SyntheseDOI 10.1007/s11229-015-0751-z

Conceptual analysis and natural kinds: the caseof knowledge

Joachim Horvath1

Received: 5 December 2014 / Accepted: 12 April 2015 Springer Science+Business Media Dordrecht 2015

Abstract There is a line of reasoning in metaepistemology that is congenial to natu-ralism and hard to resist, yet ultimately misguided: that knowledge might be a naturalkind, and that this would undermine the use of conceptual analysis in the theory ofknowledge. In this paper, I first bring out various problems with Hilary Kornblithsargument from the causalexplanatory indispensability of knowledge to the naturalkindhood of knowledge. I then criticize the argument from the natural kindhood ofknowledge against the method of conceptual analysis in the theory of knowledge. Anatural motivation for this argument is the following seemingly plausible principle: ifknowledge is a natural kind, then the concept of knowledge is a natural kind concept.Since this principle lacks adequate support, the crucial semantic claim that the con-cept of knowledge is a natural kind concept must be defended in some more directway. However, there are two striking epistemic disanalogies between the concept ofknowledge and paradigmatic natural kind concepts that militate against this seman-tic claim. Conceptual analyses of knowledge are not affected by total error, and theproponents of such analyses are not subject to a priori conceptual obliviousness. Iconclude that the argument from natural kindhood does not succeed in underminingthe use of conceptual analysis in the theory of knowledge.

Keywords Conceptual analysis Natural kinds Knowledge Homeostatic cluster view Natural kind concepts Hilary Kornblith

B Joachim Horvathjoachim.horvath@uni-koeln.de

1 Department of Philosophy, University of Cologne, Richard-Strauss-Str. 2,50931 Cologne, Germany

123

Synthese

1 Introduction

There is a certain temptation in metaepistemology that is congenial to naturalismand hard to resist, even though it is ultimately misguided. This temptation can bebrought out as follows. Suppose you are an epistemologist working on the theory ofknowledge and God tells you one day that knowledge is a natural kind. Your reactionto this divine revelation might be the following: All these years of hard armchair workon the analysis of knowledge, all this wrestling with tricky cases and counterexamples,all these countless refinements and improvements of my analysis of knowledge: it wasall a giant waste of time! I should have gotten out of the armchair a long time ago andstudied knowledge just like any other empirical phenomenon. Even though it wouldbe tempting to react in this way (see, e.g., Heller 1996, p. 335; Kornblith 2007, p. 47;Kumar 2014, p. 442; Ludwig 2013, p. 232), this temptation should nevertheless beresisted, as I will argue in this paper.

The philosopher who has given in to this temptation more than any other philoso-pher is Hilary Kornblith (cf. Kornblith 1999, 2002, 2007). The main point of his bookKnowledge and Its Place in Nature (2002) is to argue for the radical metaepistemologi-cal claim that knowledge is a natural kind. According to Kornblith, knowledge shouldtherefore be investigated with the methods of empirical science, and not by meansof armchair conceptual analysis or intuitions. Even though Kornblith also advancesother objections to conceptual analysis in epistemology (cf. Kornblith 2007, 2013),the argument from the natural kindhood of knowledge arguably takes center stagein his particular brand of naturalized epistemology. He summarizes the argument asfollows in the final chapter of Knowledge and Its Place in Nature:

I have been urging that knowledge is a natural kind and thus that a properunderstanding of the nature of knowledge requires a certain sort of empiricalinvestigation. It is a mistake to investigate our intuitions about knowledge or ourconcept of knowledge because these may be importantly incomplete or impor-tantly mistaken or both. (Kornblith 2002, p. 163; my emphasis)1

As it stands, the argument challenges any kind of apriorism in the theory of knowledge,and not just the use of conceptual analysis or intuitions, because Kornblith draws thevery general conclusion that the natural kindhood of knowledge requires a certain sortof empirical investigation. In this paper, I want to address the more specific questionwhether the natural kindhood of knowledge undermines the use of conceptual analysisin the theory of knowledge. I will conclude that the metaphysical claim that knowledgeis a natural kind does not by itself undermine the methodological claim that conceptualanalysis is an adequate method for analyzing knowledge. To challenge the latter claim,one must focus on the concept of knowledge instead, and argue that it is a natural kindconcept like water or gold.2 However, the prospects for such an argument are dim,because knowledge behaves very unlike paradigmatic natural kind concepts.

1 Other proponents of naturalized epistemology tend to concur. For example, Victor Kumar claims: Ifknowledge is a natural kind, then the satisfaction conditions for knowledge cannot be discovered througharmchair reflection of the sort that is characteristic of traditional conceptual analysis. (Kumar 2014, p. 442).2 I follow the usual convention of indicating reference to concepts with small caps.

123

Synthese

2 The argument from the natural kindhood of knowledge

The standard view in the theory of knowledge is that conceptual analysis aims at ananalysis of knowledge in terms of an illuminating3 set of individually necessary andjointly sufficient conditions for knowledge,4 of the kind that potentially5 reveals theessence or nature6 of the property or kind knowledge.7 Proposed analyses of knowl-edge are standardly expressed by (metaphysically) necessary biconditionals, such asnecessarily, something is knowledge if and only if it is a justified true belief (cf.Williamson 2007; Malmgren 2011). If successful, the argument from natural kindhoodwould undermine the status of conceptual analysis as a proper method for seeking ananalysis of knowledge in this sense.

Let us now make this argument more explicit:(1) Knowledge is a natural kind.(2) If knowledge is a natural kind, then illuminating necessary and sufficient condi-

tions for knowledge can only be figured out with empirical methods.(3) Conceptual analysis is not an empirical method.(C) Illuminating necessary and sufficient conditions for knowledge cannot be figured

out with conceptual analysis.Since the argument is clearly valid, let us therefore consider its premises. The least con-troversial premise should be (3), that conceptual analysis is not an empirical method.Indeed, the method of conceptual analysis is typically seen as a paradigm of an a priorimethod, and given that a priori methods are standardly understood as non-empiricalmethods, premise (3) simply follows from these widely held assumptions. Even thoughthere are a few dissenting voices (cf. Micevic 2000, 2005; Schwitzgebel 2008), I willsimply take premise (3) for granted in the following.

