Philosophy and Phenomenological Research Vol. LXXI, No. 2, September 2005
Replies to Alvin Goldman, Martin Kusch and William Talbott
University of Massachusetts, Amherst
I want to thank my three critics for their careful and thoughtful discussion of my work. I am grateful for their challenging comments, and for this opportu- nity to respond to their remarks.
1. Reply to Alvin Goldman
I. I Epistemology and the concept of knowledge Alvin Goldman and I agree on a great deal, but, in spite of our common commitment to naturalism, we disagree about proper method in philosophy. On Goldmans view, epistemology begins with conceptual analysis: we seek to understand our concept of knowledge. When that analysis reveals that knowledge is reliably produced true belief, an empirical investigation of the mechanisms of belief production and retention becomes the focus of episte- mological investigation. Without a prior analysis of the concept of knowl- edge, there is no reason for epistemologists to be interested in the mecha- nisms of belief production and retention. Conceptual analysis thus provides the starting point, and the raison dztre, for further philosophical work.
Goldmans view of conceptual analysis is thoroughly naturalistic. Con- cepts are not platonic entities of any sort, on Goldmans view, nor is our access to them by way of some mysterious process of rational intuition. Rather, concepts are mental entities which play an important role in our men- tal economy, and while intuitions about hypothetical cases provide us with one fallible route to understanding our concepts, carefully controlled experi- mental work is required if we are fully to understand them. Thus, on Gold- mans view of philosophy, neither the conceptual analysis with which epis- temology begins, nor the subsequent work which it motivates, is pursued by a priori means. All the same, philosophical methodology is importantly dif- ferent from method in the sciences, on this view. As Goldman comments,
Any imaginable case can shed useful light on our concept of knowledge, so long as the con- cept can be applied to the case and can generate intuitive responses, thereby indicating some-
BOOK SYMPOSIUM 427
thing relevant about that concepts contours. This highly permissive approach-which often strikes scientists as quite odd (because their own projects are of a different nature)+annot be explained by Kornbliths natural kinds approach. 
Goldman is quite right about this. At the same time, I believe that Gold- man may have underestimated the extent to which I believe that standard phi- losophical practice should be modified. My objection to an investigation of our concept of knowledge is not that it inevitably fails to satisfy naturalistic scruples; there is a great deal of work in the cognitive sciences on the charac- ter of our concepts, and there is nothing a naturalist should object to in this. Goldmans project of conceptual analysis, as I see it, is entirely continuous with work in the cognitive sciences. My objection to an investigation of our concept of knowledge is, instead, that our concept of knowledge is of little, if any, epistemological interest. Epistemologists should be engaged with an attempt to understand the nature of knowledge itself, rather than our concept of knowledge.
Just as chemists (qua chemists) should, and do, have no interest in folk chemical concepts, epistemologists should have no interest in our folk con- cept of knowledge, on my view. Early chemists did not attempt to elucidate the folk concept of gold or water; instead, they investigated gold and water themselves. Our folk concepts are, at times, a product of ignorance and mis- understanding. Chemists who wish to understand the nature of the physical world would be ill-advised to begin their studies with a careful investigation of the contours of our ordinary chemical concepts. Better, by far, to look at the chemical phenomena themselves rather than our pre-theoretic concepts of those phenomena.
Historically, there have been many misunderstandings about the nature of knowledge.* It was once widely held that knowledge required certainty; on some conceptions, it required infallibility. I have no doubt that views of this sort may well have been reasonable. Descartes view of knowledge was clearly influenced by his entirely reasonable, but mistaken, belief that things firm and lasting in the sciences would never be attained without the benefits of derivation from a foundation of beliefs which are immune to error. If we
Goldman says that An epistemological naturalist must either disavow the standard prac- tice (giving reasons for that disavowal) or tell a story that somehow reconciles the prac- tice with the precepts of naturalism. Kornblith takes the latter tack.  I may well have encouraged this misunderstanding I do say, as Goldman points out, There is room within a naturalistic epistemology for the practice of appeals to intuition, suitably under- stood. My suitable understanding, however, involves certain modifications of the standard practice, and so simply embracing either of the alternatives Goldman offers, without further elaboration, would be misleading. Nor need one look to the distant past to see examples of conceptions of knowledge which build in substantive errors. For example, I argue at length, in Chapter Four, that accounts of knowledge which require some sort of reflection on the quality of ones evidence fail to capture the real phenomenon of knowledge.
are interested in the nature of knowledge itself, however, we do best to follow the lead of work in the sciences: look at the phenomena, not at our concept of the phenomena. Investigations of ones concept of knowledge, just as inves- tigations of ones concept of aluminum or of a gene, shield one from the best source of accurate information about the phenomena we seek to understand.
