Or: ...is it the case that what the Tuoba call `---' is the same thing as what modern scientists call `***'?)
Generating a range of quite varied examples proved to be a fairly easy task. A rainy afternoon yielded several dozen. As I was constructing the examples, however, I was struck by the fact that I myself had no firm intuitions about most of them. Initially, this was no great cause for concern, since it often happens in linguistics that when one plays with sample sentences long enough, one develops a sort of grammatical "tin ear" and can no longer tell which ones sound grammatical and which do not. To get around the problem of my tin ear, I presented the examples to several of my graduate students and several of my colleagues. This was not a great success. Many of them claimed to have no firm intuitions either. Though in some cases, some people claimed to have firm intuitions in favor of an affirmative answer, while others claimed to have firm intuitions in favor of a negative answer. That was a bit of a bother, of course, but it wasn't a surprise. On questions like this, one expects to find philosophy departments full of people with tin ears, particularly a department like mine. So I resolved that I would have to do it the hard way, and run a little experiment with naive subjects. Fortunately, we have lots of them available at my university. Our Introduction to Philosophy classes are very popular. Thus I wrote up a pilot version of a questionnaire containing various fictional examples of the sort displayed above, and asked students in several Introduction to Philosophy classes to fill them out at the beginning of the class period. After doing some preliminary analysis on about a hundred responses, however, I decided to drop the project. For what I found was that the intuitions of these naive subjects were no more firm or consistent than those of my graduate students and colleagues. There was little agreement on the cases; some students protested that they really didn't know what to say; and small variations in the wording of the questions seemed to produce a substantial shift in the sorts of answers I got. Perhaps there were some systematic and widely shared principles about reference underlying the very messy data that I was collecting. But if so, I couldn't find them. My conclusion from this little pilot study was that there probably are no systematic, widely shared principles that guide people's judgements when they are presented with questions about the reference of terms in mistaken theories.
I don't pretend for a moment that the little bit of data I've collected is sufficient to establish my conclusion. And I would certainly be pleased to see the results of a much more serious and careful study. However, it is perhaps worth noting that it would not be at all anomalous if the facts about people's intuitions are as I believe them to be. The assumption on which my experiment was based is that the judgements people offer are guided, in part, by a largely tacit theory ("folk semantics"), in much the same way that commonsense intuitions about the motions of middle sized physical objects are guided by a largely tacit theory. And, as noted earlier, it is far from obvious that this assumption is correct._ But even if it is correct, there is certainly no a priori reason to suppose that the theory must entail some definite judgement about every imaginable case. Lots of theories, tacit and explicit, are incomplete and say nothing at all about some range of cases. Indeed, commonsense theories should perhaps be expected to remain silent on cases that are far removed from the workaday situations in which such theories earn their keep. Where nature doesn't itch, folk theory has no need to scratch.
Suppose I am right, and that the commonsense theory underlying our intuitions about reference says little or nothing about the reference of terms embedded in theories that we take to be seriously mistaken. What follows about the debate over eliminativism? If we assume, as I have been, that terms like `belief' and `desire' can be viewed as part of the theoretical vocabulary of an opulent, largely tacit, commonsense theory, and that the theory is seriously mistaken in a variety of ways, then if I am right about folk semantics, it follows that questions like:
Does `___ is a belief' refer to anything?
will not be answered either positively or negatively by folk semantics. Folk semantics leaves these questions unsettled. But now if we also assume that what folk semantics says about reference is all there is to say about the matter, in much the way that what the grammar of an idiolect says about grammaticality in the idiolect is all there is to say about the matter, it follows that claims like:
`___ is a belief' refers to X.
`___ is a belief' refers to something.
are simply indeterminate -- they are neither true nor false. And that looks like bad news for both sides in the eliminativism debate, since beliefs exist if and only if `___ is a belief' refers to something. So the upshot of the argument seems to be that the eliminativists' basic ontological claim, the claim that beliefs and desires do not exist, is neither true nor false. It simply has no determinate truth conditions._ Since I had spent a good part of my time over the last two decades trying to determine whether eliminativism was true or false, this was a rather startling conclusion. And there were more surprises to come.
Some years earlier, while working on a book concerned mainly with issues in epistemology, I had elaborated an argument aimed at showing that the reference relation favored by our intuition is just one rather idiosyncratic member of a large family of more or less similar relations._ The fact that our intuitions pick out that particular relation rather than one of the other members of the family is, I argued, little more than an historical accident. As the indeterminacy argument sketched in the last few paragraphs was taking shape, I began to realize that if my claim about the idiosyncrasy of reference was correct, it might have some quite unexpected implications for the eliminativism debate. For the idiosyncrasy of reference suggests that even if I am wrong about the indeterminacy of the eliminativists' thesis, the thesis may nonetheless turn out to be idiosyncratic and uninteresting. To explain how I reached that conclusion, I'll have to begin by summarizing the argument for the idiosyncrasy of the intuitive reference relation.
That argument starts with a hunch, albeit a widely shared one. While there are lots of theories of reference on the market these days, my hunch is that the accounts that do the best job at capturing people's relatively firm and stable intuitions about reference are not those that follow the path staked out by Ramsey and Lewis, but rather those that tell what Lycan calls a "causal-historical" story. The basic idea of these theories, as we saw earlier, is that words get linked to things in the world via causal-historical chains. The first step in creating such a chain is a "grounding" or a "reference fixing" - an event or process (or, more commonly, an array of such events or processes) in which a term is introduced into a language to designate an object or a kind of objects. Following this there is a series (often a very long series) of reference preserving transmissions, in which the term is passed from one user to another, preserving the reference that was fixed when the term was introduced. But, of course, not just any way of introducing a term into a language will count as grounding the term on a particular object or kind of objects, and not just any way of passing a term from one user to another will count as a reference preserving transmission. The legitimate groundings and transmissions will be those embedded in causal-historical chains that are sanctioned by intuition. When one looks carefully at the class of groundings and the class of transmissions that pass this test, however, it appears that in each class the allowable events are a mixed bag having at best a loosely knit fabric of family resemblances to tie them together. The grounding of `water' (or some ancestor of the term) on water was surely markedly different from the grounding of `helium' on helium. The heterogeneity of intuitively acceptable groundings looks even more extreme when we consider the various ways in which predicates like `kangaroo' `asteroid,' `mutation,' `electron,' `quark,' `superconductivity' and `strange attractor' were introduced. The class of intuitively acceptable reference preserving transmissions is comparably diverse. What ties all the reference determining causal-historical chains together is not any natural property that they share. Rather, what ties them together is simply the fact that commonsense intuition counts them all as reference fixing chains.
But if it is indeed the case that common sense groups together a heterogeneous cluster of causal-historical chains, then obviously there are going to be lots of equally heterogeneous variations on the commonsense theme. These alternatives will depart from the cluster favored by common sense, some in minor ways and some in major ways. They will link some words, or many, to objects or extensions different from those assigned by commonsense intuition. In so doing, they will characterize alternative word-world links, which we might call REFERENCE*, REFERENCE**, REFERENCE***, and so on. And the only obvious complaint against these alternative schemes for nailing words onto the world is that they do not happen to be the schemes sanctioned by our commonsense intuition. Thus the word-world mapping that will be captured by the correct theory of reference will be a highly idiosyncratic one. It will be one member of a large family of word-world mappings, which stands out from the rest only because it happens to be favored by intuition.
The rather arbitrary contours of the intuitively sanctioned reference relation may come into sharper focus if we reflect on the provenance of our semantic intuitions. We have been assuming that our intuitions are largely determined by a set of tacit principles. If this is right, where did those principles come from? The short answer is that no one really knows. But it is a good bet that these principles, like principles underlying our intuitions about grammaticality or morality or politeness, are culturally transmitted and acquired by individuals from the surrounding society with little or no explicit instruction. Though the range of principles people can acquire may be genetically constrained, there is every reason to suppose that other cultures could perfectly well internalize different principles of folk semantics, just as they may internalize different principles of grammar or politeness or morality. Thus, the fact that our intuitions pick out the particular word-world relation that we call "reference" rather than one of the many others in the envelope of genetic possibility is largely the result of historical accident, in very much the same way that details of the grammar of our language or elements of our principles of politeness are in large measure the result of historical accidents.
That's the end of my summary of the argument that convinced me that the intuitive reference relation is idiosyncratic and arbitrary. Suppose the conclusion is correct. What implications would it have for eliminativism? The answer that I once found convincing ran as follows: The central claim of eliminativism is that beliefs and other intentional states do not exist. But that claim is true if and only if predicates like `--- is a belief' refer to nothing. Well, suppose that `--- is a belief' doesn't refer to anything. How interesting a result would this be? Would it be anything to worry about? To think that it would, we must suppose that there is something interesting or important about reference - the intuitively sanctioned word-world mapping. For surely there are lots of alternative word-world mappings on which predicates like `--- is a belief' is related to something ontologically unproblematic. So, even if `--- is a belief' refers to nothing, it may well REFER* (and REFER**, and REFER***) to lots of things. If the difference between reference and REFERENCE* is simply that one of these idiosyncratic mappings happens to have gotten itself embedded in our folk semantics, while the other has not, then it's hard to see why we should care whether or not the extension of `--- is a belief' is empty. If the only thing that distinguishes reference from REFERENCE**, REFERENCE*** and various other word world mappings is an historical accident, then the fact that `--- is a belief' refers to nothing just isn't very interesting or important. But if that isn't interesting, then neither is the fact that beliefs don't exist.
The point can be put in a slightly different way by considering a family of doctrines that are related to eliminativism. On the account we have been working with, eliminativism is true if and only if `--- is a belief' refers to nothing. Let ELIMINATIVISM* be a doctrine that is true if and only if `--- is a belief' REFERS* to nothing; let ELIMINATIVISM** be a doctrine that is true if and only if `--- is a belief' REFERS** to nothing; and so on. Clearly some of these ELIMINATIVISM-stars are bound to be true, while others will be false. Suppose that ELIMINATIVISM***** happens to be one that turns out to be true. Is that anything to worry about? It's hard to see why it should be. But unless there is some interesting and important difference between reference and REFERENCE*****, it is hard to see why eliminativism should be any more worrisome than ELIMINATIVISM*****. If, as I have argued, the only noteworthy difference between reference and various of the REFERENCE-stars is that the former happens to be sanctioned by our intuitions, then eliminativism is neither a troublesome doctrine nor a particularly interesting one.
I confess that I was never entirely comfortable with this argument. The conclusion seemed rather wild, even by my very permissive standards. But since I couldn't see anything clearly wrong with the argument, I decided to write it up, read it around, and see if my colleagues could convince me that I had made a mistake._
Objections to the arguments that I sketched in the last section weren't long in coming. The one that I found most unsettling I heard first from John Searle; it was reinforced, in a slightly less strident tone of voice, by Christopher Gauker. Searle and Gauker focused on the second argument, the one that starts with the claim that the intuitive reference relation is an idiosyncratic relation, and concludes that eliminativism is an uninteresting doctrine not worth worrying about. What Searle and Gauker pointed out was that this argument seems to be generalizable in a way that leads to lots of strikingly implausible conclusions. As I set it out, the argument is aimed at propositional-attitude-eliminativism, the claim that beliefs and desires don't exist. But consider some claims that have been hotly debated in other intellectual domains. Most physicists, I gather, accept the claim that the Big Bang occurred, and the claim that Black Holes exist, though there are a few who deny one or the other of these. Those who deny them advocate ontological views that are analogous to eliminativism; indeed, we might well label their views "Big-Bang-eliminativism" and "Black-Hole-eliminativism". Now, as Searle and Gauker noted, for each of these eliminativist doctrines we can construct an argument entirely analogous to the argument which allegedly showed that propositional-attitude-eliminativism is nothing to worry about. For just as the REFERENCE-star relations (REFERENCE*, REFERENCE**, etc.) can be used to formulate a family of claims parallel to propositional-attitude-eliminativism, so too, they can be used to formulate a family of claims parallel to Black-Hole-eliminativism:
Black-Hole-eliminativism is true if and only if `--- is a Black Hole' refers to nothing.
Black-Hole-eliminativism* is true if and only if `--- is a Black Hole' REFERS* to nothing.
Black-Hole-eliminativism** is true if and only if `--- is a Black Hole' REFERS** to nothing.
and so on. According to the argument that Searle and Gauker were criticizing, there is nothing special about the reference relation. It's just an historical accident that our intuitions happen to favor reference rather than one of the equally idiosyncratic REFERENCE-stars. If that's sufficient to show that propositional-attitude-eliminativism isn't interesting or worth worrying about, then it looks like we should draw exactly the same conclusion about Black-Hole-eliminativism (and about Big-Bang-eliminativism, and while we're at it, we might as well throw in God-eliminativism, also known as atheism). But this, Searle and Gauker insisted, is just mad. If it really is the case that Black Holes don't exist, or that the Big Bang didn't occur, it would be enormously interesting, and would pose a major challenge to contemporary physics. Of course it is worth worrying about!
In most areas of intellectual activity, there is a fair amount of disagreement about how puzzling or implausible the consequences of a view must be in order to constitute a reductio ad absurdum of the view. And in philosophy, more than in most other disciplines, one person's modus ponens is another person's modus tolens. As will come as no surprise, I tend to be on the ponens end of the spectrum. If there is a plausible argument in favor of a doctrine, it requires a substantial collection of implausible consequences before I'm convinced that the doctrine producing those consequences should be rejected. But the Searle / Gauker argument pushed me passed my limit. Clearly something had gone very wrong. This conviction that there must be some major mistake in the way I was thinking about these issues was reinforced by another objection to my views that I had heard from Frank Jackson.