3 This qualification is supposed to rule out circular or irrelevant necessary conditions, such as the conditionof being self-identical or being such that 2 + 2 = 4. In fact, I think that an adequate account of philosophicalanalysis requires a more substantial and specific condition than the fairly vague requirement of being illumi-nating (cf. Horvath, manuscript). But for our present purposes, this condition should work reasonably well.4 For example, in his An Introduction to Contemporary Epistemology Matthias Steup writes: For ananalysis to be correct, the analysans must specify conditions that are individually necessary and jointlysufficient for the analysandum. [] For an analysis to be successful, it must be illuminating [] (Steup1996, pp. 2728).5 Why this hedged formulation? Because the connection between a philosophical analysis and the essenceof the relevant property or kind is less straightforward than is commonly assumed (cf. Horvath, manuscript).In particular, the nature of this link depends on ones general metaphysical commitments concerning themetaphysics of properties and essences. For example, if one takes properties to be mere sets of possibilia(cf. Lewis 1986, Chap. 1.5), then all necessarily co-instantiated properties will be identical, and thus anecessary biconditional that states necessary and sufficient conditions for X will effectively just tell usthat a certain property Xwhich can be expressed in at least two different waysis necessarily self-co-instantiated. However, such general truths about properties arguably do not belong to the essence of anyparticular property, and thus the necessary biconditional in question would not reveal the specific nature ofthe property of being X (cf. Fine 1995). In this case, an analysis of X would be more like an informativeidentity claim, such as Hesperus is Phosphorus, than like a claim about the essence of being Xbecausebeing the set of all possible Xs might already exhaust the latter.6 I use the terms essence and nature interchangeably in this paper.7 I indicate reference to properties or kinds with italics, and I will mostly gloss over the difference, if any,between properties and kinds (unless explicitly noted otherwise).

123

Synthese

3 Knowledge as a natural kind?

What can be said in favor of premise (1) of the argument from natural kindhood, i.e.,the claim that knowledge is a natural kind? Let us consider how Kornblith (2002,Chap. 2) argues for this seemingly radical thesis.

Kornbliths key move is to point out that knowledge plays a robust explanatoryrole in cognitive ethology, which is a branch of behavioral biology. On the basis ofbehavioral evidence about animals such as ravens or chimpanzees, Kornblith arguesfollowing many cognitive ethologiststhat the ascription of knowledge, instead ofmere true belief, is often indispensable for explaining sophisticated forms of animalbehavior, for example, co-operative hunting behavior in ravens (cf. Kornblith 2002,p. 31). The key idea is that the animals in question could not behave in the way theydo unless certain knowledge-states were among the causally relevant antecedents oftheir behavior.

How do these considerations from cognitive ethology support the claim that knowl-edge is a natural kind? On Kornbliths preferred account of natural kinds, somethingis a natural kind just in case it is a homeostatic cluster of properties,8 i.e., a cluster ofproperties that is mutually supporting and reinforcing in the face of external change(Kornblith 2002, p. 61). Such homeostatic clusters of properties display a degree ofcausal stability that is not found in just any random collection of properties (ibid.).Therefore, homeostatic clusters are able to support various inductive inferences or nat-ural (causal) laws. They also explain the characteristic surface-properties of naturalkinds, such as the liquidity of water, which is explained by the homeostatic charac-ter of H2O molecules and their causal interactions with other H2O molecules underordinary conditions of pressure and temperature. Natural kinds are thus understoodas particularly stable nodes in the causal network of the world, i.e., they are individu-ated in causalexplanatory terms. Given that Kornblith subscribes to a causal view ofnatural kinds, it makes sense to identify natural kinds and their essential features bythe role they play in our best causal explanations. And given that knowledge seems toplay an indispensable role in the causal explanation of sophisticated animal behavior,it thus makes sense to conclude that knowledge is a natural kind.

Kornbliths key empirical claim that knowledge is an indispensable factor in causalexplanations is open to various objections (see, e.g., Pernu 2009), but I do not wantto take issue with this part of his argument here. For the sake of the argument, I willassume that he is completely right about that. What seems more committal from aphilosophical point of view is the crucial inference from playing an indispensablecausalexplanatory role to being a natural kind. The main justification for thismove is the assumption, adopted via Boyds (1988,1991) homeostatic cluster accountof natural kinds, that natural kinds are individuated in causal terms.

However, the homeostatic cluster view has a striking feature that makes it especiallyproblematic in the context of a methodological argument against conceptual analysis.On the homeostatic cluster view, natural kinds are causally stable clusters of properties.Since causal stability is a matter of degree, it can be more or less perfect. As a result,

8 The account is mainly developed in Kornblith (1993), and it is basically a version of Richard Boydsaccount of natural kinds (cf. Boyd 1988, 1991).

123

Synthese

there will be instances of natural kinds that lack some of the properties in the relevantcluster, e.g., borderline instances of biological species. The consequence is that theproperties in a homeostatic cluster do not specify necessary and sufficient conditionsfor membership in the relevant kind. Here is how Boyd sums up this point:

The natural definition of one of these homeostatic property cluster kinds is deter-mined by the members of a cluster of often co-occurring properties and by the(homeostatic) mechanisms that bring about their co-occurrence. [] In casesof imperfect homeostasis in which some of the properties in the cluster are absentor some of the mechanisms inoperative it will sometimes happen that neithertheoretical nor methodological considerations assign the object being classifieddeterminately to the kind or to its complement, with the result that the homeosta-tic property-cluster definition fails to specify necessary and sufficient conditionsfor kind membership. (Boyd 1991, pp. 141142)

Recall that the standard goal of conceptual analysis is to come up with an analysisin terms of illuminating necessary and sufficient conditions. Someone who relies ona homeostatic cluster view of natural kinds, like Kornblith, would thus be unable tocontribute to that goal, for reasons that have nothing specifically to do with the rejec-tion of a priori methods. For, if knowledge were a homeostatic cluster kind, then nomethod could possibly deliver a set of illuminating necessary and sufficient condi-tions for knowledge, simply because knowledge would not be the sort of thing thathas illuminating necessary and sufficient conditions. Therefore, it seems that Kornblithwould have to reject conceptual analysis not primarily because of its a priori character,but rather by (implicitly) rejecting its very goal of specifying illuminating necessaryand sufficient conditions.9 However, this is not how Kornblith actually argues in thepassage quoted above, where he seems to hold on to the standard goal of conceptualanalysis and merely objects to its non-empirical character. To reject both the standardgoal (illuminating necessary and sufficient conditions) and method (conceptual analy-sis, intuition) of analyzing knowledge would make Kornbliths naturalism even moreradical than it already is.

Given Kornbliths own emphasis on the rejection of conceptual analysis qua apriori method, it would thus seem fitting to adopt the more orthodox conception ofnatural kinds by Putnam (1975) and Kripke (1980). This conception allows for thediscovery of a posteriori identities, such as water is H2O or gold is the element withatomic number 79, that specify the underlying microstructure of the natural kindsin question.10 Since a posteriori identities of this sort entail illuminating necessary

9 One might object that conceptual analysis could also pursue the weaker goal of providing a clusteranalysis, e.g., in the sense of Searle (1969, Chap. 7), which would seem to be compatible with thehomeostatic cluster account of natural kinds (thanks to an anonymous reviewer for raising this issue).The basic idea is that something only needs to satisfy sufficiently many (but not all) of the analyzingfeatures that a cluster analysis of some category K specifies in order to qualify as an instance of K. From amethodological point of view, however, this can only be seen as a highly revisionary proposal, at least withrespect to the category of knowledgeit certainly does not reflect how most epistemologists conceive oftheir own attempts at analyzing knowledge.10 In fact, Kornblith explicitly endorses the basic contours of the PutnamKripke conception of naturalkinds and natural kind concepts (Kornblith 2002, pp. 1213, fn. 17 & 18).