Goldman suggests that even if one thought that this described a coherent project-and 1 will have more to say about this momentarily-it is a project which would itself have to begin with conceptual analysis. After quoting a passage in which I note that knowledge requires more than just true belief, Goldman asks:
Where does the assertion that knowledge is more than just true belief come from? What licenses it? Surely it doesnt come from cognitive ethology. It would have to come, one sup- poses, from a semantico-conceptual account of the term knowledge. But many would say that this is precisely what philosophy, in its analytic phase, aims to provide. So that job is not taken over by biological science, as Kornblith often suggests that it is. 
Similarly, Goldman asks how we know which of the indefinitely many natu- ral kinds-including the various kinds in mineralogy or chemistry- knowledge is supposed to be. The answer, on Goldmans view, could only be provided by the kind of conceptual analysis he favors and which I would have us eschew. There is, I believe, an important disagreement between us on this issue.
Imagine an early chemist interested in the nature of acids.3 The term acid was widely used before there was any real understanding of what it is that makes something an acid. So this chemist has vinegar (which is a dilute solution of acetic acid), hydrochloric acid, aqua regia (a mixture of hydro- chloric acid and sulphuric acid) available in his laboratory, and he is trying to determine what, if anything, these various substances have in common. He believes they are all members of a single natural kind, and he is interested in determining what it is that makes them members of that kind. He has some views about what these substances have in common-many of which are mistaken-but instead of analyzing his concept of acid, he turns to the work- bench and tries to figure out what these substances actually have in common. No one doubts the coherence of this project.
Now imagine that another investigator hears about this project and announces that he wishes to help out. He too is going to find out what all acids have in common, and he has a number of samples of would-be acids which will form the basis of his investigation. Now suppose that the sam- ples which this investigator is examining include shoes, ships, sealing wax and his pet dog. Clearly something has gone wrong. This second investigator
For an illuminating discussion of the origin and development of chemical terminology, see Maurice Crossland, Historical Studies in the Language of Chemistry, Dover, 1978.
BOOK SYMPOSIUM 429
is not engaged in the same project as the first, and it will be immediately obvious to anyone looking on that this is so. The same is true if this inves- tigator has samples which are members of a single natural kind, but one nowhere in the vicinity of an acid: say, a dog, a cat, a cow and a sheep. How are we to explain the mistake that this investigator is making?
We might, following Goldman, ask for some sort of semantico-conceptual account of what an acid is, and use this to show that the investigation of shoes, ships and so on, or dogs, cats and so on, is not of the same sort as the first, chemical, investigation. There is a mistake which our second investiga- tor makes here, on this view, and it has to do with a failure of semantic com- petence. Im not sure whether I would want to disagree with this? but notice that the amount of conceptual analysis needed to rule out the bizarre or mis- guided investigator is utterly trivial. What is needed is not a detailed and fine- grained investigation of the concept of an acid; one certainly wouldnt want to devote two thousand years to arguing about the precise contours of the con- cept before ruling out these mistakes and getting on with the real work of studying acids. No such detailed investigation is necessary.
My view is that a proper study of knowledge requires no more-and no less-by way of conceptual analysis than is needed in the chemical case. And I think this marks a significant disagreement with Goldman. As Goldman sees it, there is a highly non-trivial and detailed investigation of our concept of knowledge which is required to do epistemology; as I see it, we will typi- cally require no such investigation at all. Goldman agrees with me that the methods involved in the chemical case are importantly different from those standardly used in philosophy: the hypothetical cases used to probe our intui- tions involve a method which is far more permissive than the methods used in scientific investigations. My view is that philosophy should not use such highly permissive methods, and it would do better to emulate successful sci- ences.