As I interpreted it, Jackson's objection focused on the first of the two arguments developed in the previous section. That argument starts with the claim that our intuitions about the reference of terms embedded in theories that we now take to be mistaken may simply be indeterminate, because the "folk semantic" theory that underlies our intuitions has nothing to say about such cases. From this, it concludes that if folk psychology is indeed a mistaken theory, then the eliminativists' ontological claims are also indeterminate. Claims like: "There are no such things as beliefs." are neither true nor false. What Jackson noted is that there is something rather odd about the way in which our semantic intuitions get to play a central role in determining whether or not beliefs exist. For, Jackson noted, if beliefs do in fact exist, then no doubt they have been around for a very long time -- probably far longer than people have had semantic intuitions. To drive home the point, Jackson also suggested an analogy between propositional-attitude-eliminativism and Big-Bang-eliminativism. Suppose it turns out that our current theory about the Big Bang is wrong in various ways. If so, then by an argument parallel to the first argument in the previous section, it seems to follow that if our semantic intuitions remain silent on such cases, then whether or not the Big Bang occurred is indeterminate. If, on the other hand, our semantic intuitions are determinate about cases like this, then the Big Bang really did occur, or (if the intuitions go the other way) it really did not. But this, Jackson protested, is a very strange conclusion. For if the Big Bang did occur, it occurred billions of years before there was anything around that had semantic intuitions. So any view that entails that our semantic intuitions play a central role in settling whether or not the Big Bang occurred is at least a bit paradoxical.
There are various moves that might be made in response to the argument I'm attributing to Jackson, and thus I don't think the argument is as compelling as the one offered by Searle and Gauker._ One of the reasons that I found Jackson's argument particularly interesting, however, is that it can be generalized in a way that challenges those whose views on what a theory of reference is supposed to do follows the proto-science line that I sketched in Section 6.2. On that view, it will be recalled, our semantic intuitions and the folk semantic theory that underlies them do not play a central role in determining the correct account of reference. Rather, the reference relation is whatever reference-like word-world relation that turns out to play an important explanatory role in linguistics (or perhaps in some other area of science). Suppose this is right. How would it connect with the eliminativism debate? Well, eliminativism as we have been construing it is the view that beliefs (and other propositional attitudes) don't exist. And that's true if and only if `--- is a belief' refers to nothing. Now let's suppose, as we have been throughout, that terms like `--- is a belief' are embedded in an opulent tacit theory ("folk psychology") and that this theory turns out to be mistaken in lots of ways. Do beliefs exist? It looks like the answer is that we don't know yet. For our colleagues in the linguistics department have not yet settled what reference is. They haven't determined which word-world relation it is that is going to prove useful in their explanatory endeavors. Nor has there been any resolution of the matter in other parts of the university. Cognitive psychologists, anthropologists, historians of science and others who may invoke the reference relation are in no better position to set out the details of the word-world relation that will facilitate their inquiries. Optimists may think that linguists or others have at least a rough and ready story to tell about the reference relation. But even if that's right (and I rather doubt it) they haven't settled on enough of the details to say when or whether terms embedded in false theories succeed in referring. Until they do, the findings of psychologists, neuroscientists, metaphysicians and others who (allegedly) have shown that folk psychology is seriously mistaken won't be enough to settle whether `--- is a belief' refers. And thus we won't know whether or not beliefs exist.
Now at this point one can well imagine someone who has heard Jackson's argument protesting that something has gone disastrously wrong here. Linguistics, particularly that part of the discipline that deals with reference, is a tiny twig on the tree of science. Yet if the line of thought in the previous paragraph is correct, then this twig on the tree of science (or perhaps some other twig) gets to play a fundamental role in determining whether beliefs exist. Moreover, the oddness generalizes just as it did in Jackson's argument. For it is entirely possible, indeed quite likely, that at some point in the future physicists will become convinced that some important claims in the currently accepted theory about the Big Bang are mistaken. Some of these physicists may begin thinking that perhaps the Big Bang did not occur at all. How should they settle the matter? Well, if we adopt the strategy set out in the previous paragraph, it looks like they should take their best available theory, troop down to the Linguistics Department, and wait patiently until the linguists have reached some firmer view on what reference really is. Needless to say, this is a hopelessly implausible scenario. The scenario is even more peculiar if, as is entirely possible, it turns out that different disciplines find different word-world relations to be most useful to their projects. If linguistics explicates reference in one way, and anthropology (or sociology of science, or ...) explicates it in a different way, then it may turn out that the physicists in our little story will get conflicting advice. If they listen to the anthropologists, they will have to conclude that the Big Bang did occur, while if they listen to the linguists, they will have to conclude that it didn't. Once again, it looks like we've ended up with a singularly implausible conclusion.
The upshot of all of this is that when plugged into the overall account of the eliminativism debate that I have been working with, both of the accounts of what a theory of reference is supposed to do that were developed in Section 6 lead to consequences that strain credulity (at least my credulity) to the breaking point. Having reached this point in my own thinking, I could no longer suppress the conclusion that there was a quite catastrophic mistake lurking somewhere in the views I had been defending. Unfortunately, however, while both the Searle / Gauker argument and the Jackson argument make it clear that something has gone wrong, neither the arguments nor the philosophers who suggested them could offer a convincing diagnosis of where exactly the mistake had occurred. Nor, at first, could I. After a while, however, I began to suspect that the real source of my difficulties were to be found in a place I'd never thought of looking.
As I now see it, the crucial mistake in the line of reasoning that I've been recounting in this essay, the misstep that ultimately led to the quite preposterous consequences encountered in the previous section, was one that was taken very early on. It was the step in which I proposed that questions about the existence of entities posited by false theories could be productively addressed by focusing on what the theory of reference tells us about the terms used in such theories. This strategy of trading substantive scientific or metaphysical questions concerning the nature or the existence of entities (questions in the "material mode" as Carnap used to say) for apparently equivalent semantic questions concerning the terms we use in talking about those entities (Carnap's "formal mode" questions) is sometimes called the strategy of semantic ascent._ Resorting to semantic ascent was the crucial error that ultimately led to disaster. More accurately, it was half of the crucial error. The other half was to pay insufficient attention to the distinction between deflationary and non-deflationary accounts of reference. On my current view, semantic ascent and appeal to the theory of reference can be of no help at all in addressing ontological issues. Since eliminativism is an ontological doctrine, appeal to the theory of reference can play no role in determining whether the doctrine is true.
Obviously this new view of mine will require some unpacking. Let me start by focusing on the strategy of semantic ascent. The general form of the principle that justifies semantic ascent (for reference -- there's an analogous story to be told about truth) might be put roughly as follows:
(1) (x) Px iff `P__' refers to (or is satisfied by) x.
Less formally, what (1) says is that an entity is a P if and only if the predicate `P__' refers to it. (1) can be viewed as an axiom schema whose instances include claims like:
(2) (x) x is a Black Hole iff `Black Hole' refers to x
(3) (x) x is a belief iff `belief' refers to x.
Obviously, (2) and
(4) ~ (Ex) `Black Hole' refers to x
(5) ~ (Ex) x is a Black Hole.
Similarly, (3) along with
(6) ~ (Ex) `belief' refers to x
(7) ~ (Ex) x is a belief.
So if we have a theory of reference which tells us that terms in seriously mistaken theories do not refer, and if we are prepared to grant that `belief' is such a term, then these claims along with (3) will entail that beliefs don't exist. And that is pretty much the way my ill-fated line of reasoning began.
But what about (3) and the other instances of (1). What justification do we have for them? No justification was offered on the many occasions when I've invoked (or presupposed) instances of (1) in my arguments, either in this Chapter or elsewhere. I've always simply assumed that instances of (1) are so obvious that they need no justification; indeed, they hardly need to be stated. But are they? As I now see it, the answer turns on whether or not the notion of reference being invoked is a deflationary notion. If it is, then instances of (1) are indeed trivial. But if reference is understood as a deflationary notion, then instances of (1) are also quite useless for our purposes. They provide no help at all in determining whether theoretical terms embedded in a seriously mistaken theory refer to anything. If, on the other hand, reference is taken to be a non-deflationary notion, then instances of (1) may well be helpful in settling whether or not terms in a mistaken theory refer. But on non-deflationary accounts of reference, (1) and it's instances are themselves problematic. If they are true, they are not obviously true. They aren't premises we get for free. So if we are going to invoke them in arguments for or against eliminativism, we are going to need an argument that supports them. And I, for one, haven't a clue about what the argument might be.
In defending this cluster of claims, I'll begin by elaborating a bit on what I mean by "deflationary" accounts of reference. These accounts, which are modeled on deflationary accounts of truth, take terms like `refers' `denotes' and `designates' to be quasi-logical devices that earn their keep by facilitating semantic ascent and descent. On these accounts, the schema (1) or something like it captures the entire meaning of `refers'. It tells us all there is to tell about the nature of the reference relation. So on these accounts (3) and its kin really are trivial and obvious; indeed on some versions of the deflationist story, they are analytic._ Thus, if `refers' is read in a deflationary way, then if we can establish that `belief' refers to nothing (= (6)), we can infer that there are no beliefs. But of course, on a deflationary story, reference is not determined by a complex causal chain, or by any other empirically investigatable "naturalistic" relation. So we can't attempt to determine whether `belief' refers by exploring whether anything lies at the other end of an appropriate reference fixing causal chain. Rather, (3) itself provides us with our only way of finding out whether `belief' refers to anything. The answer is yes if and only if something is a belief. But now what shall we say about brain states (or functional states, or dispositional states, or whatever else you might fancy) that have some, but not all of the features attributed to beliefs by folk psychology? Does `belief' refer to them? The only answer we get from deflationary accounts of reference is that it does if and only if they are beliefs. Obviously, there is no hope of making progress here. We are tugging at our own bootstraps.
Let's turn, now, to non-deflationary accounts of reference. For our purposes, any account of reference that isn't deflationary will be "non-deflationary." This includes causal-historical accounts, description accounts, and perhaps others as well. Some of these accounts seem to offer a promising way of breaking out of the circle sketched in the previous paragraph. If reference is some naturalistic relation then we can try to use the techniques of science and history to determine whether or not a problematic term refers to something. But it's my contention that on non-deflationary accounts we need some argument for accepting (3) and other instances of (1). We don't get them for free.
Perhaps the easiest way to make the point is to focus on causal-historical accounts of reference, though everything I say in the next few paragraphs could be recast, mutatis mutandis, if we were to focus instead on description based accounts. As we saw in Section 7, given any detailed specification of a causal-historical reference relation -- a relation mapping words on to entities and classes or kinds in the world -- it is an easy matter to construct alternative relations that map words to the world in a different way. By fiddling with the specification of what counts as an acceptable grounding, or with the specification of what counts as a reference preserving transmission, or by making various other modifications, we can generate an array of word-world mapping relations (REFERENCE*, REFERENCE**, etc.) that differ from the specified relation. Some of the alternatives may depart from the specified relation in only one case, some in a few cases, and still others may depart massively. Let us call the causal-historical relation that, according to our favorite theory, determines the actual reference relation "C-H-link-r". Also, let's use "C-H-link-R*" as a label for the causal-historical relation that determines REFERENCE*, "C-H-link-R**" as a label for the relation that determines REFERENCE**, etc. Using this terminology, (1) is equivalent to:
(C-H 1) (x) Px iff C-H-link-r (`P__', x).
Less formally, what this says is that an entity is a P if and only if it stands in the C-H-link-r relation to the predicate `P__'. But, of course, for each of the alternative word-world mappings, there is a schema analogous to (C-H 1):
(C-H 1*) (x) Px iff C-H-link-R* (`P__', x).
(C-H 1**) (x) Px iff C-H-link-R** (`P__', x).
Clearly, it can't be the case that all of these schemata are correct. Since C-H-link-r and C-H-link-R* are different relations, there is at least one predicate, `P__', and one entity, x, such that if `P__' stands in the C-H-link-r relation to x, then `P__' does not stand in the C-H-link-R* relation to x, or vice versa. Why, then, should we suppose that (C-H 1) is correct, and that all the others are mistaken? Surely it is not obvious that, for example, all and only pigs stand in the C-H-link-r relation to the predicate `___ is a pig'; nor is it obvious that all and only Black Holes stand in the C-H-link-r relation to the predicate `___ is a Black Hole'. Let me stress that I am not claiming that these instances of (C-H 1) aren't true. It is, I suppose entirely possible that they are. My claim is simply that we need an argument to show us that they are true. Contrary to what I once assumed, these claims are far from trivial.
Presumably a plausible argument aimed at establishing the truth of instances of (C-H 1) would rely heavily on one or the other account, sketched in Section 6, of what it is to get a theory of reference right. For it is these accounts that show us how to justify the claim that C-H-link-r is to be identified with reference. Those who favor the account in Section 6.1 will try to show how the fact that C-H-link-r is sanctioned by intuition (or folk semantics) can be used to establish that all instances of (C-H 1) must be true. While those who favor the account in Section 6.2 will try to show how the fact that C-H-link-r is central to the explanatory purposes of linguistics (or anthropology or cognitive psychology) guarantees that all instances of (C-H 1) are true. Unfortunately, I know of no argument along these lines that is even remotely plausible. And without such an argument, the attempt to use semantic ascent to determine whether or not the entities invoked in false theories exist grinds to a halt.