123

Synthese

biconditionals such as necessarily, something is water iff it is H2O, they do allow forpursuing the traditional goal of conceptual analysis, i.e., to provide illuminating neces-sary and sufficient conditions, with empirical meansunlike the homeostatic clusterview. In other words, the PutnamKripke conception of natural kinds would enableKornblith to only reject conceptual analysis as an appropriate method for theorizingabout knowledge, while holding on to the standard goal of providing illuminatingnecessary and sufficient conditions for knowledge (although on a scientific basis andnot through some form of armchair analysis).

Adopting the PutnamKripke conception would, however, threaten to undermineKornbliths argument for the claim that knowledge is a natural kind, because thisargument crucially depends on the inference from playing an indispensable causalexplanatory role to being a natural kind. Since the PutnamKripke conception is notcommitted to a causal individuation of natural kinds, it does not by itself support the keyinference from having a robust causalexplanatory profile to being a natural kind.On this conception, something that clearly is a natural kind, like water or gold, mightbe causally inert in some other possible world, e.g., in a world with very different lawsof nature, while something that has a robust causalexplanatory profile in the actualworld, like knowledge (if Kornblith is right), might nevertheless fail to be a natural kind.

But maybe Kornbliths argument for the natural kindhood of knowledge does notrequire such a tight connection between the causalexplanatory role of knowledge andthe causal individuation of natural kinds. Maybe the fact that a given kind K has arobust causalexplanatory profile should rather be seen as a fallible criterion for thenatural kindhood of K. Understood in this way, Kornbliths argument would merelyrequire that the robust causalexplanatory profile of knowledge makes the naturalkindhood of knowledge sufficiently likely, without actually entailing that knowledgeis a natural kind.

Yet given the orthodox PutnamKripke conception, which does not claim any con-stitutive link between causal efficacy and natural kindhood, it seems difficult to moti-vate even a merely probabilistic relation between having a robust causalexplanatoryprofile and being a natural kind. For this would apparently require that most of thecategories that have a robust causalexplanatory profile also happen to be natural kinds.Since the total number of categories that have a robust causalexplanatory profile isvast, maybe even infinite, such a claim is difficult to evaluate. But on the face of it,many categories that have a robust causalexplanatory profile are not happily classifiedas natural kinds. Think, for example, about the many causalexplanatory categoriesfrom the social realm, such as money, power, citizenship, or military force, or aboutthe robust causalexplanatory profile of many artifactual kinds, such as key, hammer,screwdriver, or lawnmower. In fact, even some mathematical categories, which typi-cally do not figure on anyones list of natural kinds, have a robust-causal explanatoryprofile, as I will argue in the following section for the case of primeness. So arguably,there is a large number of prima facie counterexamples to the inference from havinga robust causalexplanatory profile to being a natural kind, and this challenges theclaim that most of the categories that have a robust causalexplanatory profile are nat-ural kinds. For this reason, the idea that having a robust causalexplanatory profile is avalid probabilistic criterion for natural kindhood does not seem very promising either.

123

Synthese

In sum, it is hard to see how Kornbliths argument from the indispensable causalexplanatory role of knowledge in cognitive ethology to the natural kindhood ofknowledge can be supported in the intended way. On the one hand, the argumentfrom the causalexplanatory role of knowledge to the natural kindhood of knowledgecrucially relies on the homeostatic cluster view of natural kinds. Yet this view under-mines the standard goal of analyzing knowledge in terms of illuminating necessary andsufficient conditions, and so the argument aims too broadly. Opting for the orthodoxPutnamKripke conception of natural kinds would help to avoid the latter problem,but this conception fails to provide a metaphysical link between causalexplanatoryindispensability and natural kindhood. Absent such a link, however, it seems difficultto support the crucial inference from having a robust causalexplanatory profile tobeing a natural kind.

4 Natural kindhood and conceptual analysis

How should we assess premise (2) of the argument from natural kindhood? That is,how should we assess the conditional if knowledge is a natural kind, then illuminat-ing necessary and sufficient conditions for knowledge can only be figured out withempirical methods?

Here is an initial reason to be skeptical about premise (2). The claim that knowledgeis a natural kind is a metaphysical claim about knowledge, and not a claim about ourepistemic relation to knowledge, and also not a claim about our concept of knowledge.And why should a particular view about the metaphysics of knowledge have anyspecific implications for the epistemology or semantics of knowledgein this case:negative implications for conceptual analysis? The metaphysical status of individualobjects, for example, does not have any specific implications for the analysis of propernames, and the metaphysics of spacetime does not suggest any particular view aboutthe meaning of indexicals like here or now. Why should this be otherwise in caseof the metaphysics of knowledge and the meaning of the word knowledge, or thecontent of the concept knowledge?

One suggestive answer might be: if knowledge is a natural kind, then knowledgemust be a natural kind concept, and natural kind concepts are not a priori analyz-able in terms of illuminating necessary and sufficient conditions.11 According to the

11 Note that the converse argument from knowledge is a natural kind concept to knowledge is anatural kind clearly fails, because a natural kind concept may fail to pick out any kind at all, as in the caseof phlogiston, or it may pick out a disjunctive, non-natural kind, as in the case of jade. One might objectthat it is not clear what exactly makes phlogiston or jade a natural kind concept in the first place, giventhat they actually fail to pick out a natural kind (thanks to an anonymous reviewer for raising this issue).First, of course, what makes them natural kind concepts is the fact that they are equally nondescriptive asparadigmatic natural kind concepts, such as mass or gold. But since nondescriptiveness is not sufficientfor being a natural kind conceptgiven that one can even refer to a non-natural kind like bachelorhoodwith a nondescriptive concept (see below in the main text)there must be some further reason why it islegitimate to regard phlogiston and jade as natural kind concepts. A plausible suggestion would be thatthe way the concepts phlogiston and jade were introduced is completely analogous to the way certainparadigmatic natural kind concepts were introduced, and this, together with their nondescriptiveness,suffices to regard them as natural kind concepts. With some amount of idealization, we can say that,for example, the concept phlogiston was introduced to refer to a certain natural phenomenon that was(incorrectly) thought to play a particular theoretical role in chemistryjust like the concept mass was

123

Synthese

standard view, natural kind concepts are nondescriptive concepts12 that are not con-stitutively associated with any descriptive features or inferential relations (cf. Kripke1980; Putnam 1970, 1975; Soames 2002).13 Therefore, we need not have any explicitor implicit representations of illuminating necessary and sufficient conditions in orderto possess a natural kind concept. And the representations that we have can be highlymisleading with respect to necessary and sufficient conditions for kind membership.Even subjects who competently possess a natural kind concept, such as water orgold, are therefore prone to all kinds of ignorance and error concerning necessaryand sufficient conditions for membership in the relevant kind. Apparently, this is whatKornblith has in mind when he suggests that our concept of knowledge [] may beimportantly incomplete or importantly mistaken or both (see quote above).