As a result, many of the counterexamples which Goldman offers to my account, involving a deity which doesnt interact with its environment, or Swampman, are ones which my account is not designed to address. We may have intuitions about these cases, although I believe that in many such cases these intuitions are far more variable than they are made out to be, but it is not clear to me that a proper account of what knowledge is needs to be responsive to them. I see such cases as on a par with imaginary cases about
As I see it, individual investigators here must have a certain recognitional capacity in order even to begin: they must be able to recognize at least some samples of the stuff they wish to examine. I dont think that the proper way to understand this is by viewing this recognitional capacity as peculiarly semantic or conceptual, but, as I see it, this is not where the important issue is between Goldman and me. Suppose we say that this is a semantic or conceptual ability. The real issue is Just how substantial the conceptual inves- tigation must be.
430 HILARY KORNBLITH
acids. Just as chemists neednt strive to produce an account of what an acid is which will answer to intuitions which investigators may have in the absence of a real understanding of the phenomena, epistemologists should, I believe, focus on the real phenomenon of knowledge and stop worrying about what their pre-theoretical intuitions say about extremely bizarre, counter- nomologicals cases. Once we view the object of investigation as knowledge itself rather than our concept of it, these cases no longer seem relevant.
I return to this issue in my response to William Talbott.
1.2 Concepts of knowledge and knowledge itself Different individuals, and different societies, may well have different concepts of knowledge. Sameness of concept is not required for sameness of referent, and so even if there is a good deal of variation in the concepts individuals have, they might all, nonetheless, be referring to the very same thing when they use the term knowledge. But suppose, for the sake of argument, that there is a great deal of agreement among individuals in their concepts of knowledge. We might still ask whether this concept is of any epistemologi- cal interest. My own view, unlike Goldmans, is that it is not.
Suppose, as I have argued, that knowledge constitutes a real kind in nature: suppose, that is, that there is a kind of cognitive state which plays an important causal role in nature, and that proper explanations of various natu- ral phenomena must appeal to such states. Our concept of knowledge might, of course, characterize this kind accurately or inaccurately. But if we are inter- ested in knowledge itself, then there is no reason to examine our concept of it. If our concept of knowledge mischaracterizes the real phenomenon, then examining our concept will just elicit misinformation about it. Since our concepts often fail to capture fully and accurately the nature of real phenom- ena, we would do well to look at the phenomena themselves rather than our concepts.
A different possibility is that our term knowledge picks out a kind with an interesting theoretical unity to it, but what unifies the kind is not some natural phenomenon at all; rather, the phenomenon of knowledge turns out to be some sort of social phenomenon. Consider, for example, the property of being married. This is a genuine property, and it confers real causal powers. Married individuals can do various things that unmarried individuals cannot. But unlike the property of having a kidney, the property of being married is one whose very existence is dependent upon certain social structures and
Notice that the Gettier case is quite different from these other two since it is certainly not counter-nomological. The Gettier case thus needs, on my view, to be taken more seri- ously than the others, although having firm intuitions about a case, even about a nomologically possible case, does not assure that ones intuitions accurately capture fea- tures of the kind under investigation.
BOOK SYMPOSIUM 43 1
social conventions. Marriage is an institution which was humanly created, not a feature of the world which was discovered, existing independent of human conventions.
Now if the property of knowing something to be true is a social property rather than a natural property, that would be an interesting thing. (I discuss this further in my response to Martin Kusch.) But it would still not suggest that epistemologists should examine our concept of knowledge rather than knowledge itself. Human beings often have many false beliefs about social kinds, and there are certainly many social institutions which are widely mis- understood. Sociologists who are interested in understanding the nature of social institutions do not merely examine our concepts of those institutions, even if our concepts of them may play an important role in making them the kinds of institutions they are. The fact that our concepts may, in the case of social kinds, play this important role in helping to create the kind itself means that if knowledge is a social kind, then an investigation of our concept of knowledge will be relevant to understanding knowledge itself in a way that it would not if knowledge were a natural kind. It would not, however, suggest that an examination of our concepts could simply substitute for an examina- tion of the phenomenon.