During a seminar on these matters, Hartry Field suggested what might seem to be a tempting strategy for dealing with the difficulty. The basic idea is to make the semantic ascent principle_ a constraint on any acceptable non-deflationary account of reference. Thus, for example, a theorist who accepts the account in 6.1 might say that the right word-world relation (the one that really is reference) is the one that meets a pair of constraints. First, it must be a relation that does a good job at capturing our intuitions about a wide range of cases, and second, it must satisfy the semantic ascent principle. Any relation which doesn't satisfy that principle just will not count as the reference relation. The imposition of this constraint might itself be justified by appeal to intuition, since (1) is arguably the most intuitively obvious fact about reference. Theorists who accept the account in 6.2 can make much the same move. On their view it is the sciences (linguistics or anthropology or what have you) that get to say which word-world relation reference is. But there is a constraint on their endeavors. Whatever relation they come up with must satisfy the semantic ascent principle. If it doesn't, then it isn't the reference relation. This move has the added attraction of ending the worry that different sciences might find different word-world relations important in their explanatory pursuits. This could still happen, of course. Linguistics might characterize its favored word-world relation in one way, and anthropology could characterize it's favored relation in a very different way. However, the relations characterized will not count as reference unless they both satisfy the semantic ascent principle. And if they both satisfy the principle, then, since relations are individuated extensionally, they are the same relation. The relation may be described or picked out very differently in different sciences. But if the relations favored in different sciences all satisfy (1), then for any given term the favored word-world relations must all pick out exactly the same referents. Problem solved!
Well, not quite. For as Field also pointed out, the strategy of building in semantic ascent as a constraint on reference really just hides the problem, it doesn't solve it. Perhaps the easiest way to see the point is to consider a situations in which a pair of theorists agree about the Premises of the eliminativists' argument but disagree about the conclusion. Theorist A grants that folk psychology is mistaken in various ways, and thus that there is nothing in the world that has all of the properties that folk psychology attributes to beliefs. But Theorist A does not think that this shows that beliefs don't exist. For Theorist A also has a theory of reference which specifies a relation, R (for concreteness, we can imagine it is a causal-historical relation of the sort championed by Lycan). In addition, Theorist A has a pretty persuasive argument that the predicate `--- is a belief' stands in the R relation to certain neurophysiological states (or functional states, or whatever). "So," he argues, "we can conclude that beliefs do exist, since `belief' stands in the R relation to these neurophysiological states, and R is the reference relation, and the reference relation satisfies the principle of semantic ascent. Q.E.D." Theorist B, an eliminativist, is not persuaded. "Look," she replies, "all you've done is establish that R is not the reference relation. It can't be, because, as you have shown, `belief' stands in the R relation to these neurophysiological states, and these neurophysiological states aren't beliefs. Thus your R relation fails to satisfy semantic ascent. So it's not reference. Q.E.D. to you too!" Obviously this dispute isn't going to get settled until we determine whether or not the neurophysiological states are beliefs. And appeal to the theory of reference is going to be of exactly no help in that since we can't tell whether the proposed reference relation satisfies the semantic ascent constraint unless we already know whether the neurophysiological states are beliefs. I think the moral to be drawn from this little tale is clear. If satisfying semantic ascent is proposed as a requirement that any acceptable account of reference must meet, then appeal to the theory of reference can be of no help in resolving contested ontological questions. For without some independent way of settling these contested ontological questions, we have no way of knowing whether the principle of semantic ascent is satisfied. And if we have some independent way to resolve contested ontological questions, we don't need the theory of reference._
In my own earlier work, and in the work of most other authors who were (or still are) seriously tempted by eliminativism, the strategy of semantic ascent and an explicit or implicit appeal to description based theories of reference played a central role in getting from the Premises of the eliminativists' argument to the conclusions. A central goal of the last six Sections has been to explain why I now think this route is singularly unpromising. My project of deconstructing the eliminativists' deconstruction is not yet complete, however. For in some of the literature on eliminativism there are hints of a very different strategy for filling the gap between the eliminativists' Premises and their conclusions. Rather than relying on a theory of reference, this alternative strategy invokes the notion of "constitutive" or "conceptually necessary" properties. The central idea is that some of our concepts require, as a matter of logical or conceptual necessity, that any object to which the concept applies must have certain properties. We would not apply the concept to an object, or count the object as falling within the category that the concept specifies (or, at least, we ought not to do so) unless the object has the constitutive properties. Thus, for example, it might be urged that being unmarried and being male are constitutive properties for the concept of a bachelor, or that having a negative charge is constitutive for the concept of an electron. So if something is not male and unmarried, then it can't be a bachelor, and if something does not have a negative charge, then it can't be an electron. Both of these would be "conceptually impossible."_
It's easy to see how the notion of a constitutive property might be used to fill the gap between the Premises and the conclusions in the eliminativists' argument. If it can be shown that a certain property is conceptually necessary for having beliefs, or for having propositional attitudes in general, and if science (or philosophical argument) can demonstrate that no one has that property, it follows that no one has beliefs or propositional attitudes. However, it's my view that this strategy is even less promising than the one that relies on description theories of reference. My skepticism is based on a pair of distinct though related considerations.
The first is that it is far from obvious how a theorist who wants to invoke this strategy could establish that one or another property is indeed conceptually necessary for something to count as a belief, or for someone to count as having beliefs. Obviously, the mere fact that lots of people think, or claim or presuppose that all beliefs or believers have a certain property is not enough to show that the property in question is conceptually necessary. Nor would it be sufficient to show that many people would refuse to apply the term "belief" to a state or organism that lacked the property. For it might simply be the case that most people (or indeed all people) happen to have some strongly held opinions about beliefs, and that these opinions are false. There was, after all, a time at which most people would have refused to apply the term "star" to an object that did not have the property of rotating around the earth. But as we now know, they had some deeply entrenched false opinions about stars and about the earth, and it was these opinions rather than any conceptual impossibility that was responsible for their refusal.
Actually, if we look at the literature in which the notion of conceptually necessary properties is invoked, there is no serious effort to establish what most people claim about beliefs, nor is there any evidence offered about the percentage of speakers who are willing or reluctant to apply the term `belief' to particular cases. What we find instead are just assertions that one or another property strikes the author as essential or constitutive. Here is a typical example from Andy Clark:
It is plausible to require that any being who can be said to grasp a concept C must be capable of judging that she has made a mistake in some previous application of C.... It seems to me conceptually impossible for a being to count as grasping a concept and yet be incapable of ever having any conscious experience involving it. That is, part of what we mean when we say that someone grasps the concept `dog' is that, on occasion, the person has conscious mental experiences which involve that very concept.... If I am right, the very idea of a True Believer thus builds in two demands (consciousness and the ability to issue genuine judgements about its own past performance) which scientific investigations might reveal not be met in specific cases._
In another publication, where he offers a rather different account of what is conceptually necessary for having a thought or a belief, Clark launches his defense of his proposal as follows: "It remains to pump our intuitions. Here are mine."_
Now the striking thing about these claims and the intuitions that putatively support them is how little agreement there is among the theorists offering them. Some writers maintain that "rationality" and the potential for "logical combination" are essential._ Others deny it._ Some insist that "causal efficacy" or having "causal powers" is essential._ Others insist that it is not._ Some writers agree with Clark's consciousness requirement, though many others find it to be intuitively implausible. And I have yet to find anyone whose intuitions match Clark's in requiring "the ability to issue genuine judgements about its own past performance," since, as Clark readily acknowledges, it is almost certainly the case that household pets and other non-human animals fail to satisfy this requirement. Thus the condition would rule that it is "conceptually impossible" for the family dog to have beliefs!
Since appeal to intuition seems to be the preferred methodological option for those attempting to justify the claim that one or another property is constitutive or conceptually necessary for belief, the existence of widely divergent intuitions obviously poses a major obstacle for anyone who hopes to use the strategy we're considering for linking the Premises of the eliminativists' argument with the conclusions. But I am inclined to think that this practical problem is actually just a symptom of a much deeper theoretical problem. In the current philosophical environment, the very existence of properties that are constitutive or conceptually necessary for the application of one or another concept is hardly an assumption that can be taken for granted. For if some properties are constitutive for a concept, and others are not, and if, as Clark and others suggest,_ claims about concepts are interchangeable with claims about what our words mean, then it seems there must also be some sentences that are true entirely in virtue of meaning, and others whose truth or falsity depends in part on the way the world is. That is, there must be some sentences that are analytic and others that are synthetic. If, for example, being unmarried is constitutive or conceptually necessary for being a bachelor, then presumably "All bachelors are unmarried" is analytic. And if having causal potency is conceptually necessary for being a belief, then "All beliefs are causally potent" is analytic. Starting more than forty years ago, however, Quine and others offered some enormously influential arguments aimed at showing that the analytic / synthetic distinction is untenable._ On Quine's view, and on the view of many other philosophers as well, there are no sentences that are true solely in virtue of their meaning. If this is right, then there are no constitutive or conceptually necessary properties. So it is hardly surprising that those who think that there are conceptually necessary conditions for being a belief or a believer have trouble agreeing about what they are.
I don't propose to review the arguments against the existence of the analytic / synthetic in this essay. Indeed, for current purposes I don't even need to assume that the conclusion of those arguments is correct, though as it happens I think it is. All that is needed here is the observation that the very existence of analytic truths, and thus of constitutive or conceptually necessary properties, is hotly disputed and highly problematic. So those who want to fill the gap in the eliminativists' argument by invoking the notion of constitutive properties owe us some further argument. They must either make it plausible that the arguments against the existence of the analytic / synthetic distinction are mistaken, or that the notion of a conceptually necessary property can be made sense of without presupposing or entailing the existence of analytic sentences. Having done that, they must go on to defend specific claims about the properties that are conceptually necessary for beliefs, and (if it turns out that intuitions are relevant -- it's far from obvious that they are) they must explain away the intuitions of those whose intuitions run counter to these claims. Obviously accomplishing all of this would be no easy task. If there are philosophers who chose to follow this path in defending eliminativism, I certainly wish them well. But I don't propose to hold my breath until they succeed._
From the eliminativists' point of view, the results that we've reached so far have been entirely negative. Even if we grant the most promising version of the eliminativists' Premises, I've argued, the conclusions don't follow. Nor is it likely that we can fill the gap by appealing to a theory of reference or by invoking the notion of a constitutive property. But, of course, the fact that the eliminativists' conclusions don't follow from their Premises doesn't indicate that those conclusions are wrong. Moreover, it hardly seems plausible that the eliminativists' Premises are totally irrelevant to claims about the existence or non-existence of beliefs and other propositional attitudes. For surely if it is the case that our folk theory about the propositional attitudes is seriously mistaken, that is something that should be taken into account in determining whether propositional attitudes exist. But how? This is, of course, just a special case of a much more general question that has now moved to center stage in our exercise in deconstruction: How are we to go about deciding whether or not the entities posited by any false theory exist? If we can't turn to the theory of reference or exploit claims about constitutive properties, how are these ontological issues to be settled?
An answer that might seem attractive at this point is suggested by a strategy that has been much discussed in the philosophy of science in recent years. What we need is some way to determine whether or not it would be rational to conclude that a given type of entity exists, in the face of various sorts of findings about the shortcomings of the theories in which entities of this type play a role. One way to proceed would be to see if we can extract some normative principles -- principles of rational ontological inference or decision making -- from a study of actual historical cases in which these sorts of decisions have already been made.
Here's how this project might unfold. We begin by assembling a collection of historical cases in which it was ultimately decided that the posits of mistaken theories did not exist, and another collection of historical cases in which it was decided that the posits of mistaken theories did exist, despite the mistaken claims that the theories had made about them. Caloric, phlogiston, Vulcan and witches might be plausible candidates for the first set; stars, atoms, planets and brains might be candidates for the second. With these two sets of cases in hand, we can begin to look for salient similarities and differences. Are there apparently relevant features of the defective theory, or of the problematic posit, or of the relation of the theory to other theories, that are present in most or all the cases in one set, and absent in most or all the cases in the other? If the answer is yes, we can use these features to formulate candidate principles specifying when it is rationally appropriate to retain items in the ontology of a mistaken theory, and when it is appropriate to "eliminate" them. The tentative principles can then be tested against additional historical cases that were not part of the original data base. If the principles endorse the decisions that were actually made in these additional cases, this will count in favor of the principles. While if the principles endorse the opposite decision in one or more of these cases, it will indicate that we should make some adjustments in the principles and try again. Once we have formulated a reasonably robust set of principles that handles lots of the clear historical cases, we could set about testing the principles against our intuitions about a range of hypothetical cases, provided of course that we have such intuitions and are inclined to trust them. In pursuing this strategy, we need not always modify our principles when they come into conflict with an actual historical case. If a given set of principles for rational ontological decision making does a good job at capturing most actual cases, we may decide that in a few historical cases people made the wrong decision; their ontological inference was not a rational one. Similarly, we need not always modify our principles when they come into conflict with a firmly held intuition about a hypothetical case. Rather, we may decide that our intuition in that case is misguided.