However, it does not follow from the fact that a kind K is a natural kind that any con-cept of K must be a natural kind concept. Gold is a natural kind, but the element withatomic number 79 is a descriptive concept of gold (see also Ludwig 2013, p. 233).Of course, analyzing the concept the element with atomic number 79 only pro-vides us with the rather trivial necessary biconditional something is the element withatomic number 79 iff it is the only thing that is an element and has atomic number 79.

Footnote 11 continuedintroduced to refer to a certain natural phenomenon that was (correctly) thought to play a particulartheoretical role in physics. And again with some amount of idealization, we can say that the concept jadewas introduced by ostension to particular instances of jade, with the (unsuccessful) intention of referringto that kind of stuff or natural kind (depending on the conceptual sophistication of those who introducedthe concept)just like the concept gold was introduced by ostension to particular instances of gold, withthe (successful) intention of referring to that kind of stuff or natural kind (see also Soames 2007).12 According to some accounts of natural kind concepts, they must at least have a minimal descriptive corein order to solve the so-called qua-problem (see, e.g., Devitt and Sterelny 1999). For example, one mighttry to fix the reference of the concept tiger to the natural kind tiger by ostension to actual tigers, whichis one important way to fix the reference of natural kind concepts. But then it may still be indeterminatewhether tiger refers to tigers, animals, living beings, or material objects. For this reason, natural kindconcepts may need a certain minimum of descriptive features in their content, such as being an animal. Asa consequence, one may come to know certain trivial facts about tigers merely on the basis of analyzing theconcept tiger, e.g., that tigers are animals. This is still a far cry from illuminating necessary and sufficientconditions for being a tiger, however.13 Natural kind concepts are also standardly regarded as rigid designators (cf. Kripke 1980), i.e., asconcepts that have the same referent in all possible worlds (where they have a referent at all). However, itis an open question whether the notion of rigiditywhich was primarily developed for singular termscan also be extended to general terms or general concepts, like water, gold, or bachelor (cf. Besson2010; Schwartz 1980, 2002; Soames 2002, Chap. 9). For example, if one identifies the reference of generalconcepts with their extension, then the reference of most natural kind concepts will clearly not be the samein all relevant possible worlds. There surely could have been, e.g., more or less water or gold than thereactually is, so the extension of water and gold changes across possible worlds. What if one identifies thereference of general concepts with the relevant property or kind instead, e.g., with the property of being goldin case of the concept gold? This assigns a reference to gold that does indeed remain constant across allrelevant possible worlds. But the same holds for general concepts that are clearly not natural kind concepts.For example, if the concept bachelor has the property of being a bachelor as its referent, then it surelyrefers to that property in all possible worlds where it has a referent at allfor, which other property shouldit refer to if not to the property being a bachelor? For these reasons, I put the issue of rigidity aside in thispaper. The whole work in an argument against conceptual analysis is done by the nondescriptiveness ofnatural kind concepts anyway.

123

Synthese

But the apparent difference between a concept like the element with atomicnumber 79 and the concept knowledge can be understood in purely epistemic terms.While the descriptive content of the concept the element with atomic number 79is cognitively transparent, such that coming to know that content requires only min-imal reflection, the descriptive content of the concept knowledge is not cognitivelytransparent in this way, and coming to know it apparently requires a substantial amountof a priori reflection.

The dissociation between the epistemic properties of a concept and the metaphysicalstatus of its referent is further substantiated by the fact that we can also use nondescrip-tive concepts in order to refer to paradigmatic non-natural kinds. For example, we couldintroduce a nondescriptive concept of bachelorhood by using the merely reference-fixing description men with that marital status, while demonstratively referring to agroup of bachelors (cf. Kripke 1980, pp. 5758 on merely reference-fixing descrip-tions).

Therefore, even if knowledge is indeed a natural kind, it does not followthat every concept of knowledge is also a natural kind concept. One mightobject that we are not considering arbitrary concepts of knowledge here, such asErnest Sosas favorite philosophical subject matter or the main topicof KNOWLEDGE AND ITS PLACE IN NATURE. Rather, we are concerned with the pre-theoretical lexicalized concept knowledge that we standardly express with the ordi-nary English word knowledge. Even if it does not strictly follow from the natural kind-hood of K that our lexicalized concept of K is a natural kind concept, one might never-theless argue that this inference enjoys strong inductive support. For, in case of para-digmatic natural kinds, such as water, gold, tiger, or aluminum, the relevant lexicalizedconcepts are indeed natural kind concepts. So given the assumption that knowledge is anatural kind, isnt it at least highly probable that knowledge is a natural kind concept?

The answer is: only if the characteristic features of paradigmatic natural kinds areprojectible to the case of knowledge, and this seems highly questionable. For on theface of it, the kind knowledge is very unlike paradigmatic natural kinds. In case ofparadigmatic natural kinds, like water or gold, illuminating necessary and sufficientconditions for kind membership are identified at the level of underlying microstructure.For example, the underlying microstructure of water is identified at the molecular level,as being composed of H2O molecules, the underlying microstructure of gold is identi-fied at the atomic level, as being the element with atomic number 79, and the underlyingmicrostructure of tiger is presumably identified at the biochemical level, as being aspecies with a certain genetic make-up.14 The various instances of knowledge do notshare this paradigmatic feature of being unified by their underlying microstructure.

One reason for this important dissimilarity might be that cognitive kinds, likeknowledge, belief or intention, are multiply realizable at the underlying neurologi-cal or biochemical level, which is a widely held view in the metaphysics of mind

14 It should be noted, however, that the natural kindhood of biological species is a highly controversialissue in contemporary philosophy of biology (cf. Bird and Tobin 2015, Sect. 2.1). In particular, it is notclear whether the nature of biological species can be understood in terms of their intrinsic, microstructuralpropertiesor whether it must instead be understood in extrinsic, relational terms (cf. Okasha 2002; LaPorte2004).

123

Synthese

(cf. Putnam 1967; Fodor 1974; Bickle 2013).15 Accordingly, it makes sense, at leastin principle, to ascribe such cognitive states to a wide range of different creaturesfrom humans to ravens, chimpanzees, octopuses, or even aliens. For this reason, wecannot expect that the instances of these cognitive states have anything interesting incommon in terms of their underlying microstructure (see also Bird and Tobin 2015,Sect. 2.3).