Finally, it might be that if we were to undertake an investigation of the phenomenon of knowledge, we would find not only that it is not a natural kind, but that it is not even an important social kind. It could turn out, at least in principle, that the various things we speak of as exemplifying knowledge turn out to have no interesting properties of any sort in common at all. In this case, I believe, our concept of knowledge would be a historical curiosity, of no epistemological interest at
My own view, of course, is the first of these. Those who reject the view that knowledge is a natural kind may have one of these other views in mind, or they may favor yet a different view. It would be useful, I believe, for those who believe that the concept of knowledge has an important role to play in epistemology to do more than, I believe, has already been done to explain just what that role is.
2. Reply to Martin Kusch
2. I Belief and language I have argued that knowledge is not the exclusive possession of adult human beings; prelinguistic humans, and non-human animals as well, have knowl- edge. Now some philosophers have argued that this is mistaken because lan- guageless animals, on their view, cannot have beliefs, and hence cannot have
Epistemology itself might be of no more than historical interest if this were to be the case, and if a similar fate befell other epistemic terms of appraisal.
432 HILARY KORNBLITH
knowledge. Kusch is on my side here; he agrees that languageless animals may have beliefs. But just as some philosophers allow that languageless animals may have knowledge, but then go on to draw a distinction between human knowledge and what they regard as the distinct and inferior state of mere animal knowledge, Kusch wants to insist, following Hans-Johann Glock? that languageless animals have nothing more than simple beliefs, where these are different in kind from the beliefs of humans. I am deeply sus- picious of attempts to draw such distinctions.
Why do Kusch and Glock believe that the beliefs of languageless animals are different in kind from those of adult human beings? According to Kusch,
First, in the case of animals, belief ascriptions do not create intensional contexts. Second, thought-ascriptions to animals are restricted to perceptible features of the environment. Third, animal concepts are confined to perceptible features of the environment. Fourth, animals are incapable of satisfying one of the two criteria whch we standardly use in attributing concepts to humans. They may be able to apply principles of classification, but not to explain them. And finally, animals beliefs lack the kind of holistic-inferential context that characterizes human linguistic thought. 
Let me take these points in order. Kusch contends that animal belief ascriptions do not create intensional
contexts. I dont see why we should think that this is so. A dog chasing a squirrel might see it run up a tree and come to believe that the squirrel is now in the tree. But even if that particular squirrel happens to be my daughters favorite animal, the dog does not believe that my daughters favorite animal is now in the tree. Just as human belief ascriptions do not allow for the sub- stitution of co-referring terms salva veritate, animal belief ascriptions do not either.
Kusch claims that animal thought ascription is confined to perceptible fea- tures of the environment, and animal concepts are similarly restricted. This is an interesting claim and it would be nice to have some evidence on this issue one way or the other. In the case of pre-linguistic infants, for example, the claim is almost certainly false. As Susan Gelman has argued: there is reason to think that human beings have an innate disposition to regard certain kinds as determined by underlying properties which are only imperfectly correlated with easily available perceptual features; in a word, we have an innate dispo- sition to think of the world as divided into kinds which have Lockean real essences. If this is right, and Gelman offers a great deal of evidence for think- ing that it is, one need not have a language in order to have concepts which go beyond merely perceptual features of objects. It would not be surprising if
Animals, Thoughts and Concepts, Synthese 123(2000), 35-64. I will use this term, following Kusch, for languageless animals. The Essential Child: Origins of Essentialism in Everyday Thought, Oxford University Press, 2003.
BOOK SYMPOSIUM 433
some non-human animals had similar conceptual resources. Kusch and Glock offer no reason to think otherwise, although it must be admitted that this is an empirical issue. What is clear, however, is that there is no reason in prin- ciple why languageless animals may not have concepts and thoughts which go beyond perceptual features.
The suggestion that animals are able to apply principles of classification, but not able to explain them and that this grounds a distinction in kind between human and animal belief simply begs the question against the view that the difference between having a language and lacking one cannot ground a difference between kinds of beliefs. More than this, the suggestion is implau- sible on its face. People who are mute cannot explain principles of classifica- tion either, but this does not impugn the status of their beliefs.