The strategy set out in the previous paragraph is something of a hybrid. The part of the story that relies on intuitions resembles the "reflective equilibrium" account of the justification of inferential principles that was given it's most famous formulation by Nelson Goodman._ The part that attends to actual historical cases is modeled on the strategies for uncovering normative principles of scientific reasoning and inquiry that are sometimes labeled "normative naturalism."_ Obviously there is room for many variations on the pattern that I've sketched, some stressing historical cases and down playing the role of intuition, others going in just the opposite direction._ Some researchers have used a normative naturalist strategy to test discursively formulated methodological principles,_ while others, most notably Herbert Simon and his colleagues, have attempted to produce computer simulations of the reasoning processes that have led to important discoveries in the history of science._ Also, as Simon and his co-workers have urged, there are ways to supplement the largely historical approach to locating principles of scientific reasoning with experimental studies in which scientists are asked to solve unfamiliar problems._
There has been relatively little work along normative naturalist lines aimed explicitly at uncovering principles of rational ontological inference that might be useful in assessing the eliminativists' argument. The most interesting study I know of in this area is Robert McCauley's exploration of the way in which tensions between theories at the same and at different "levels of analysis" lead to theoretical and ontological elimination._ McCauley offers no detailed definition of "level of analysis," though the intuitive idea is familiar enough. "Broadly speaking, chemistry is a higher level of analysis than subatomic particle physics, .... biology is an even higher level, and psychology higher than that."_ Building on the work of Wimsatt,_ McCauley reviews various episodes in the history of science and offers an interesting and provocative conclusion. When a new theory comes into conflict with an older one at the same level of analysis, and when the new theory is better than the old one, "the superior theory eliminates its competitor (and its ontology)."_ However, when theories at different levels of analysis come into conflict, the story is quite different. "The history of science reveals no precedent for theory replacement or elimination in interlevel contexts."_ On McCauley's view, this finding has clear implications for the version of the eliminativist argument offered by Paul Churchland. Both common sense psychology and cognitive psychology "operate at different levels of analysis" from neuroscience._ Thus, even if it does turn out, as Churchland insists it will, that neuroscience and common sense psychology are "incommensurable," it would, McCauley maintains, be incorrect to "conclude ... that such incommensurability requires the elimination of one or the other."_
I am not entirely comfortable with McCauley's analysis of the historical record, since he is often less careful than one might wish about distinguishing between the "elimination" of theories and the "elimination" of their ontology. But that's not a worry I propose to pursue here. For current purposes, what's important about McCauley's work is that it illustrates the way in which a normative naturalist might begin to make progress in determining what conclusion it would be rational to draw from the eliminativists' Premises. If McCauley is right, and if the eliminativist defends the Second Premise by arguing for a "radical incommensurability"_ or "logical inconsistency"_ between neuroscience and folk psychology, no elimination of the ontology of folk psychology is rationally warranted. This is, of course, only one among many ways in which eliminativists might try to defend the Second Premise. So there's lots more work to be done. Still, one might hope that for each sort of defense that eliminativists offer for the Second Premise, normative naturalists could determine whether the eliminativists' argument, fleshed out in that way, rationally supports the ontological conclusion that eliminativists wish to draw._
The normative naturalist strategy that I've been recounting is certainly an intriguing one, and perhaps it will ultimately succeed. But for two rather different reasons, I am inclined to be more than a bit skeptical. The first reason for my skepticism turns on the normative status of the principles of reasoning (or inference or inquiry) that the normative naturalist strategy sanctions. If normative naturalists are right, then the principles produced by the process they recommend are rational principles; the conclusions to which the principles point are conclusions we ought to accept, provided of course that we accept the premises. But why, exactly, ought we to accept them? What gives these principles their normative force? This is a complex issue on which there has been some vigorous debate in recent years._ Since I have staked out my own position elsewhere,_ I'll restrict myself to two brief observations. First, it is a mistake to ask why the normative naturalist strategy yields principles of rational inference, since there isn't one normative naturalist strategy, there are many whose relation to one another is best viewed as a loose family resemblance. And while it may be the case that all of these strategies will converge in sanctioning some principles, it is very likely indeed that there will be other principles on whose normative status they will disagree. So a serious defense of the normative naturalist strategy must say why one specific member of this family of strategies is the right one -- the one that can be counted upon to produce principles of reasoning that we ought to accept. And that, needless to say, is no easy task. My second observation is that some attempts to defend the normative credentials of normative naturalist procedures (including Laudan's and my own) have a decidedly relativist flavor. The principles sanctioned by normative naturalism are claimed to be hypothetical imperatives. They are principles that we ought to follow if we have certain goals. If our goals are different, the normative naturalist procedure may well sanction different principles. However, in many intellectual controversies, including the debate over eliminativism, it is far from clear that all parties to the dispute share the same goals. And if they don't, then normative naturalism may recommend different conclusions to different people. I am inclined to think that these two observations justify at least a bit of skepticism about the normative naturalist approach in general.
My second reason for being skeptical about the normative naturalist strategy focuses more sharply on what we can expect normative naturalism to tell us about patterns of ontological inference, including the sort of inference that could get us from the eliminativists' Premises to their conclusions. Even if we grant that the output of the normative naturalist procedure will be principles that rational people ought to follow, it is important to realize that there is no guarantee that this approach will succeed in resolving our problem about what to conclude from the eliminativists' Premises. For it is entirely possible that there simply are no normative principles of ontological reasoning to be found, or at least none that are strong enough and comprehensive enough to specify what we should conclude if the Premises of the eliminativists' arguments are true. We might well find that in many cases in which ontological inferences have been made there are no features of the theories involved, the sorts of errors they make, or the properties attributed to the items in their ontology that are systematically correlated with the ontological conclusions that have been accepted in the history of science. We might also find that facts about the defective theory and its problematic posits do not suffice to generate clear and intersubjectively consistent intuitions about how the ontological decision ought to go. In those historical cases in which widely accepted ontological conclusions were drawn, it is reasonable to suppose that there must be some explanation for which way the decision went. And if, in some of those cases (or many), normative principles remain silent, then other factors must have been involved. But these additional factors may be significantly different in different cases. Moreover, in some cases it might turn out that the outcome was heavily influenced by the personalities of the people involved, or by social and political factors in the relevant scientific community or in the wider society in which the scientific community is embedded. If this is right, then the normative naturalist strategy may well fail to produce principles of reasoning that tell us what ontological conclusion to draw about entities invoked in mistaken theories.
While I don't pretend to have enough evidence to mount a conclusive case, it is my bet that this outcome is exactly what we would find if we launched a serious and systematic study of the relevant historical cases. Here is a brief and admittedly speculative list of some of the kinds of cases that I suspect a careful scrutiny of actual examples will uncover:
i) Don't care's:
In some cases it just doesn't much matter to anyone whether theorists conclude that some of the entities invoked in a mistaken theory really do exist, though their properties are quite different from those attributed to them by the old theory, or whether they conclude that there are no such things -- that some of the theoretical terms in the old theory didn't refer to anything -- and that the phenomena that need explaining are best explained by invoking different theoretical entities. In these "don't care" cases the decision will be made more or less arbitrarily. This may well be what happened with electrons which were, according to Fine, originally conceived of not as particles but as units of quantity of electric charge._ It may also be what is happening now with genes. Modern molecular genetics recognizes nothing that has all of the important properties attributed to genes by Mendel, De Vries, T. H. Morgan, Muller, and the other pioneers of modern genetics. Should we conclude that there are no such things as genes? Or should we conclude that genes do exist, though Mendel and others were seriously mistaken in many of the claims they made about them? There has, it seems, been no firm decision reached in the relevant scientific communities. Eventually, one suspects, the matter will be resolved, though it may be that little hangs on the issue, and no one much cares which way the decision ultimately goes._
ii) Implicit previous agreements:
In some cases there may be an implicit agreement in the relevant scientific community that some property or set of properties are the essential ones for some posited entity, and that if it turns out that nothing has those properties then everyone in the community would agree that the entity doesn't exist. I used to think that the history of the aether might well be an example of this sort of case. What was important about the aether, I thought, was that it was uniform and stationary, and thus could provide "the fundamental frame for inertial systems."_ So when Michelson and Morley showed that this conception of the aether was untenable, everyone would agree that there is no aether. It appears, however, that the historical situation was actually considerably more complicated, and that some theorists advocated retaining the aether, even though they agreed that it was "undetectable in principle."_ So perhaps this case is not a good example of a widely shared implicit agreement.
iii) Social and political factors internal to the relevant science:
According to Kim Sterelny, Elliott Sober once conjectured that had Lavoisier wished to be viewed less as a radical innovator and more as a conservative, he might have retained a venerable old term rather than introducing a new one. And rather than maintaining that there is no such thing as phlogiston, he might instead have claimed that Priestly, Stahl and earlier theorists were simply mistaken about lots of the properties that they attributed to phlogiston. So if Lavoisier had a somewhat different personality, what we now call `oxygen' would be called `phlogiston' instead._ I have no idea whether Lavoisier's temperament and the pragmatic or political consequences of being thought to be radical or conservative played any significant role in this case. (I've checked with Sober, and he doesn't have any evidence bearing on the case; indeed, he can't recall making the comment.) But it is my guess that personalities and the micro-politics of scientific communities often play an important role in situations like this._ In some situations it is easier to get a grant or a promotion or to enhance one's reputation in the scientific community by announcing the discovery of a new entity or denying the existence of one previously claimed to exist. In other situations, it is more politically expedient to conclude that entities of a certain sort don't have some of the properties previously attributed to them, and that experimental results or other phenomena can best be explained by attributing some rather different properties to those entities. Which conclusion the scientific community ultimately accepts may well be determined, in some cases, by factors like these.
iv) Broader social and political factors:
During the last decade or two a growing number of serious and well educated people have claimed that witches really did exist at the time of the Inquisition, and that they still do. However, they also maintain that witches make no pact with the devil, cast no evil spells and do not practice black magic (or ride on broomsticks!). These allegations are myths, misunderstandings or slanders spread about witches by people who know little about witchcraft, and also, less innocently, by religious and political enemies. Contemporary people who take themselves to be witches claim that they, like witches in earlier centuries, are practitioners of
an ancient nature-religion ... in which the earth [is] worshipped as a woman under different names and guises throughout the inhabited world.... [Witches] try to create for themselves the tone and feeling of an earlier humanity, worshipping a nature they understand as vital, powerful and mysterious.... Above all, witches try to `connect' with the world around them. Witchcraft ... is about the tactile, intuitive understanding of the turn of the seasons, the song of the birds; it is the awareness of all things as holy, and that, as is said, there is no part of us that is not of the gods.... The Goddess, the personification of nature, is witchcraft's central concept.... The Goddess is very different from the Judeo-Christian god. She is in the world, of the world, the very being of the world.... Witchcraft is a secretive otherworld, and more than other magical practices, it is rich in symbolic, special items._
Modern witches practice some rather unusual rituals, which they believe to be similar to rites practiced by witches in earlier centuries. Most of them are performed in the nude._ It is hardly surprising that these people have a penchant for secrecy, since their neighbors tend to view such behavior as bizarre, immoral or threatening. Many of those who practice these rituals also maintain that some of the women who were burned at the stake as witches very likely were witches, though they had not had any commerce with the devil and had not done any harm to anyone. Others who were burned as witches probably were not witches at all, they contend, but simply demented old women or victims of various social or political vendettas.
In the liberal democracies of the late twentieth century, the assertion that witches exist and that some of the people who were executed for being witches really were witches, though they had harmed no one and made no pact with the devil, is no doubt a rather eccentric thing to say, though the consequences of saying it are (one hopes) likely to be benign. In 16th century Europe, however, the consequences of making the same statement might well have been very different. At that time people who held the view that none of the women accused of being witches had made a pact with the devil or caused any harm, and that these women ought not to be tortured or put to death, might be much more effective if they insisted that witches are myths -- that they simply do not exist -- and thus that all of the women accused of being witches are falsely accused. That, near enough, is what many of them did say, and what most of us say as well.
Similar sorts of political considerations color the contemporary debate over homosexuality. While no one denies that same sex sexual activity occurs, there is some reason to believe that many widely held beliefs about sexual preference, homosexuality and homosexual people are false. This has led some to conclude that there is no such thing as homosexuality or that there are no homosexual people. Others (including some gay and lesbian activists) argue that this is exactly the wrong conclusion to draw. Rather, they insist, homosexuality and homosexuals do indeed exist, though many claims made about them are false. I think it is pretty clear that this dispute is in large measure a political dispute. Both sides agree that there is no kind or condition that has all the features commonly attributed to homosexuality. What is really in dispute is how hatred, prejudice, and discrimination in this area are best confronted and overcome._
The skeptical conjecture that I've been trying to make plausible in the last few pages is that the normative naturalist strategy will not uncover principles of rational ontological inference that are rich enough to tell us, in lots of the most interesting cases, what ontological conclusions we ought to draw when we come to believe that some previously accepted theory is seriously mistaken. In support of this conjecture, I've suggested that in many historical cases the resolution of ontological questions can be explained in part by the personalities of those involved, or by social and political factors in the relevant scientific community or in the surrounding society. This is a descriptive claim about the ways in which ontological questions have in fact been resolved. One might, however, be tempted to think that there is also a normative conclusion to be wrung from this descriptive claim. And since I find such temptations hard to resist, let me sketch how the argument might go.
Suppose it is the case that social and political negotiations of the sort that I've been sketching played an important role in determining the ontological conclusions that were accepted, in the history of science, about atoms, stars, electrons and brains, and about caloric, phlogiston and the aether. Surely it is rational for us now to believe that stars and electrons, etc., exist, and that caloric and phlogiston, etc. don't. But if it is rational for us to hold these beliefs, and if social and political negotiations played an essential role in determining what we now believe, then such negotiations are sometimes an important part of the process leading to rational scientific belief. Moreover, it seems plausible to assume that any process that leads to the formation of a rational belief counts as a rational process. If that's right, then social negotiations are sometimes an important part of the rational process of belief revision in science.