In addition to that, no one has ever made a plausible suggestion concerning theunderlying microstructure of the kind knowledge. Kornbliths own supposedlyempirical identification of knowledge with reliably produced true belief (cf. Korn-blith 2002, pp. 6263) is clearly not a proposal in terms of microstructure, nor does ithave the flavor of a new or surprising scientific discovery that one finds in the case ofparadigmatic natural kindsin fact, the very same proposal was already made manyyears before on the basis of armchair conceptual analysis (cf. Goldman 1986).

So if knowledge is a natural kind, then it is at best a rather atypical natural kind, onewhose nature is not hidden at some underlying micro-level that calls for sustainedempirical investigation. For this reason, one should be wary of projecting features ofparadigmatic natural kinds, such as the existence of lexicalized natural kind concepts,onto atypical cases like knowledge. Moreover, as I will argue in the following section,there are strong direct considerations against the claim that knowledge is a naturalkind conceptconsiderations that would trump any inductive considerations of thesort just considered. In sum, there is no good reason for accepting the inference fromknowledge is a natural kind to knowledge is a natural kind concept.

What is more, premise (2) of the argument from natural kindhood derives its primafacie plausibility from the more general claim that if some category K is a natural kind,then illuminating necessary and sufficient conditions for K can only be figured outwith empirical methods. However, if we accept causalexplanatory indispensabilityas a good criterion of natural kindhood, as Kornblith suggests, then this more generalclaim becomes subject to a number of counterexamples. For there are cases wherea kind figures indispensably in causal explanations, and should thus be regarded asa natural kind according to the present criterion, but where an a priori analysis ofthe relevant lexicalized concept might still yield illuminating necessary and sufficientconditions. These cases undermine the crucial transition from K is a natural kind toilluminating necessary and sufficient conditions for K can only be figured out withempirical methods. Let us consider some of these cases.

A first case might be analytic functionalism in the philosophy of mind (cf. Arm-strong 1968; Lewis 1972; Shoemaker 1981; Braddon-Mitchell and Jackson 2007),which is a live theoretical option for at least some mental states. The basic idea is thata mental state M is individuated by its functional role, that is, by its typical causesand effects, and that this functional role is a priori accessible on the basis of reflectionon our concept of M. For illustration, let us assume that analytic functionalism is trueof pain. The mental state of pain would then be individuated by the typical causes

15 For the multiple realizability of knowledge it does not matter whether knowledge is a composite statethat consists of a belief that satisfies various further conditions, such as justification or truthwhich is theorthodox view in epistemology (cf. Nagel 2013), or whether knowledge is a distinctive mental state inits own right (cf. Williamson 1995, 2000; Nagel 2013).

123

Synthese

and effects of being in pain, such as grimacing-, wincing- and moaning-behavior. Thiscomplex cluster of typical causes and effects, which is considered to be a priori acces-sible on the basis of analyzing our concept pain, would then provide the materials for afunctional analysis of pain. Since the case for the causalexplanatory indispensabilityof mental kinds like pain is at least as strong as the corresponding case for knowledge,such mental kinds would qualify as natural kinds in light of the causalexplanatorycriterion. But in that case, the relevant mental kinds would be natural kinds that areamenable to a priori analysis.

Analytic functionalism is a controversial view, of course, but there are also lesscontroversial cases. Consider dispositional kinds like poison or fragility. These kindsoften play an indispensable causalexplanatory role, at least no less than knowledgedoes, and should thus be regarded as natural kinds according to the causalexplanatorycriterion. The fragility of glass explains, for example, why glass cannot be handled andtransported like other materials, such as metal or wood. It seems plausible, however,that we can figure out illuminating necessary and sufficient conditions for fragilitysolely on the basis of our grasp of the concept fragility, e.g., as a rough approxima-tion, that fragile objects tend to break when they collide with hard objects.16

Mathematical properties play an indispensable role in many causal explanations aswell.17 For example, primeness plays a crucial role in the explanation of the highlyunusual 13-year and 17-year life cycles of the cicada genus Magicicada. According toone of the leading hypotheses, the cycle length is a prime number in order to optimallyescape predators (Goles et al. 2001, p. 33).18 As Goles et al. elaborate, a prey witha 12-year cycle will meetevery time it appearsproperly synchronized predatorsappearing every 1, 2, 3, 4, 6 or 12 years, whereas a mutant with a 13-year periodhas the advantage of being subject to fewer predators. In other words, Magicicadahave their unusual 13-year and 17-year life cycles because the primeness of theselife cycles increases their survival rate and thus enhances their evolutionary fitness.Therefore, the primeness of the Magicicada life cycles is an indispensable part of thecausal explanation of these life cycles in evolutionary terms.19 In light of the causalexplanatory criterion for natural kindhood, primeness should thus be considered as anatural kind. However, an a priori analysis of the concept primeness clearly yieldsilluminating necessary and sufficient conditions for primeness, i.e., that something isprime if and only if it is only divisible by one and itself. The fact that even certainmathematical properties qualify as natural kinds according to the causalexplanatorycriterion of natural kindhood further undermines the crucial transition from K is anatural kind to illuminating necessary and sufficient conditions for K can only befigured out with empirical methods.

16 The apparent analyzability of fragility should not be confused with the seemingly more problematicidea of providing illuminating necessary and sufficient conditions for the dispositionality of fragility (cf.Bird 2007; Ellis 2001; Mumford 2004).17 Thanks to Brendan Balcerak Jackson for discussion.18 According to a competing hypothesis, the prime life cycles of Magicicada are adaptations that preventhybridization in small and isolated populations (cf. Cox and Carlton 1988; Yoshimura 1997).19 We are only considering a special science explanation here, of course, just as Kornblith does in the caseof knowledge and cognitive ethology.

123

Synthese

One might object that if it were indeed established that knowledge is a natural kind,then there would be no reason to be interested in the concept of knowledge anymore.Instead, one should investigate the natural kind knowledge itself, just as Kornblith urgeson the first page of Knowledge and Its Place in Nature.20 However, if it were also estab-lished that the concept knowledge is a descriptive concept, and not a nondescriptivenatural kind concept, then we might actually learn something about knowledge itselfby means of conceptual analysis, just as we learn something about primeness itself bydoing conceptual analysis. For in that case, conceptual analysis would arguably havethe power to reveal illuminating necessary and sufficient conditions for knowledge,and it seems hard to deny that this would be of philosophical interest.