Finally, the suggestion that animals beliefs lack the kind of holistic- inferential context that characterizes human linguistic thought is, I believe, hard to deny. If we attribute the belief that there is a dog in the yard to my neighbor, we may reasonably conclude that she also believes that there is a mammal in the yard; not so if we attribute the same belief to another dog. What this shows, however, is not that the beliefs of languageless animals are not holistically integrated with their other beliefs, but that the inferential connections they make are often quite different from those we would make. The same is true, of course, when we attribute beliefs to individuals who have theories quite different from our own, who live in times and places dif- ferent from ours, and who speak languages different from our own. This does not undermine attribution of beliefs to others different from us; it merely means that we must be careful in drawing any conclusions from such attribu- tions.
I do not wish to deny that having a language allows us to have beliefs which we could not have without one. This does not give us a reason, how- ever, for thinking that the beliefs of languageless animals are different in kind from those of animals who do have a language. There is a single kind of psy- chological state which humans and other animals both have, a kind of state which plays a crucial and characteristic role in our mental economy.
It is also unclear to me why, even if Glock and Kusch were right about this, it should be taken to show that animal belief is different in kind from human belief. There is no ques- tion, of course, that humans may have beliefs about things of which non-human animals will forever remain ignorant. But some non-human animals have sense organs very dif- ferent from any human, thereby affording them a kind of epistemic access to features of the world which we simply lack, and surely this cannot serve as a ground for claiming that the beliefs of these animals are different in kind from those of human beings. Mere difference in subject-matter or process of belief acquisition is not sufficient to ground a difference in kind.
434 HILARY KORNBLITH
2.2 Reflection and cognitive improvement
Some philosophers have argued that knowledge requires a certain sort of reflection, reflection on the epistemic merits of ones belief. Thus, on this sort of view, if I am to know that p , I must reflect on my belief that p to see whether it meets certain requirements: on one version, I must introspect to see whether p coheres with my other beliefs; on another, I must reflect on my beliefs to see whether p meets certain foundationalist standards. I argue that this reflective requirement is not well motivated. When agents reflect in this sort of way, they are extremely likely to come to believe that the belief in question passes the indicated test regardless of the manner in which the belief was originally acquired and regardless of the reasons for which the belief is retained. Reflection of this sort is ill-suited to serve as a check on the charac- ter of our belief acquisition: instead of correcting our errors, it has a tendency to produce greater self-confidence, but no greater accuracy.
In response, Kusch argues,
Komblith exaggerates the bad news from the experimental study of human reasoning. The psychological literature on this topic does not paint as dark a picture as he alleges. Indeed, the general trend in the literature on this topic seems to go towards a re-evaluation of the pessi- mismthat characterized the famous work of the seventies ... Moreover, let us not forget that cognitive illusions do not just occur in the realm of self-reflection. They occur just as well in our theorizing about worldly events. That is to say, if the existence of cognitive illusions is an argument against reflection, then it is, by the same token, also an argument against the empiri- cal study of nature. 
I did not argue, however, that introspection is in general unreliable, or that human inference is, in general, unreliable; such a suggestion, would, as Kusch rightly points out, have broad skeptical implications. Rather, what I did argue is that reflection is ill-suited for certain purposes: the manner in which introspection works makes it particularly ill-suited for the purpose of error detection. If we insist nevertheless that agents reflect on the epistemic credentials of their beliefs in order to be credited with knowledge, we will be insisting that they engage in a process which does not tend to improve the reliability of their belief acquisition; such reflection tends to produce more self-confident, but no less misguided, believers. Requiring agents to reflect on the epistemic merits of their beliefs makes sense if one assumes that it will improve their reliability; once one understands its actual effects, however, the suggestion that it is a necessary condition for knowledge loses all plausibil- ity.
2.3 Communitarian epistemology Some have argued that knowledge is, in some sense or other, essentially social. Knowledge requires that one be part of a linguistic community, on certain views, or that one be part of a community which participates in the
BOOK SYMPOSIUM 435
practice of giving and asking for reasons. I argue that knowledge has no such social requirements.