I am, I hasten to add, not entirely comfortable in urging this conclusion since the argument that supports it helps itself to lots of assumptions about rationality, and I have no idea how those assumptions could be defended._ Indeed, I suspect that debates about assumptions like these are best viewed as moves in the vehement social and political negotiations over the role and authority of science in contemporary society. But that's a theme I will have to save for some other occasion.
I don't think that the brief and rather breezy discussion of the previous section comes at all close to establishing my skeptical thesis about normative naturalism. Indeed, the only way to make a persuasive case for (or against) my claim that the naturalist strategy will not produce principles of ontological inference that are powerful enough to assess eliminativist arguments is to do lots of careful historical research and lots of detailed cognitive modeling. And that's a project for scholars with skills very different from mine. But suppose it turns out that I'm right. What then?
Well, if there are no principles of rational ontological inference to guide us in deciding what to conclude from the eliminativists' Premises, then the question will have to be settled in some other way. It is my guess that a good deal of the debate over what conclusion to draw about the existence or non-existence of the posits of commonsense psychology is best viewed as falling under heading (iii) on my list in 11.2. And support for this view can be found in some surprising places. Here is a passage in which Patricia Churchland, widely regarded as a leading advocate of eliminativism, is discussing how it is decided whether to retain or eliminate the entities and properties invoked by a theory, when that theory is reduced to a theory in a more basic science:
If a reduction is smooth, in the sense that most principles of TR [the theory that is to be reduced] have close analogues in Tb [the more basic reducing theory], then the matching of cohort terms denoting properties can proceed and identities can be claimed. Informally, a similarity-fit means that property PR of the reduced theory has much the same causal powers as the cohort property Pb of the basic theory.
Determining when the fit is close enough to claim identities between properties and entities of the old and those of the new is not a matter for formal criteria, and the decision is influenced by a variety of pragmatic and social considerations. The whim of the central investigators, the degree to which confusion will result from retention of the old terms, the desire to preserve or to break with past habits of thought, the related opportunities for publicizing the theory, cadging for grants, and attracting disciples all enter into decisions concerning whether to claim identities and therewith retention or whether to make the more radical claim of displacement._
I am also inclined to think that some of the debate over commonsense psychology falls under heading (iv), since the conclusions that scientists and scholars advocate about the existence or non-existence of commonsense mental states may well have a very significant impact on broader political and social issues. Consider, for example, the question of how to treat people who are mentally or emotionally ill. Making a persuasive political case for public funding of various forms of psychotherapy, particularly "cognitive" psychotherapy with its focus on removing false beliefs and changing unrealistic expectations, will be significantly more difficult if leading scientists and philosophers insist that there are no such things as beliefs and expectations, or that they play no role in a suitably scientific understanding of the causation of human behavior. On the other hand, those who would urge the wider use of psycho-active drugs in treating mental and social problems may find their case significantly easier to make if the eliminativists succeed in persuading lots of people that commonsense mental states are myths._
If the account of ontological decision making that I've been sketching is on the right track, and if, as we have been assuming, commonsense psychology turns out to be mistaken in important ways, then, for a while at least and perhaps for a very long while, indeterminacy looms. The question we have been pursuing since the end of Section 3, viz.
What conclusion should be drawn about the existence of intentional states?
will have no determinate answer until the political negotiations that are central to the decision have been resolved. When those negotiations are finished and we have reached a broad social consensus, the conclusion that wins will have much the same status as the conclusions that were reached in the case of stars or atoms or phlogiston._ If, in the spirit of the argument at the end of 11.2, we are prepared to say, in retrospect, that the conclusions reached about the existence of stars, atoms and phlogiston are the ones that rational people should accept, then the same will be true for the conclusion reached about the existence of intentional states, whatever that conclusion turns out to be. I've been warned more than once that the view I'm urging makes me a "social constructionist" -- or something dangerously close. If so, so be it. However, as I will argue in the section to follow, it also situates me squarely within the Pragmatist tradition that runs from Peirce to Quine to Rorty. And like Brer Rabbit in the briar patch, that's a place where I feel right at home.
Earlier versions of the ideas sketched in the previous section have been tried out in a variety of forums over the last few years. In this final section I'll try to clarify my position by considering two very different reactions that it has provoked. One of them maintains that my thesis about the role of politics and personalities in resolving scientific disputes is dangerously radical -- or would be if it weren't simply absurd. The other suggests that the role played by social and political factors in resolving ontological questions, though real enough, is utterly innocuous -- hardly more than a platitude. I think the right reaction is somewhere in between.
Those who are most alarmed by my view claim that it is anti-rational, nihilistic and ultimately self-defeating. If political, social and psychological factors, rather than evidence and principles governing rational reasoning, can determine what conclusion we accept about the existence or non-existence of intentional states (or phlogiston, or stars), the critics ask, then why don't the same factors determine the answers to questions like: Does smoking cause cancer? or Did millions of people die in the Nazi death camps? or Is the earth more than 10,000 years old? And if political and social factors do determine the answers to these questions then it seems that for all questions about the way the world is, might makes right and there can be no objective or rational inquiry. Nihilism reigns! Moreover, as one critic went on to note, if political and social factors really did determine the conclusions we should accept about all these questions, they would determine the conclusions we should accept about the correctness of the view I am advocating. So if someone wished to refute my view, he could do so by mounting a smear campaign, or making large and prestigious grants to those who opposed it, or just hiring some hit men to kill those who are sympathetic to the view and burn their manuscripts!
Fortunately, the Orwellian nightmare that the critic is conjuring can be traced to a misunderstanding of the view I am proposing. I am not a thoroughgoing skeptic about the normative naturalist strategy -- quite to the contrary. I think that over the centuries scientists have developed an increasingly sophisticated and powerful set of strategies for going about the business of reasoning and inquiry in various domains, and that conveying these strategies from teachers to students is one of the fundamental functions of education and apprenticeship in science. But expertise in this area, like expertise in most other domains, is largely tacit. The aim of normative naturalism, as I construe it, is to make explicit the strategies of reasoning and inquiry that underlie good scientific practice. I suspect that there are going to be lots of such principles. And, for those with the appropriate goals, the principles have real normative clout, since they are the best strategies of reasoning that our species has yet developed for pursuing those goals._ Moreover -- and this is my direct response to the accusation of nihilism -- I think these principles are rich and detailed enough to provide normative guidance on lots of questions. My skepticism about normative naturalism is focused quite sharply on its capacity to produce principles of rational ontological inference that are strong and detailed enough to dictate what we should conclude about the entities invoked by a theory when we come to believe that the theory is seriously mistaken. In particular, I very much doubt that the principles that the normative naturalist process will uncover will tell us what ontological conclusions to draw if the Premises in the eliminativists' argument turn out to be correct. But in claiming that normative naturalism will not produce comprehensive principles that specify a determinate answer in all cases, I am most emphatically not claiming that the principles of reasoning these strategies yield will be so radically incomplete that they will yield no determinate answer on any question, or on most. So much, then, for the accusation of nihilism.
The second reaction I want to consider is set out with great clarity in a recent paper by David Papineau._ Papineau agrees that ontological questions of the sort we've been considering may well be "indeterminate,"_ and he also agrees that in many cases "scientific decisions are inevitably determined by sociological factors." However, on his account there is little to be concerned about in all of this, since sociological factors and "scientific politics" "only come into play when a scientific term ... turns out to be vague in a way that requires remedying." Since "there is generally no objective reason to refine [the definition of the term] in one direction rather than another," political and sociological factors are free to step in. But there is nothing of empirical substance at stake when such questions arise. They are really just linguistic questions. "It is simply a matter of deciding how to use words."_ This is a refreshingly down to earth view of the situation, one which I wish I could accept. To explain why I can't, I'll have to back up a bit and sketch the argument that Papineau offers in support of his view.
Papineau begins with the assumption that the Ramsey/Carnap/Lewis account of the meaning and reference of theoretical terms is on the right track, and thus that Putnam-style causal-historical accounts are mistaken. As we saw in Section 4, the Ramsey/Carnap/Lewis view claims that the meaning of theoretical terms can be captured by definite descriptions of a certain sort, and that these descriptions determine the reference of the term. For Lewis, whose statement of the view Papineau takes to be "definitive," theoretical terms are "defined as the occupants of the causal roles specified by the theory...; as the entities, whatever those may be, that bear certain causal relations to one another and to the referents of O-terms [the old, original terms we understood before the theory was proposed]."_ A bit more formally, the meaning of a theoretical term, F1, can be given by the following definition:_
F1 =df (<\ x) (T(x))
where "T(F1) is the set of assumptions [of the theory] involving F1 that contribute to its definition" and "T(x) is the open sentence that results from T(F1) when F1 is replaced by the variable x, and <\ is the definite description operator."_ If, as is likely, T(F1) invokes other theoretical terms, say F2, ... Fn, these can be handled by existentially quantifying into the positions they occupy, yielding:
F1 =df (<\ x1) (E!x2, ..., xn) (T(x1, ..., xn))
"This says that F1 refers to the first in the unique sequence of entities which satisfies T(x1, ..., xn), if there is such a sequence, and fails to refer otherwise (where T(x1, ..., xn) is the open sentence which results when we replace F1, ... Fn by x1, ..., xn in F1's defining theory.)"_ As Papineau notes, one important consequence of this view is that theoretical terms are eliminable. "[A]ny claims formulated using such terms are simply a shorthand for claims that can be formulated without such terms..." Thus "we can eliminate theoretically defined terms from any claims in which they appear."_
This bare bones version of the Ramsey/Carnap/Lewis story requires some embellishments, since, as noted in Section 4, without them it is implausibly holistic and implausibly stringent. All of a theory's causal claims play a role in the definition of all of the theory's theoretical terms. So if any of the theory's causal claims turns out to be mistaken, all of it's theoretical terms will fail to denote. One way to avoid the problem would be to distinguish sharply between those theoretical claims that contribute to the definition of the term and those that do not. But, in agreement with Quine, Papineau maintains that "there is no obvious feature of scientific or everyday thinking which might serve to underpin this distinction."_ Like Quine, he thinks "there is no fact of the matter, for many theoretical assumptions, whether they are definitional or not."_ However, Papineau's agreement with Quine is really quite limited. For, unlike Quine, he thinks that there typically are some "core assumptions" of the theory that "unquestionably do contribute" to the definition of each theoretical term, while other assumptions of the theory "unquestionably do not contribute."_ He also thinks that there will typically be some theoretical assumptions whose definitional status is indeterminate. Papineau uses the handy notation Ty, ("y" for yes), Tn ("n" for no) and Tp ("p" for perhaps), to label these three sets of assumptions. Since the definitional status of Tp is unsettled, the definitions of theoretical terms are imprecise.
The central point of Papineau's paper is that in many cases, "the imprecision does not matter." For if a theoretical term, F, "has this kind of imprecise definition, with Ty strong enough to ensure a unique referent and Ty-plus-Tp not too strong to rule out a referent entirely,... [then] F would end up referring to the same entity however the imprecision were resolved. Given this, there is no need to resolve the imprecision. Understand F as you will, consistently with your definition including Ty and excluding Tn, and you will be referring to the same thing." Since this sort of imprecision or vagueness "does not necessarily imply any indeterminacy of reference, it does not automatically follow that any sentences involving F will lack determinate truth conditions."_
Papineau recognizes that things need not always work so nicely, however. His account of why problems arise, how they can be resolved, and why the resolution need not be disquieting is worth quoting at some length.
... I admit that there are also cases where such definitional imprecision does lead to claims which lack determinate truth-conditional content. When this happens, our discourse is flawed. So when we identify such cases, we ought to remedy the imprecision.
... [T]here are two dangers a theoretical definition of some term F must avoid, if it is to yield a term that is useful for stating truths. It must not make the definition so weak as to fail to identify a unique satisfier. And it must not make the definition so strong that it rules out satisfiers altogether. The special risk facing imprecise definitions is that the indeterminate status of the assumptions in Tp can make it indeterminate whether these two dangers have been avoided. Thus, to take the first danger, Ty might be too weak to ensure a unique satisfier by itself, but may be able to do so if conjoined with some of the assumptions in Tp. Then it will be indeterminate whether F refers uniquely. Second, it may be that nothing satisfies all of Ty-plus-Tp, but that there would be satisfiers if we dropped some of the Tp assumptions from this conjunction. Then it will be indeterminate whether F refers at all.
Cases of the second kind are perhaps more familiar. The example of belief ... fits this bill. Suppose that the assumption that beliefs have internal structure is in the Tp of "belief"'s definition. And suppose further that the connectionists are right to deny that the entities which satisfy the undisputed criteria for being beliefs ("belief"'s Ty) have no internal structure.... Then it would be indeterminate whether "belief" refers to those entities or not, since it would be indeterminate whether their lack of causal structure disqualifies these entities from being beliefs. It is plausible that many cases from the history of science and elsewhere have the same structure. Does the failure of "caloric is a fluid" mean that there is no caloric? Does the failure of "straight lines are Euclidean" mean that there are no straight lines? Does the failure of "witches have magic powers" mean that there are no witches? Does the failure of "entropy invariably increases in a closed system" mean that there is no entropy? And so on.