5 Knowledge as a natural kind concept?

What emerges from the preceding discussion is that Kornbliths rejection of concep-tual analysis with respect to knowledge hinges primarily on the semantic claim thatknowledge is a natural kind concept, and only secondarily on the metaphysical claimthat knowledge is a natural kind. Therefore, concluding that conceptual analysis fails toyield illuminating necessary and sufficient conditions for knowledge on the basis of themetaphysical claim that knowledge is a natural kind actually turns things upside down.A more straightforward way to argue for this conclusion would be to employ familiarconsiderations from the discussion about natural kind terms and semantic externalism(cf. Kripke 1980; Putnam 1970, 1975). If one had shown, in this way, that knowledgeis a natural kind concept, then one could follow the standard procedure for establish-ing that knowledge is a natural kind, namely, by investigating whether paradigmaticinstances of knowledge share some underlying featurejust as paradigmatic instancesof water, tiger, or gold (cf. Kornblith 2002, pp. 1011; Kripke 1980; Putnam 1975). Asystematic discussion of the claim that knowledge is a natural kind concept wouldthus need to examine the standard arguments that semantic externalists have advancedfor this sort of conclusion. At a minimum, this would involve a discussion of Putnams(1975) Twin Earth thought experiment, Kripkes (1980) epistemic and modal argu-ments, and Burges (1979) related arguments for social externalism. In my view, thesearguments do not apply very well to the concept of knowledge (pace, e.g., Cappelenand Winblad 1999; Kumar 2014). Since a defense of this claim would require anotherpaper or chapter (cf. Horvath 2011), my aim in this section will be more limited,namely, to highlight two epistemic features of the concept of knowledge that make fora striking disanalogy between this concept and paradigmatic natural kind concepts.

It is characteristic of natural kind concepts, like water or gold, that a prioriconceptual analysis does not reveal illuminating necessary and sufficient conditionsfor membership in the relevant kinds. On the one hand, this comes down to the familiarobservation that we are subject to all kinds of ignorance and error when we try todetermine illuminating necessary and sufficient conditions for, e.g., water or gold onthe basis of a priori conceptual analysis (cf. Kripke 1980; Putnam 1975). On the otherhand, there is the less familiar observation that the proposed conceptual analyses of

20 Thanks to Kirk Michaelian for pressing this point.

123

Synthese

water or gold did not come anywhere near identifying illuminating necessary andsufficient conditions.

First, a priori analyses of gold as, e.g., being a yellow metal (cf. Kant 1977, 10:267), or of water as, e.g., being a clear drinkable liquid, did not hit upon any of theconditions that figure in the correct scientific analysis of these kinds, which is beingthe element with atomic number 79 in case of gold, and being composed of H2 Omolecules in case of water (cf. Kripke 1980; Putnam 1975). Therefore, these a priorianalyses are not just partly wrong, or subject to some degree of ignorance, but theyare defective in the more radical sense of being affected by total error.21

Second, prior to empirical research in modern chemistry nobody even had the con-cepts that are required for grasping the correct scientific analysis of, e.g., water interms of H2O, such as hydrogen or oxygen. And this is anything but an isolatedcase. Just consider concepts like proton, electromagnetism, or chromosome thatare required for grasping the correct scientific analysis of various other natural kinds.However, proponents of a priori analyses of natural kinds did not only fail to possessthe concepts that are needed for grasping their correct scientific analysis, but there wasalso no recognizable way of acquiring these concepts solely through further a prioritheorizing. Thus, a priori analyses of natural kinds were not only affected by total error,but the proponents of these analyses were also subject to a priori conceptual oblivi-ousness with respect to the correct scientific analysis of the natural kinds in question.22

We can summarize these two observations as follows:

(O1) A priori analyses of natural kinds are affected by total error.(O2) Proponents of a priori analyses of natural kinds are subject to a priori conceptual

obliviousness.

I will now argue that neither (O1) nor (O2) applies to a priori conceptual analyses ofknowledge, which is a powerful reason to conclude that knowledge is not a naturalkind concept.

Why does (O1) not apply to a priori conceptual analyses of knowledge? There is noindication that conceptual analyses of knowledge are subject to anything like the totalerror that affects a priori analyses of, e.g., water or gold. It is true, of course, that many apriori analyses of knowledge are subject to partial ignorance or error. This can be seen,for example, from the failure of the standard analysis of knowledge as justified truebelief (cf. Gettier 1963), andmore generallyfrom the failure of most suggestedanalyses of knowledge, and also from the fact that epistemologists endorse a varietyof different analyses of knowledge that are mutually incompatible (cf. Shope 1983,2002). The majority of contemporary epistemologists agree, however, that knowledgeimplies true belief (cf. Ichikawa and Steup 2012), or that the truth of a knowledge-constituting belief must not be an accident (cf. Unger 1968; Zagzebski 1994; Pritchard

21 A proposed analysis Ap of K is subject to total error iff none of the analyzing features that figure in Apalso figure in the correct analysis Ac of K.22 A thinker T is subject to a priori conceptual obliviousness with respect to the correct analysis Ac of Kiff T possesses none (or almost none) of the concepts C1, , Cn that are needed for grasping the analyzingfeatures that figure in Ac, and there is no realistic way for T to acquire C1, , Cn solely through further apriori theorizing.

123

Synthese

2005). So these largely uncontroversial necessary features of knowledge do not seem tobe affected by any ignorance or error, as far as we can tell from our present perspective(and this is the only perspective that we can reasonably take on this issue). Even Korn-bliths own allegedly scientific analysis of knowledge on the basis of considerationsfrom cognitive ethology includes some of these widely accepted necessary features,since he identifies knowledge with reliably produced true belief (cf. Kornblith 2002,p. 58). In fact, this is the same result that Goldman, according to his methodologi-cal self-understanding (cf. Goldman 2007), had obtained many years before on thebasis of a priori conceptual analysis (cf. Goldman 1986). These observations aboutcontemporary epistemologyincluding naturalized epistemologysuggest that (O1)does not apply to a priori conceptual analyses of knowledge, for there is presently noindication that they are affected by total error.

Why does (O2) not apply to a priori analyses of knowledge? Since some of thenecessary features of knowledge that are widely accepted on the basis of conceptualanalysis are apparently not subject to ignorance or error, it follows that philosopherspossess at least some of the concepts that are required for grasping the correct analysisof knowledge, such as truth or belief. Moreover, since truth and belief are prethe-oretical concepts, it follows that even lay people possess some of the concepts that arerequired for grasping some of the necessary features of knowledge. And even if epis-temic justification should be a pretheoretical concept as well, which seems doubtful(cf. Senor 2013), there is still reason to think that philosophers did not always possessall of the concepts that are required for grasping the analyzing features of knowledge.For as a result of Gettiers (1963) refutation of the standard analysis of knowledge asjustified true belief, the justification condition will either have to be supplemented orreplaced by some additional or alternative condition X. Some of the candidate featuresfor X that epistemologists have suggested are: indefeasibility (cf. Lehrer and Paxson1969), reliability (cf. Goldman 1979), sensitivity (cf. Nozick 1981), safety (cf. Sosa1999), or aptness (cf. Sosa 2007). It seems clear that the relevant concepts of inde-feasibility, reliability, sensitivity, etc. are not pretheoretical concepts, and that evenmost epistemologists only acquired them post-Gettier. So before Gettier, even mostphilosophers failed to possess some of the concepts that are needed for grasping someof the analyzing features of knowledge. Yet isnt this simply a form of a priori con-ceptual obliviousness with respect to the analyzing features of knowledge, and thusevidence that knowledge might be a natural kind concept after all?