Kusch favors what he calls a communitarian epistemology, and the social requirements he favors are motivated by Wittgensteins famous rule- following arguments. Wittgenstein argues that the very idea of following a rule presupposes embedding in a social context: Hence it is not possible to obey a rule privately: otherwise thinking one was obeying a rule would be the same as obeying it. Epistemic normativity, and normativity generally, on this view, is essentially social.
My response to this is certainly not original. What I argue is that the dis- tinction between thinking one has obeyed a rule and actually obeying it may be motivated simply through reflection on ones own past performance. This does not guarantee, of course, that one never makes mistakes, but relying on ones community for correction does not guarantee immunity from error either. If the worry then is that without a community, one may fail to notice departures from a rule, then the move to the community will not solve the problem. If on the other hand, the original worry was simply that without a community, there is no motivation for the distinction between seeming to apply a rule and actually applying it, then the move to the community is not necessary, since reflection on ones past performance is motivation enough for the distinction. Either way, the community plays no essential role here. Social practices are not the root of normativity.
Kusch expresses dissatisfaction with this response, and he insists that, Communitarians do not reason in the way [Kornblith] assumes. [416- 171 But I do not follow what Kusch says about the way communitarians are supposed to reason.
It is the comparison between different people that purport to follow the same rule that creates the conceptual space for the distinction between getting the rule right and merely believing to get the rule right. Note that the point of the reference to others is not to establish what is cor- rect. What is at issue here is not a criterion of rightness but a condition of the intelligibility for attributing rule-following . Communitarians do not say that the individual is right if, and only if she does what the others do. Rather, communitanans insist that in order to make sense of an individual being right we must make reference to something that is external to the individual. And these external conditions are the actions and judgements of others. 
So Kusch rejects, rightly, the idea that if a community believes I am follow- ing a rule correctly then I am following the rule correctly. Instead, the move to the community creates the conceptual space for the distinction between getting the rule right and merely believing to get the rule right. But why is the community needed for this? Why cant reflection on my own past prac- tice-seeing how I have made errors in the past-make conceptual space for the distinction at issue? Kusch doesnt say.
Philosophical Investigations, Blackwell, 1958, section 202.
436 HILARY KORNBLITH
I suggest a number of different ways in which social practices might be at the root of epistemic normativity, and ways in which knowledge might be essentially social; I present arguments against each of them. What I dont see is how Kuschs communitarian view eludes those arguments.
3. Reply to William Talbott
3.1 Is epistemology parochial? I defend the view that knowledge is a natural kind. Just as we discover what gold is by examining putative samples of gold to determine what most of them have in common, I argue that we should undertake a comparable empirical investigation of the phenomenon of knowledge. On this view, Tal- bott argues, epistemology becomes parochial. ... I believe that the main dif- ference between epistemology and the sciences is that, while the sciences are expected to be parochial (true of this world, not of all possible worlds), epis- temology should not be.  Talbott suggests that I see epistemology dif- ferently:
What is knowledge? Here is Kornbliths answer to this question: Knowledge is reliably pro- duced true belief. (58) He does not claim this is a necessary truth, true in all possible worlds. It is enough if knowledge is a natural kind in this world and Kornbliths account of it is true in this world (and other similar worlds). 
But the move from the claim that epistemology is an empirical discipline to the claim that a correct epistemology would consist of contingent rather than necessary truths is one I do not endorse. The distinction between the empiri- cal and the a priori should not be identified with the distinction between the contingent and the necessary.
Consider Talbotts claim that the sciences, unlike epistemology, deal with parochial truths. There can be no doubt that some of the truths which science discovers are parochial in Talbotts sense: that is, they are contingent. It is a contingent fact that the earth has a single natural satellite, and our knowledge of this fact is clearly empirical. But it was also an empirical discovery that water is H,O, and this is a necessary truth. Some of the empirical discoveries which are made in the sciences are not parochial at all. Just as the sciences empirically discover the essential properties of natural kinds such as water, an empirical investigation of the phenomenon of knowledge may discover the very features which make it what it is. On my view, a proper account of knowledge would give us an account of its necessary properties, just as a proper scientific account of the nature of water properly identifies it as H,O. And just as water is a natural kind in every world if it is a natural kind in this world, knowledge is a natural kind in every world if it is a natural kind in this world.