There are also plausible cases of the other kind, where the indeterminacy of Tp makes it indeterminate which entity, if any, some term refers to. Thus modern microbiology tells us that various different chunks of DNA satisfy the undisputed criteria for "gene", and that further assumptions are needed to narrow the referent down....
In both these kinds of cases, some new discovery makes it manifest that the looseness in the definition of some term F is not benign after all....
The obvious remedy in this kind of situation is to refine the definition so as to resolve the question....
Is there anything which makes it right to go one way rather than another in such cases? I doubt it. Certainly if we look at the history of science, there is no obvious principle which decides whether in such cases scientists concluded that there are no Fs, or, alternatively, that there are Fs, but the assumption at issue is false of them. Consider the different fates of the terms "caloric" and "electric". Originally both of these were taken to refer to a fluid that flowed between bodies.... Later it was discovered that neither quantity is a fluid, and that the appearance of flow is in both cases a kinetic effect. The two cases are structurally similar. Yet we now say that electricity exists, but that caloric fluid does not.
If there is a pattern governing which way the terminology goes in such cases, it is probably one involving the micro-sociology of the thinkers responsible for the relevant theoretical revision. Theorists who want to present themselves as merely continuing the tradition of those who have previously studied Fs will retain the term F for the thing satisfying the basic criteria Ty but not the newly revised part of Tp. On the other hand, theorists who want to distance themselves from the existing theoretical establishment will urge that Fs do not exist, and that their new assumption identifies a hitherto unknown entity G....
The idea that scientific decisions are inevitably determined by sociological factors seems antithetical to any realist attitude to science....
However, the limited role I have ascribed to sociological factors has no such implication. For I am suggesting only that sociological factors come into play when a scientific term that was hitherto thought to have a determinate reference turns out to be vague in a way that requires remedying. When the meaning of a vague term needs refining, there will generally be no objective reason to refine it in one direction rather than another. It is simply a matter of deciding how to use words, given that our previous practice with these words has proved inadequate. So the intrusion of sociological factors at his point need cause no disquiet to the realist._
To get a bit clearer on the implications of Papineau's position for the issues that are central to this Chapter, let's focus in on the problem facing theorists (like the eliminativists) who come to believe that some of the claims in a previously accepted theory are mistaken. Suppose that F is a theoretical term in the old theory. What ought the theorists to conclude about the existence of Fs? According to Papineau, these theorists may be in one of three quite different situations.
In the first situation one or more of the claims that the theorists now take to be mistaken are part of Ty -- the set of "core assumptions" that "unquestionably do contribute to the definition" of F. In this case, according to Papineau, the definition requires the theorists to conclude that there are no Fs. If for some reason the theorists are disinclined to say "There are no Fs," they might also decide to change the meaning of F by reassigning the rejected assumptions to Tn. They could then go on saying "Fs exist." But in saying this they would not be denying what, prior to the change, someone could have asserted by saying "Fs do not exist."
In the second situation, all of the mistaken assumptions are in Tn. Here, the appropriate conclusion is that there are Fs, though the old theory made some mistaken claims about them. As in the previous case, theorists who are so inclined are perfectly free to change the meaning of F in a way that makes one of the rejected assumptions part of Ty. They could then say "Fs do not exist." But this is perfectly compatible with their earlier assertion, "Fs exist."
Finally, in the third situation, one or more of the mistaken assumptions is in Tp, though none are in Ty. In this case the imprecision in the definition of F leads to indeterminacy. Claims that invoke F, including the claim that Fs exist, have no determinate truth value. Since the imprecision matters in this case, something must be done to remedy the problem. There are two obvious options. Theorists can either decide to reassign one or more of the mistaken assumptions to Ty, in which case they will say "Fs don't exist," or they can decide to reassign all the mistaken assumptions to Tn, in which case they will say "Fs do exist." But which ever option they adopt, the claims they make after the decision to adopt it will neither deny nor affirm what they were saying when they uttered "Fs exist" or "Fs don't exist," prior to the decision. For psychological or political reasons, different theorists may prefer different decisions. But there is no factual issue at stake in these disputes. One way to make this obvious relies on the fact that, on Papineau's account, theoretical terms are eliminable. Imagine that there are two theorists, one of whom wants to continue saying "Fs exist" while the other wants to say "Fs don't exist." We need only invite each of them to restate their claims without invoking the term F, by using the Ramsey-sentence strategy. Since the two theorists have revised the old definition in different ways, the result will be two quite distinct sentences, both of which both theorists should be happy to accept. The only real dispute between them is a dispute over how to re-define the term F. Though Papineau does not make the point, his story even has a nice explanation of why, once the political dust has settled, we all ought to conclude that Fs (read: "atoms" or "electrons") do exist, or that Fs (read: "phlogiston" or "witches") do not exist. It's not because there was a correct conclusion that the political negotiations inevitably reached. It's simply that the political negotiations are the process by which we decide how to reform our language. And once a consensus has been reached, we ought to abide by it for much the same reason that we ought to use the word "dog" when we want to say something about dogs, not the word "cat".
Papineau's paper offers an attractive and very reassuring package. His view is well informed, crystal clear, and the account it provides of the role of political and psychological factors in scientific decisions is much easier to live with than the one I've proposed. It's not a package that I think we ought to buy, however, since the philosophical price we'd have to pay for the reassurance it offers is far too high. In order to make his position plausible Papineau must help himself to a pair of very fundamental assumptions. Characteristically, he is completely up front about their status as assumptions; he doesn't pretend to offer any serious arguments for them. On my view, neither of the assumptions is defensible.
The first assumption is that a description theory, something at least roughly similar to the account offered by David Lewis, is the right theory of reference for theoretical terms. As we noted in Section 5, there are many philosophers who reject this account of reference for theoretical terms, and favor instead a Putnam-style causal-historical account. More recently, the deflationary disquotational account championed by Brandom, Horwich and Field has also won some converts. So the truth of the description theory is hardly an assumption we should grant to Papineau without argument. But, as I explained in Section 6, my own reservations about the description theory go deeper than this. I don't think it makes much sense to debate which theory gives the right account of reference for theoretical terms until we get a lot clearer on what it is that the "right" theory of reference is supposed to do. Moreover, as I argued in Sections 6 and 7, there is a sensible answer to this question on which it's entirely possible that there is no correct account of reference for terms in a mistaken theory, and another sensible answer on which there may be several quite different theories of reference, all of which are equally correct. So my complaint about Papineau's first assumption is twofold. First, he has assumed a controversial account of the reference of theoretical terms. Second, his assumption presupposes that there is a correct theory of reference for various sorts of terms. And that presupposition is one I have taken pains to undermine.
The second assumption that underlies Papineau's position is that there is a philosophically defensible way to draw an analytic / synthetic distinction. The distinction he needs is not quite the traditional analytic / synthetic distinction, since Papineau does not claim that all declarative sentences are either analytic or synthetic. He allows, indeed insists, that there are some sentences whose status is indeterminate. But for Quine, and for most other critics of the analytic / synthetic distinction, the existence of indeterminate cases is of little moment. What Quine and those who follow him deny is that there are any sentences whose truth depends on nothing apart from their meaning. And on Papineau's view there are going to be lots of these sentences. Each member of Ty for a given theoretical term F (i.e. each of the "core assumptions" that "unquestionably do contribute to F's definition") generates an analytic truth. If Ty1 is one of the core assumptions for F, then
(8) For all x, if x is F then x is Ty1
is true by definition. It is analytic truths like these that mandate a negative ontological decision when one of the assumptions in Ty turns out to be mistaken. Papineau also thinks that when indeterminacy becomes problematic we sometimes make sentences like this true (indeed, analytically true) by stipulating a revised definition of F. Once that's done, we can't rationally reject the sentence without again changing the meaning. For Quine, by contrast, there are no cases of this sort. As sciences evolves, every sentence in a theory is a potential candidate for revision or rejection, whether or not it was originally introduced as a stipulation, and there is no principled distinction between revising a sentence that was originally a definition and revising any other sentence. Indeed, this egalitarian attitude that grants no sentence special privileges because of its meaning or its history is one of the central strands in Quine's version of pragmatism.
It would be easy enough to set out lots of other illustrations of the fact that Papineau's view clashes head-on with Quinean skepticism about analyticity. But there is no need, since the point is both obvious and uncontested. Papineau himself is quite explicit about the matter. He recognizes that "many of the objections to a simple analytic-synthetic distinction apply equally to my tripartite division of theoretical assumptions into analytic, synthetic and indeterminate." He is untroubled by this, however, since he thinks those objections are unconvincing, and that it is a straightforward matter to determine empirically whether various sentences are analytic, synthetic or indeterminate in a given linguistic community._
Where does all this leave us? Well, if you're prepared to accept a description theory of reference for theoretical terms, along with a version of the analytic / synthetic distinction that allows for indeterminate cases, then Papineau's view provides an account of the role of politics and personalities in scientific decision making that few will find disquieting. It also provides an attractive strategy for answering our question about the right conclusion to draw if the eliminativists' Premises are right: First, we do some meaning analysis to determine which of the three situations distinguished above obtains. If it's the first, then the eliminativists get to draw their conclusion; if it's the second, they don't. And if it's the third, then the conclusion itself is defective -- it's vague in a way that prevents it from having a determinate truth value. So we can sit back and let the political dust settle, confident that the only real issue is how to redefine our propositional attitude terms. On the other hand, if you're not prepared to accept Papineau's assumptions, and if I'm right in guessing that normative naturalism won't produce principles that are strong enough to be of help, then until the social negotiations reach a consensus there is no determinate answer to the question of what to conclude if the eliminativists' Premises are correct. And there is more at stake in these debates than how to redefine a word. What's at stake is whether or not we ought to accept the existence of propositional attitudes.
Let me close with a few brief historical observations that may help to locate my view in philosophical space. My disagreement with Papineau is in many ways quite similar to Quine's dispute with Carnap over the role of language in settling ontological questions._ On Carnap's view, scientific inquiry, when rationally reconstructed, can be divided into two quite different sorts of activities. One of these is the choice (or sometimes the construction) of the language or "linguistic framework" in which our inquiry will be conducted. Within a given framework, certain ontological claims (like "Numbers exist" or "Propositions exist") will be trivial analytic truths. Frameworks will also include "meaning postulates" which specify or entail analytic conditionals (like (8)) stating necessary and/or sufficient conditions for being an entity of a certain sort. When a framework has been adopted there will remain lots of questions, including lots of ontological questions, that are far from trivial and must be resolved by empirical or logical investigation. Those investigations constitute the second stage of scientific inquiry whose job it is to develop and test hypotheses or theories within the linguistic framework that has been chosen. Sometimes, as science progresses, it may turn out that the linguistic framework being used is awkward or inconvenient or perhaps even defective. When this happens, investigators may decide to adopt another linguistic framework. The question of which linguistic framework to adopt is not a "theoretical" question, according to Carnap. It is "a matter of a practical decision concerning the structure of our language. We have to make the choice whether or not to accept and use the forms of expression in the framework in question."_ Thus, for example, to accept a linguistic framework containing the apparatus for talking about familiar, "spatio-temporally ordered system of observable things and events" is "to accept rules for forming statements and for testing, accepting, or rejecting them." And "the efficiency, fruitfulness, and simplicity of the use of the thing language may be among the decisive factors" in deciding whether to accept it._ "The acceptance cannot be judged as being either true or false because it is not an assertion. It can only be judged as being more or less expedient, fruitful, conducive to the aim for which the language is intended._ But, as we've seen, the acceptance of a framework commits us to accepting lots of sentences, including both existential claims and conditionals like (8). On Carnap's view, we accept these because the framework that they are part of is useful for our purposes. Other statements that the framework makes available we accept or reject on the basis of evidence along with the framework's rules for testing, accepting and rejecting such statements.
As I read him, the central thesis in Quine's critique of Carnap is that this distinction between two different kinds of statements -- those that are mandated by our choice of linguistic framework, which we accept because the framework is useful, and those that are not mandated by the framework and that we accept on the basis of evidence -- is not a distinction that can be drawn in a principled way when we look at the history of science, or at the evolution of our own beliefs. If we could draw this distinction in a principled way we would have all that is needed to distinguish the analytic from the synthetic. Or, going the other way round, if we could distinguish analytic from synthetic statements, there would be little problem in distinguishing between those scientific developments that count as changes in language from those that count as changes in theory. But, Quine maintains, we can do neither.
Carnap ... and others take a pragmatic stand on the question of choosing between language forms, scientific frameworks; but their pragmatism leaves off at the imagined boundary between the analytic and the synthetic. In repudiating such a boundary I espouse a more thorough pragmatism. Each man is given a scientific heritage plus a continuing barrage of sensory stimulation; and the considerations which guide him in warping his scientific heritage to fit his continuing sensory promptings are, where rational, pragmatic._
It will come as no surprise that I side with Quine in this controversy, while Papineau's view is much closer to Carnap's. But there is one respect in which my pragmatism may be more radical than Quine's. In talking about the factors that "influence our inclinations to adjust one strand of the fabric of science rather than another" Quine mentions "conservatism," "the quest for simplicity," and "the continuing barrage of sensory stimulation."_ But there is no mention of the role of psychological and political considerations in determining either the inclinations of an individual or the ultimate consensus in a community. I have argued that both of these factors often do play important roles in inquiry. I have also argued, more tentatively, that at least sometimes they ought to.