It seems plausible that, pre-Gettier, even most epistemologists were subject to a lim-ited, partial from of conceptual obliviousness vis--vis the correct analysis of knowl-edge (whatever exactly it may be). But given that epistemologists did already possesscrucial concepts like truth or belief, this was very different from the completeconceptual obliviousness that is characteristic of proponents of a priori analyses ofnatural kinds. More importantly, however, there is no indication that philosophers wereever subject to a priori conceptual obliviousness concerning the analyzing features ofknowledge. Unlike concepts like hydrogen or atomicnumber that were only shapedand acquired through a sustained process of empirical theorizing, the acquisition oftechnical concepts in the theory of knowledge, like indeefeasibility, sensitivity, oraptness, does not seem to depend on empirical information or theorizing in any sub-stantial sense. Rather, these concepts were shaped and acquired within the very process

123

Synthese

of a priori reflection on the analyzing features of knowledge that is so characteristic ofthe method of conceptual analysis. Therefore, proponents of conceptual analyses ofknowledge were never subject to anything like the a priori conceptual obliviousnessthat we find in proponents of a priori analyses of natural kinds. At most, epistemologistswere (and probably still are) subject to a limited form of conceptual obliviousness thatcan be cured by further a priori theorizing. These considerations suggest that (O2) doesnot apply to proponents of a priori conceptual analyses of knowledge, which is anotherdisanalogy between the concept knowledge and paradigmatic natural kind concepts.

To conclude, the concept of knowledge lacks two epistemic features that are char-acteristic of paradigmatic natural kind concepts. The first of these features is that apriori analyses of paradigmatic natural kinds are affected by total error. The secondfeature is that proponents of a priori analyses of natural kinds are subject to a prioriconceptual obliviousness. These two features constitute a striking disanalogy betweenknowledge and paradigmatic natural kind concepts, like water or gold, that makesit quite unlikely that they all belong to the same semantic category.

6 Conclusion

The metaepistemological temptation that I described at the beginning of this papershould be resisted: there is no convincing argument against conceptual analysis in thetheory of knowledge from the claim that knowledge is a natural kind.

First, the argument from the indispensable causalexplanatory role of knowledgeto the natural kindhood of knowledge is problematic in several ways. The argumenteither proves too much by ruling out illuminating necessary and sufficient conditionsfor knowledge on metaphysical grounds alone (insofar as it relies on the homeostaticcluster account of natural kinds), or the argument loses the metaphysical connectionbetween causalexplanatory indispensability and natural kindhood (when it resorts to anon-causal theory of natural kinds). Absent such a connection, the inference from hav-ing a robust causalexplanatory profile to being a natural kind remains unsupported.

But what if knowledge were indeed a natural kind? Wouldnt this tell stronglyagainst conceptual analysis in the theory of knowledge? To support this worry, onemight appeal to the following seemingly plausible principle: if K is a natural kind, thenthe concept of K is a natural kind concept. And as we learnt from Putnam and Kripke,natural kind concepts like water or gold are not amenable to a priori conceptualanalysis. This principle fails, however, because one can even refer to paradigmaticnatural kinds with purely descriptive concepts, such as the element with atomicnumber 79 in the case of gold. It is true that lexicalized concepts of paradigmaticnatural kinds are typically natural kind concepts, but one cannot simply draw anyinductive conclusions about the concept of knowledge from that fact, because even ifknowledge is indeed a natural kind, it is a rather atypical natural kind. Moreover, thecriterion of causalexplanatory indispensability allows for various counterexamples tothe transition from K is a natural kind to K is not amenable to conceptual analysis,e.g., in the case of dispositional or mathematical categories. In order to assess theviability of conceptual analysis in the theory of knowledge, one should therefore focusdirectly on the semantic question whether knowledge is a natural kind concept or not.

123

Synthese

However, there are two striking epistemic disanalogies between the conceptknowledge and paradigmatic natural kind concepts that make it quite unlikely thatknowledge belongs to the same semantic category. First, in contrast to proposed apriori analyses of natural kinds, a priori conceptual analyses of knowledge are notaffected by total error. And second, proponents of conceptual analyses of knowledgeare not subject to a priori conceptual obliviousness with respect to the analyzing fea-tures of knowledge.

We should thus resist the temptation to draw negative conclusions about conceptualanalysis from the natural kindhood of knowledge, for this kind of inference is fraughtwith difficulties and problems. This point presumably applies to other important philo-sophical categories as well, such as free will, justice, action or truth. At any rate, oneshould expect that at least some of the considerations that are relevant to the case ofknowledge also apply to those other cases. With respect to the case of knowledge, Iconclude that conceptual analysis remains unshaken by Kornbliths argument from thenatural kindhood of knowledge. Conceptual analysis continues to be a viable methodin the theory of knowledge, irrespective of the largely orthogonal question of whetherknowledge is a natural kind or not.

Acknowledgments I would like to thank Brendan Balcerak Jackson, Thomas Grundmann, Frank Hof-mann, Jonathan Jenkins Ichikawa, Stan Husi, Jens Kipper, Hilary Kornblith, Kirk Michaelian, WolfgangSchwarz, Anand Vaidya, and various anonymous reviewers for numerous helpful comments on this paperand its non-identical predecessors. The paper originated from a critical comment on Hilary Kornblithswork at the 2nd Cologne Summer School in Philosophy in August 2007 at the University of Cologne.Special thanks to Hilary for extensive discussion and plenty of encouragement. I also want to thank theparticipants of Hilarys doctoral colloquium at the University of Massachusetts Amherst in March 2008Kristoffer Ahlstrom-Vij, Jeremy Cushing, Jeff Dunn, Meghan Masto, Alex Sarch, Kirk Michaelian, IndraniBhattacharjee, and Hilary Kornblithfor their generous engagement with my paper and very valuable com-ments. Thanks also to Brendan Balcerak Jackson, Magdalena Balcerak Jackson, Alma Barner and WolfgangSchwarz for their very helpful comments in a reading group session of the Emmy Noether IndependentJunior Research Group Understanding and the A Priori in June 2009, which was kindly hosted by theUniversity of Cologne and generously supported by the Deutsche Forschungsgemeinschaft DFG (GermanResearch Foundation). Additional thanks to the DFG for supporting my research on this paper as part of theproject Eine Verteidigung der Begriffsanalyse gegen die Herausforderungen des Naturalismus (A Defenseof Conceptual Analysis against the Challenges from Naturalism), which was kindly hosted by the Universityof Cologne from 2007 to 2010 (under the auspices of Thomas Grundmann).