BOOK SYMPOSIUM 437
So my view is not the one which Talbott attributes to me. I believe that epistemology is an empirical discipline which may discover necessary truths about knowledge.
3.2 Universal concepts
Talbott suggests that the central concepts of a proper epistemology are universal, and he questions whether my own conception of epistemology would allow such a result. What does it mean to say that a concept is univer- sal?
In Putnams Twin Earth examples and in related examples from Burge we see that the content of many concepts, including natural kind concepts, exhibit a kind of parochiality. Our concept water picks out H,O, the Twin Earthians concept water picks out XYZ. But it seems to me that no such example can be constructed for the concept true .... Let us say that a concept that is not susceptible to different extensions in Twin Earth cases is universal. 
But this way of looking at things is, I believe, confused. Consider the case of water. As the Twin Earth story is typically told, our word water refers to H20, while when Twin Earthians use the term water, they refer to XYZ. What should we say about concepts? The story about Twin Earth has it that people on earth associate the idea of clear, colorless, liquid found in our lakes and rivers with the term water, and Twin Earthians associate the same idea with their uses of the term. Thus, although we associate the same ideas with the term water-we have the same concept, if you will-our uses of the term refer to different kinds. This much is unobjectionable.
But the traditional story involves a fair bit of idealization. Different indi- viduals no doubt associate different ideas with the term water. Here on earth and in the current year, rather than 1750, many individuals will associate the idea of H20 with the term water; others will associate some description more like the one in the traditional story. But, and this is part of the point of the story Putnam and Burge tell, the idea one associates with the term-the concept of water-does not serve to determine the reference of the term; indeed, the descriptions one associates with the term may not even be true of the referent of the term. Thus, within a given language community all of whom use the term water to refer to a given kind, there may be many differ- ent concepts of water-many different ideas associated with the term. It is thus a mistake to talk about the concept of water in a given language community.
It is equally a mistake to talk about the concept of truth, justification, reliability or knowledge, as Talbott does. Different individuals, right here on earth, are likely to have quite different epistemological concepts, even if their epistemological terms co-refer. I argue, in the first chapter of Knowledge and its Place in Nature, that it is a mistake to try to analyze ones ordinary, pre- theoretical concept of knowledge, because ones concept of knowledge may be
438 HILARY KORNBLITH
indicative of gaps in ones understanding of the phenomenon; ones concept of knowledge may also include substantive mistakes about knowledge, just as early investigators concepts of acid, for example, included substantive errors about it. I argue that little is to be gained in epistemology by examining our concepts. We should be looking at the phenomenon of knowledge itself, rather than our concept of it, if we wish to make any progress in epistemol- ogy.
Some individuals will have epistemic concepts which are universal, in Talbotts sense; others will not. But it makes no sense to talk about whether the concept of knowledge, for example, is universal, since different indi- viduals will have different concepts of it even when they use the term to talk about one and the same thing.