_. Philosophers are not the only ones to advocate eliminativism, nor were they the first. J. B. Watson, one of the founders of behaviorism in psychology, disparaged the mental states mentioned in commonsense psychological explanations, states like beliefs and desires, as "medieval conceptions," and "heritages of a timid savage past," of a piece with "superstition," "magic" and "voodoo." (Watson, 1930, pp. 2-5.) B. F. Skinner, perhaps the most famous behaviorist, tells us that "mental life" is an "invention," and that "mental or cognitive explanations are not explanations at all." (Skinner, 1974, pp. 114-115.)
_. In this Chapter, I will capitalize "Premise," "First Premise" and "Second Premise" when referring to the Premises of the eliminativists' argument.
_. See Chapter 2, fn. 6.
_. Stich (1991a), (1992), (1993b); Stich and Warfield (1995).
_. See Chapter 3, Section 4.
_. McCloskey (1983a), p. 308.
_. McCloskey (1983a), p. 311.
_. McCloskey (1983a), p. 301.
_. McCloskey (1983a), p. 302.
_. McCloskey (1983a), p. 310.
_. McCloskey (1983a), p. 310.
_. Clement (1983).
_. Quoted in McCloskey (1983b), p. 124. For a quote from Galileo's De Motu, written in about 1590, see McCloskey (1983b), p. 127.
_. See, for example, Chomsky (1965), Ch. 1, and Chomsky (1975).
_. See Chapter 3, Section 4.
_. For the former view, see Harman (1973); for the latter see Fodor (1975).
_. For quasi-pictorial representations, see Kosslyn (1994); for holographic representations, see Eich (1982) and Metcalf (1989); for mental models, see Johnson-Laird (1983); for connectionist representations, see Hinton, McClelland and Rumelhart (1986).
_. The most explicit example I know is Andy Clark's recent proposal that in order to counter the eliminativists we should "reject outright the idea that folk psychology is necessarily committed to beliefs and desires as being straightforwardly causally potent. (Clark, 1993, p. 211). Others who suggest similar moves include Jackson and Pettit (1990) and (forthcoming), and Horgan and Graham (1990).
_. Sellars (1956).
_. Sellars (1956), p. 317.
_. For a more extended discussion of Sellars' "myth of Jones," see Chapter 3, Section 2.
_. See, for example, P. S. Churchland (1980) and (1986), pp. 386-399.
_. Fodor (1975), Chs. 1, 3 and 4.
_. P. S. Churchland (1986), p. 396.
_. Hooker (1975), p. 217. Quoted in P. S. Churchland (1986), p. 396.
_. Van Gelder (1991), p. 56.
_. Fodor (1985), p. 23.
_. See, for example, Loar (1983) and Stalnaker (1984).
_. O'Brien (1991) p. 173.
_. A precursor of the Ramsey, Stich and Garon argument can be found in Stich (1983), pp. 237-242. However, back in those days there were no up-and running radically holistic models of memory.
_. Davies (1990) and (1991).
_. The quote is from a useful summary of Davies' argument in Clark (1993), p. 197. Clark calls this assumption "the demand for conceptual modularity." Conceptual modularity certainly does not entail a full fledged Sellarsian language of thought, though perhaps it comes closer than the propositional modularity invoked in Ramsey, Stich and Garon's argument.
_. Smolensky (1988).
_. Wilson (1985), quoted in Stich (1983), p. 236.
_. For a critique of this argument, see Horgan and Woodward (1985).
_. The quote is from Putnam (1975); for Kripke's arguments see Kripke (1972).
_. Roughly speaking, one set of properties supervenes on another if the presence or absence of properties in the first class is completely determined by the presence or absence of properties in the second. For a detailed discussion of various ways in which this account can be made more precise, see Chapter 5, Section 4.
_. This is, near enough, the "principle of psychological autonomy" that I advocated in Stich (1978a). For some additional discussion of the principle, see Stich (1983), Ch. 8.
_. Fodor (1987), Ch. 2, and Fodor (1991). As Fodor notes, the arguments in these two sources are very different.
_. Burge (1986); Van Gulick (1989); von Eckardt (1993), ch. 7.
_. See Stich (1983), Chapter 4. For a critique of this argument, see Fodor (1987), Chapter 3.
_. Fodor (1978), p. 223.
_. Stich (1983), Ch. 8.
_. Or, on a slightly different version of the view, only if they have identical inferential properties. This is, needless to say, a very demanding condition to impose on content identity. Indeed, it is so demanding that Jones, Mullaire and I (1991) claimed it was a straw man -- a view that no one had every really held. But, as Michael Devitt (forthcoming, Ch. 1) has recently demonstrated, we were wrong. Harman (1973, p. 14) clearly asserts the view, and both Papineau (1987, p. 98) and Block (1991, p. 260) come very close indeed.
_. For one widely discussed version of this argument, see Fodor (1987), Ch. 3.
_. For some detailed examples, see Stich (1983), pp. 66-72 and Stich (1982), pp. 185-88.
_. See, for example, Kim (1989a) and (1993), and Bickle (1992).
_. Van Gulick (1993). See also Kim (1989a) and (1989b). Both Kim and Van Gulick are critical of the argument, though
their objections are quite different.
_. The first quote is from Fodor (1987), p. 98; the second is from Fodor (1984), p. 32.
_. Fodor (1987), p. 97.
_. Perhaps this is a good place to note that some authors use the term "eliminativism" for the view that folk psychology is a seriously mistaken theory, rather than for either of the ontological theses that I take to be eliminativism's central claims. Other authors seem to slide back and forth between the "ontological" reading and the "badly mistaken theory" reading, a practice that is easy to understand if they think that the former reading follows straightforwardly from the latter.
Yet another distinction worth noting separates two versions of ontological eliminativism, one of which claims that beliefs (and other propositional attitudes) don't exist, while the other claims that believers (and desirers etc.) don't exist. If we assume that beliefs are a sort of internal state, and that someone is a believer if and only if she has some beliefs, the distinction is of no importance. However, there are some writers who reject this account of what it is to be a believer, because they think it is a mistake to view propositional attitudes as states or "entities." But, of course, these writers do not deny that humans and other organisms are entities. So eliminativists can pose their challenge in a way that is acceptable to these writers by claiming that no humans or other organisms are believers or desirers -- that predicates of the form `___ believes that p' and `___ desires that q' are never true of them. For more on this distinction, see Stich (1991a), Sec. III, and O'Leary-Hawthorne (1994), Sec. I.
_. For a bit of evidence in support of this guess, see fn. 70.
_. The papers include Lewis (1966), Lewis (1970) and Lewis (1972). The quote is from Lewis (1972), p. 209.
_. Lewis (1972), p. 207.
_. Lewis (1972), p. 208. This is the second time I've quoted this passage from Lewis. The first time was in Stich (1983), pp. 15-16, where I was laying the groundwork for a series of arguments for eliminativism.
_. Lewis (1972), p. 209.
_. Lewis (1972), pp. 208-9.
_. Lewis (1972), p. 209.
_. Lewis (1972), p. 211.
_. Lewis (1970), p. 85.
_. Lewis (1972), p. 209.
_. Lewis (1972), p. 210.
_. Lewis (1972), p. 210, emphasis added.
_. The idea that the eliminativists' claim may be indeterminate even when all the scientific facts are in will become something of a leitmotif in this Chapter. It arises in another context in Section 7, and gets a rather radical twist in Section 11.
_. Lewis (1972), p. 212.
_. Lewis (1972), pp. 212-3. It might be thought odd that Lewis treats all mental state terms (indeed all theoretical terms) as names rather than predicates or functors. But as he notes (Lewis, 1970, p. 80):
No generality is lost, since names can purport to name entities of any kind: individuals, species, states, properties, substances, magnitudes, classes, relations, or what not. Instead of a T-predicate `F ____', for instance, we can use `____ has F-hood'; `F-hood' is a T-name purporting to name a property, and `____ has ____' is an O-predicate. It is automatic to reformulate all T-terms as names, under the safe assumption that our O-vocabulary provides the needed copulas....
_. Lewis (1972), p. 213.
_. Lewis (1972), p. 214.
_. Lewis (1972), p. 212. For more on this view of folk psychology, see Chapter 3.
_. Though I may have been the only person tempted by eliminativism who relied explicitly on Lewis's account, other writers clearly favor similar accounts. According to Paul Churchland:
Our common-sense terms for mental states are theoretical terms of a theoretical framework (folk psychology) embedded in our common-sense understanding, and the meanings of those terms are fixed in the same way as are the meanings of theoretical terms in general. Specifically, their meaning is fixed by the set of laws/principles/generalizations in which they figure. (1984, p. 56)
Similarly, Patricia Churchland writes:
The meaning of an expression for an individual is a function of the role that expression plays in his internal representational economy -- that is, of how it is related to sensory input and behavioral output and of its inferential/computational role within the internal economy. Sparing the niceties, this is the network theory of meaning, otherwise known as the holistic theory or the conceptual-role theory. (1986, p. 344)
This sort of account has also been influential among authors on the other side of the issue, who are not in the least tempted by eliminativism. Here's an example from Colin McGinn:
It is very plausible that the platitudes of commonsense psychology, as they relate mental states to each other and to behavior, implicitly define the terms contained therein: it is thus a necessary, indeed analytic, truth that entities are mental if and only if they satisfy those platitudes. In consequence if (per impossibile) the commonsense theory should be false, then mental states would simply not exist - for the corresponding terms have no meaning, and hence no denotations, beyond their role in that theory. (1978, p. 150).
_. Lycan (1988a), p. 32.
_. Lycan (1988a), pp. 31-32.
_. Lewis offers his theory as an account of both the reference and the meaning of theoretical terms, and this is certainly the most natural way to construe it. Causal-historical theories, by contrast, are generally offered only as accounts of reference. So one might try to preserve something like Lewis's account of meaning and add on a causal-historical account of reference. This is, near enough, the strategy pursued in "dual factor" theories. Since reference rather than meaning does all the work in eliminativists' arguments, I won't have much to say about theories about the meaning of theoretical terms.
_. Devitt and Sterelny (1987), Secs. 3.3, 4.2, 5.1 and 5.2.
_. Or theories - since it might turn out that different accounts of reference are appropriate for different sorts of terms.
_. The inspiration for this example is in Kripke (1972), p. 282, though Kripke does not include the bit about the fellow who actually was swallowed by a big fish. I'm not sure who first added this wrinkle to the tale, though if memory serves, I first heard it from Kripke.
_. See Stich (1972) and (1975), and Soames (1984). The inspiration for this account of syntax can be found in Quine (1953a). There is, of course, no guarantee that there will be a unique simplest way to capture speakers' intuitions. If there are several equally simple theories, then on this view of grammar they are all correct.
_. For more on these two views of grammar, see Chapter 3, Section 4.
_. This account is similar to "external" accounts of folk psychology discussed in Chapter 3. The closest analog is the view second from the left in the box in Figure 3.1.
_. It is important to note that there is a substantive empirical assumption lurking in the background here. If we adopt the view that the job of a theory of reference is to describe the tacit theory that characterizes a word - world mapping and that plays a central role in producing our intuitions about various examples, then we are assuming that ordinary folk actually have an internally represented theory of reference guiding their intuitions. And there is certainly no guarantee that this assumption is correct. It might be the case that our intuitions about the sorts of examples that philosophers are wont to discuss (Jonah, witches, stars, and the rest) are guided by a heterogeneous set of principles about translation, paraphrase, charitable interpretation, historiographical convenience, inferential practice, and a host of other considerations. If I understand him correctly, this is the view urged by Rorty (1979, pp. 284-295) who maintains that reference "is not something we have intuitions about." (p. 292). I am inclined to think that Rorty may be right, and thus that much recent philosophical work on reference is built on a foundation of quicksand. But that's a topic for another occasion.
There are passages in Field (1994, Sec. 4), which might be interpreted as agreeing with Rorty, and denying that we have intuitions about reference. However, Field tells me that his view is rather different from Rorty's. For more on Field's view, see fn. 94.
_. The best sustained statement and defense of this view of grammar is in Laurence (1993). For Chomsky's view, see Chomsky (1965), (1975) and (1992).
_. Though there is a fascinating literature on this topic. See, for example, Murdock (1980).
_. The most explicit endorsement of this view that I've found is in Devitt and Sterelny (1987), p. 170:
Just as it is the task of economics to explain such properties as price and value ..., it is the task of linguistics to explain such properties as truth and reference....
In the following paragraph they describe "the semantic properties of linguistic symbols" as "the properties revealed by linguistics." (Emphasis is mine.) Unfortunately, they offer no references to the work of linguists in which the nature of semantic properties is revealed. More recently Devitt (1994 and forthcoming) seems to have adopted a less sanguine view of what linguistics has revealed about the nature of semantic properties.
_. For fitness, see Sober (1984); for grammaticality, see Fodor (1981b) and Laurence (1993); for space-time, see Sklar (1974).
_. Cummins (1989), Chs. 1 and 2, argues persuasively that different branches of cognitive science will require different "representation" relations. This is not quite the same point I'm making here, since the representation relations that Cummins has in mind are mind-world mappings, not word-world mappings. But the points are closely related, and I am indebted to Cummins for getting me to see this possibility more clearly.
_. See fn. 80.
_. Much the same follows for the claim that believers and desirers don't exist -- i.e. that no one believes or desires anything. See fn. 51.
_. Stich (1990), Ch. 5. I am indebted to Peter Godfrey-Smith who was, I think, the first to stress that the reference relation picked out by our intuition is a quite "baroque" and idiosyncratic word - world mapping. See Godfrey-Smith (1986).
_. For much the same arguments in a more confident tone of voice, see Stich (1991a) and (1992).
_. Perhaps the most persuasive response to the Jackson argument would be one which adopted the general outlines of Papineau's position, which is discussed at length in Section 12.
_. See Quine (1960), Sec. 56.
_. "Generally speaking, any expression of the form "`...'" designates ..." is an analytic statement provided the term "..." is a constant in an accepted framework. If the latter condition is not fulfilled, the expression is not a statement." (Carnap 1950/1956, p. 217) For some very sophisticated recent discussion of the deflationary account, see Brandom (1984), Field (1986 and 1994) and Horwich (1990).
_. I.e. (1) or something like it. For obvious reasons, this schema is also sometimes called "the disquotation principle".
_. I'm grateful to Hartry Field for much helpful conversation on these issues.
This might be an appropriate place to say a bit more on how I interpret Field's current view about reference. As noted in fn. 80, there are passages in Field (1994) which suggest that Field, like Rorty, denies that ordinary folk have any intuitions about reference. However, Field tells me that this is not what he intended in those passages. Rather, he is inclined to think that people generally do have a tacit internalized theory of reference, but that this theory is really quite minimal. It contains little more than the semantic ascent (or disquotation) schema and also, perhaps, some information on how to use it. This internally represented schema underlies lots of intuitions about reference. But the intuitions in question are the "trivial" ones -- i.e. the ones that are just instances of the schema, like (2) and (3). In the philosophical literature on reference, by contrast, the focus is on a quite different set of intuitions which are evoked by setting out a scenario like the story about Jonah recounted in 6.1, or like Kripke's scenario in which we discover the following fact:
(F) The incompleteness theorem was proved by a man baptized "Schmidt" and who never called himself anything other than "Schmidt"; a certain person who called himself "Godel" and who got a job under that name at the Institute of Advanced Study stole the proof from him. (Field, 1994, p. 261)
Against the background of stories like these we
are asked: "Who is it natural to say that we have been referring to
when we use the name `Godel'?" On Field's view, our answers to such
aren't at the most basic level about reference but about our inferential practice. That is, what Kripke's example really shows is that we would regard the claim (F) as grounds for inferring "Godel didn't prove the incompleteness theorem" rather than as grounds for inferring "Godel was baptized as `Schmidt' and never called himself `Godel'". Reference is just disquotational. It does come in indirectly: from (F) I can indirectly infer
"Godel" doesn't refer to the guy that proved the incompleteness theorem.
But that isn't because of a causal theory of reference
over a description theory, but only because I can infer
Godel isn't the guy that proved the incompleteness theorem
and then "semantically ascend". (Field, 1994, pp. 261-262)
In claiming that "reference is just disquotational" Field is not merely claiming that our internalized theory of reference is exhausted by something like (1). For he also holds that the disquotational account of reference is the only notion of reference that we really need in the sciences (including semantics and logic). So on Field's view, it doesn't much matter whether we answer the question posed in Sec. 6 by appeal to folk semantics (as in 6.1), or by appeal to proto-science (as in 6.2). In both cases, we will get the same, disquotational, account of reference.
_. Parts of this Section are borrowed from Stich and Warfield (1995).
_. As Michael Devitt has pointed out to me, it looks like the strategy of semantic ascent, or something very much like it, is lurking in the background here too, linking claims of the form:
We ought not apply the concept C to an object lacking some property that is constitutive for that concept.
to claims of the form:
The object in question can't be a C.
_. Clark (1993), pp. 216-218. For some other examples, see Davies (1991), p. 239 ff; Evans (1981), p. 132; Rey (1991), p. 221.
_. Clark (1991), p. 217.
_. Rey, op. cit.
_. Fodor denies that rationality is essential for having a belief. Indeed, he writes: "I accept -- in fact, welcome -- what amounts to the conclusion that people can believe things that are arbitrarily mad. Fodor (1987), p. 88. Clark apparently denies that "logical combination" is conceptually necessary.
_. Fodor (1987), Chapter 1; Clark (1991); Egan (1995); and perhaps Dretske (1989), p. 1. Clark is clearest in claiming that causal potency is a conceptual necessity. Fodor tells us that it "seems to me intuitively plausible" that having causal powers is an "essential property of the attitudes" (p. 10). Dretske and Egan both think that folk psychology is committed to causal potency, though it's not clear that they think it is conceptually necessary. There is also an extensive literature attempting to say what the causal potency requirement comes to. See, for example, LePore and Loewer (1987); Fodor (1989); Horgan (1989) and (1993); Heil and Mele (1993).
_. Jackson and Pettit (1988) and (forthcoming). Clark seems to have changed his mind on this one. See the quote from Clark (1993) in fn. 18.
_. See the quote displayed two paragraphs back.
_. See, for example, Quine (1953b), White (1950), Goodman, (1949) Harman (1967). In reading the books and articles of those who invoke the notion of a constitutive property, it is easy to get the feeling that one has fallen into a time warp. These philosophers write as though the notion of constitutive properties were entirely unproblematic, and they give no indication that they have ever heard of Quine and his assault on the analytic / synthetic distinction. There is, for example, not a single reference to Quine in Clark (1989), (1989/1990), (1991), or (1993), nor in Davies (1990) or (1991).
_. It's perhaps worth noting that throughout this section there is a sense in which I have been stacking the deck in favor of the eliminativist, since I have identified constitutive properties with conceptually necessary properties. If eliminativists can show that there are some properties that are conceptually necessary for having beliefs, and if they can show that people fail to have one or another of these properties, then the eliminativists win. But if they can't show that there are any conceptually necessary properties for having beliefs, then since we have been assuming that the Second Premise of the eliminativists' argument is correct (i.e. that folk psychology is a seriously mistaken theory) the issue remains unsettled. For the opponents of eliminativism to win using an analogous argument, they would have to defend the claim that there are properties that are conceptually sufficient for having beliefs, and that some people have these properties. Needless to say, I don't recommend holding one's breath until the project succeeds in this case either.
_. Goodman (1965). The term "reflective equilibrium" is due to Rawls (1971). For further discussion of the reflective equilibrium strategy, see Daniels (1979) and (1980), Cohen (1981), Stich (1988) and (1990), Ch. 4, and Stein (forthcoming).
_. I believe that the term "normative naturalism" was first introduced by Larry Laudan, though I use the label rather more broadly than Laudan does. See Laudan (1987), Laudan (1990), and Laudan, Laudan and Donovan (1988). For an excellent discussion of the historical roots of normative naturalism and its relation to other projects in the "naturalized epistemology" tradition, see Kitcher (1992).
_. For a discussion of some of these alternatives, see Kitcher (1992). Thagard (1988), Ch. 7, proposes an integrated strategy in which intuitions and historical studies both play a role.
_. E.g. The "Rule of Predesignation" which requires scientists to "prefer theories that make successful surprising predictions over theories which explain only what is already known," or the Popperian Principle that requires scientists to "propound only falsifiable theories." Both of these are discussed in Laudan (1987). The most systematic and impressive effort to determine which discursively formulated principles have actually been adhered to in the history of science is the project proposed in Laudan, et. al. (1986). The results are reported in Donovan, Laudan and Laudan (1988).
_. See Simon (1966), Simon (1973), and Langley, Simon, Bradshaw and Zytkow (1987). Thagard (1988) and (1992) has developed a rather different approach to the computer simulation of episodes in the history of science.
_. See, for example, Qin and Simon (1990) and Dunbar (1989).
_. McCauley (1986).
_. McCauley (1986), p. 72.
_. Wimsatt (1976a) and (1976b).
_. McCauley (1986), p. 75.
_. McCauley (1986), p. 79-80. Emphasis is McCauley's.
_. McCauley (1986), p. 79. Emphasis is McCauley's.
_. McCauley (1986), p. 79.
_. McCauley (1986), p. 77.
_. McCauley (1986), p. 77.
_. For another interesting illustration of the normative naturalist strategy applied to ontological questions, see Enc (1976). On Enc's view there is an important distinction to be drawn between terms that purport to refer to ostensible objects (i.e. objects that it is possible to point to) and terms that do not purport to refer to ostensible objects. `Gold' and `magnet' are examples of the former; `electron' and `phlogiston' are examples of the latter. According to Enc, "the burden of reference for [non-ostensible theoretical terms] will be carried by the kind-constituting properties attributed to the object and by the explanatory mechanism developed in the theory." (p. 271) Scientists will deny that a non-ostensible object or kind exists if and only if it turns out that nothing has all of the kind constituting properties. Macdonald (1995) proposes that we can use Enc's analysis to assess eliminativist arguments. But I am more than a bit skeptical, since I have no idea how we determine which properties of intentional states are the "kind constituting properties." I also have my doubts about the correctness of Enc's account itself since, so far as I can see, it entails that scientists should have decided that atoms and electricity don't exist.
_. See, for example, Doppelt (1990), Giere (1985), Kitcher (1992), Laudan (1990), Nickles (1986), Rosenberg (1990), Thagard (1988).
_. See Stich (1990), Ch. 6, and Stich (1993a)
_. Fine (1975), pp. 24 ff.
_. For an interesting discussion of this case, see Kitcher (1982) and (1984), p. 343 ff, and Waters (1994).
_. Gillispie (1960), p. 509.
_. Gillispie (1960), p. 510.
_. As Enc notes (1976, p. 267), advocates of the Kripke-Putnam account of reference might well be committed to saying that phlogiston is oxygen, since the term `phlogiston' was a term originally introduced to refer to the substance (or "principle") that is responsible for combustion and calcination (the process in which a heated metal loses its metal-like properties like its brilliance and its cohesiveness). And oxygen is the substance that is responsible for both of these phenomena. In his interesting attempt to model the conceptual changes in chemistry from Stahl to Lavoisier, Thagard (1992, Ch. 3) points out that the stuff we would call "oxygen" and that Priestly sometimes described as "dephlogisticated air" was regarded as a compound by Lavoisier. One of its components was the element or principle that he called "principe oxygine;" the other component was "the matter of fire and heat" which Lavoisier, along with most of his contemporaries, also called "caloric"!
_. For a similar view, see Papineau (forthcoming), Sec. 4. Part of the relevant passage is quoted in Section 12 of this Chapter.
_. Luhrmann (1989), pp. 45-8. Luhrmann's book is an ethnographic study of witches in contemporary England. The views recounted in this quote are widely shared in the groups she studied. I'm grateful to Philip Kitcher for drawing my attention to Luhrmann's work.
_. For details, see Luhrmann (1989), pp. 48-9.
_. For some useful discussion, see Stein (1992a) and some of the other essays in Stein (1992b). For a detailed and fascinating discussion of the political negotiations involved in deciding whether or not a controversial psychiatric condition, Binge Eating Disorder, really exists, see McCarthy and Gerring (1994). And for a richly textured account of the social and political underpinnings of the debate about the existence of Multiple Personality Disorder, see Hacking (1995).
_. Though I do have lots of views about how they can't be defended. See Stich (1990), Chs. 2, 4 and 6.
_. Churchland (1986), pp. 283-4. I am grateful to Ted Warfield for bringing this passage to my attention.
_. I owe this observation to Warren Dow.
_. Those negotiations do not take place in isolation, of course. They are carried on at the same time as the empirical and philosophical debates, sketched in Section 3, about the virtues and shortcomings of commonsense psychology. And it seems unlikely that any consensus will emerge on what ontological conclusion to draw if folk psychology were to turn out to be wrong in ways that few believe it will. To see the point, consider the analogy with phlogiston. Suppose that phlogiston theory had turned out to be wrong, but that it's defects were quite different from what we now take them to be. Should we conclude that phlogiston exists or not? Obviously the answer depends on the details. But there are lots of imaginable cases in which we would have no idea what to say. The answer is just as indeterminate now as it was before Lavoisier. Since there is no need to decide what conclusion to accept in all possible cases, those engaged in the process typically don't even try.
_. For further discussion see Laudan (1987) and (1990) and Stich (1990), Ch. 6, and (1993a).
_. Papineau (forthcoming).
_. Papineau (forthcoming), Sec. 2.
_. Papineau (forthcoming), all quotes are from Sec. 7.
_. Lewis (1972), p. 211 and p. 208.
_. For the remainder of this Section I will follow Papineau in using "F", "F1", "G", etc. as dummy names of words, and using underlining, as in "F", "G", etc. for dummy names of a theoretical property or other entity.
_. Papineau (forthcoming), Sec. 2.
_. Papineau (forthcoming), Sec. 3.
_. Papineau (forthcoming), Sec. 3.
_. Papineau (forthcoming), Sec. 1.
_. Papineau (forthcoming), Sec. 6.
_. Papineau (forthcoming), Sec. 4.
_. Papineau (forthcoming), Sec. 4.
_. Papineau (forthcoming), Sec. 7.
_. Papineau (forthcoming), Sec. 6.
_. Though there are also some important differences. A detailed account of the analogies and disanalogies would be an intriguing project, but it's one that I'll have to save for another occasion. The central papers in the debate between Carnap and Quine are: Carnap (1950), (1952) and (1955), and Quine (1936), (1951), (1953b) and (1963).
_. Carnap (1950), p. 207.
_. Carnap (1950), p. 208.
_. Carnap (1950), p. 214.
_. Quine (1953b), p. 46.
_. Quine (1953b), p. 46.