3.3 Knowledge as natural kind
Talbott argues that knowledge is not a natural kind, or at least that the term knowledge does not secure reference in the way in which natural kind terms do. As Putnam points out, if we were to discover that the objects we call cats turned out to be remote-controlled robots from Mars, we would have to conclude that cats are not animals; we would have to change our beliefs about the kind of objects cats are. But, Talbott argues,
... the discovery that cats were remote-controlled robots would not change our beliefs about the nature of knowledge. Before the discovery, Kornblith and I would have believed cats had knowledge. The discovery that they were remote-controlled robots would make it evident that they dont know anything. What is anchoring our concept of knowledge in this case? The other presumed instances in the actual world? To see that they are not, suppose you were to discover that all supposedly sentient beings on earth (except you) were robots remotely controlled from Mars. If you made this discovety, you might well conclude that you were not a human being, because you were not of the same kind as all those other things you took to be human beings. However, you could not reasonably think that the robots all had knowledge and you did not. Although far-fetched, it is possible to imagine that you are the only being on earth with knowl- edge. This sort of thought experiment shows, I believe that the nature of knowledge is not determined the way a natural kind is because the concept of knowledge is not parochial in the way that the concept of cat or human being is. [425-261
This is, I believe, an interesting case. Let me introduce two further imagi- nary cases. In the first case, let us imagine that the term metal was intro- duced into English by way of the following baptismal event: someone stand- ing over a large variety of samples of copper, aluminum, gold, silver, platinum and a variety of other substances intones, Let me call these sub-
The robot-cats first made their appearance, I believe, in It Aint Necessarity So, reprinted in Putnams Mathematics, Mutier and Method: Philosophical Papers, vol. 1, Cambridge University Press, 1975; the paper was originally published in 1962. Many of htnams other papers on this issue are reprinted in Mind, Language and Reality: Phi- iosophicaf Papers, vol. 2, Cambridge University Press, 1975.
BOOK SYMPOSIUM 439
stances and others relevantly similar, metals. I will suppose that our bap- tizer has noted some important similarities among these substances in the course of an investigation of their physical properties, and, although he has made some errors, and the substances in front of him include some that we would not now regard as metals, we may, nonetheless, take this individual to be using the term metal to refer to the very same natural kind which we refer to when we use that word. This is, I take it, the standard sort of story which Putnam, for example, tells about the introduction of natural kind terms. A term may refer to a kind even though the samples which were ini- tially baptized using the term were not all members of the kind.
Now consider a second case. Once again, an individual engaged in an in- vestigation of the physical properties of various substances comes to note certain striking similarities among a number of them, and he decides to intro- duce a term to characterize those substances which have the relevant similari- ties. As in our first case, if you were to ask this investigator just what the relevant similarities are, he would make a number of mistakes. But now for the difference. Just before our investigator introduces the new term, a prank- ster in his laboratory replaces all but one of the samples he has been examin- ing with cleverly disguised replicas, all of which are made of wood. Once again, the investigator points at the objects on his workbench and intones, Let me call these substances and others relevantly similar, metals. When our investigator leaves for the day, the prankster removes the wooden repli- cas, and puts the samples of metal back on the bench where they had been.
I dont believe that in this second case we should conclude that the newly introduced term metal refers to wood, even though the majority of the items ostended were wood rather than metal. By the same token, I dont believe that this kind of case shows that metal is not a natural kind term, or that the nature of metals is not determined by features of local samples with which the baptizer interacts. And I dont believe that Talbott would disagree with me about any of this. Talbott seems to accept the main lines of Putnams account of natural kinds and natural kinds terms. Precisely how such an account should be couched to accommodate this sort of example raises an interesting question, but Talbotts example is not designed to show that there is a problem for Putnams account of natural kinds and natural kind reference; it is designed to show that terms such as knowledge do not work in the same way as natural kind terms do. Talbotts example, however, is very simi- lar to my second example involving the term metal. So it cant be taken to show that knowledge works in a way different from paradigmatic natural kind terms.
Putnam is at pains to point out that baptizers may have many false beliefs about the kinds they baptize. A proper understanding of the nature of a lund may come about as the result of extended investigation, but it is not a p e q -
440 HILARY KORNBLITH
uisite for successful reference. At the same time, not all false beliefs are equal. In some cases, if we discover that a baptizer has a false belief about many of the samples used to introduce a kind term, this should be taken to show that he has false beliefs about the kind baptized; this is the case in the first of my two examples. In other cases, certain kinds of mistakes about the samples present during a baptismal ceremony will show that they did not even play a role in fixing the reference of the term introduced, even assuming that the term was successfully introduced; this is the case in the second of my two examples, where the prankster has surreptitiously replaced the intended objects of ostension. Some will think that this suggests a descriptive com- ponent in a proper account of reference; others will think that it merely requires a more complicated non-descriptive account. But we need not settle that issue here. Talbotts example does not show that knowledge secures its reference in any way different from paradigmatic natural kind terms, nor does i t show that knowledge is not a natural kind.
BOOK SYMPOSIUM 441