References

Armstrong, D. (1968). A materialist theory of the mind. London: Routledge & K. Paul.Besson, C. (2010). Rigidity, natural kind terms and metasemantics. In H. Beebee & N. Sabbarton-Leary

(Eds.), The semantics and metaphysics of natural kind terms. New York: Routledge.Bickle, J. (2013). Multiple Realizability. In E. N. Zalta (Eds.), The Stanford Encyclopedia of

Philosophy. Retrieved March 5, 2015, from http://plato.stanford.edu/archives/spr2013/entries/multiple-realizability/.

Bird, A. (2007). Natures metaphysics: Laws and properties. Oxford: Clarendon Press.Bird, A., & Tobin, E. (2015). Natural kinds. In E. N. Zalta (Eds.), The Stanford Encyclopedia of Philosophy.

Retrieved February 27, 2015, from http://plato.stanford.edu/archives/spr2015/entries/natural-kinds/.Boyd, R. (1988). How to be a moral realist. In G. Sayre McCord (Ed.), Essays on moral realism. Ithaca,

N.Y: Cornell University Press.Boyd, R. (1991). Realism, anti-foundationalism and the enthusiasm for natural kinds. Philosophical Studies,

61(1), 127148.

123

Synthese

Braddon-Mitchell, D., & Jackson, F. (2007). The philosophy of mind and cognition: An introduction. Oxford:Blackwell.

Burge, T. (1979). Individualism and the mental. Midwest Studies in Philosophy, 4, 73121.Cappelen, H., & Winblad, D. (1999). Intuitions. Facta Philosophica, 1, 197216.Cox, R. T., & Carlton, C. E. (1988). Paleoclimatic influences in the evolution of periodical cicadas (Insecta:

Homoptera: Cicadidae: Magicicada spp.). American Midland Naturalist, 120(1), 183193.Devitt, M., & Sterelny, K. (1999). Language and reality: An introduction to the philosophy of language.

Cambridge, MA: MIT Press.Ellis, B. (2001). Scientific essentialism. Cambridge: Cambridge University Press.Fine, K. (1995). Senses of essence. In R. B. Marcus, W. Sinnott-Armstrong, D. Raffman, & N. Asher (Eds.),

Modality, morality and belief: Essays in honour of Ruth Barcan Marcus (pp. 5373). Cambridge:Cambridge University Press.

Fodor, J. (1974). Special sciences (or: The disunity of science as a working hypothesis). Synthese, 28(2),97115.

Gettier, E. (1963). Is justified true belief knowledge? Analysis, 23(6), 121123.Goldman, A. (1979). What is justified belief? In G. Pappas (Ed.), Justification and knowledge (pp. 123).

Dordrecht: D. Reidel.Goldman, A. (1986). Epistemology and cognition. Cambridge, MA: Harvard University Press.Goldman, A. (2007). Philosophical intuitions: Their target, their source, and their epistemic status. Grazer

Philosophische Studien, 74, 126.Goles, E., Schulz, O., & Markus, M. (2001). Prime number selection of cycles in a predator-prey model.

Complexity, 6(4), 3338.Heller, M. (1996). The mad scientist meets the robot cats: Compatibilism, kinds, and counterexamples.

Philosophy and Phenomenological Research, 56(2), 333337.Horvath, J. (2011). In defense of conceptual analysis. Dissertation. University of Cologne.Horvath, J. (manuscript). Philosophical analysis: The concept grounding view.Ichikawa, J. J., & Steup, M. (2012). The analysis of knowledge. In Edward N. Zalta (Eds.), The Stan-

ford Encyclopedia of Philosophy. Retrieved April 9, 2013 from http://plato.stanford.edu/archives/win2012/entries/knowledge-analysis/.

Kant, I. (1977). In J. W. Ellington (Ed.), Prolegomena to any future metaphysics. Indianapolis: HackettPublishing.

Kornblith, H. (1993). Inductive inference and its natural ground: An essay in naturalistic epistemology.Cambridge, MA: MIT Press.

Kornblith, H. (1999). Knowledge in humans and other animals. Philosophical Perspectives, 13, 327346.Kornblith, H. (2002). Knowledge and its place in nature. Oxford: Clarendon Press.Kornblith, H. (2007). Naturalism and intuitions. Grazer Philosophische Studien, 74, 2749.Kornblith, H. (2013). Is there room for armchair theorizing in epistemology? In M. Haug (Ed.), Philosophical

methodology: The armchair or the laboratory? (pp. 195216). London: Routledge.Kripke, S. (1980). Naming and necessity. Cambridge, MA: Harvard University Press.Kumar, V. (2014). Knowledge as a natural kind term. Synthese, 191(3), 439457.LaPorte, J. (2004). Natural kinds and conceptual change. Cambridge: Cambridge University Press.Lehrer, K., & Paxson, T. (1969). Knowledge: Undefeated justified true belief. Journal of Philosophy, 66(8),

225237.Lewis, D. (1972). Psychophysical and theoretical identifications. Australasian Journal of Philosophy, 50(3),

249258.Lewis, D. (1986). On the plurality of worlds. Oxford: Blackwell.Ludwig, K. (2013). Methods in analytic epistemology. In M. Haug (Ed.), Philosophical methodology: The

armchair or the laboratory? (pp. 217239). London: Routledge.Malmgren, A.-S. (2011). Rationalism and the content of intuitive judgements. Mind, 120(478), 263327.Micevic, N. (2000). Deep and superficial a priori. Acta Analytica: Philosophy and Psychology, 15, 89112.Micevic, N. (2005). Empirical concepts and a priori truth. Croatian Journal of Philosophy, 5, 289315.Mumford, S. (2004). Laws in nature. London: Routledge.Nagel, J. (2013). Knowledge as a mental state. In T. S. Gendler & J. Hawthorne (Eds.), Oxford Studies in

Epistemology (pp. 272308). Oxford: Oxford University Press.Nozick, R. (1981). Philosophical explanations. Cambridge, MA: Harvard University Press.Okasha, S. (2002). Darwinian metaphysics: Species and the question of essentialism. Synthese, 131(2),

191213.

123

Synthese

Pernu, T. K. (2009). Is knowledge a natural kind? Philosophical Studies, 142(3), 371386.Pritchard, D. (2005). Epistemic luck. Oxford: Oxford University Press.Putnam, H. (1967). Psychological predicates. In W. H. Capitan & D. D. Merrill (Eds.), Art, mind, and

religion (pp. 3748). Pittsburgh: University of Pittsburgh Press.Putnam, H. (1970). Is semantics possible? Mind, language and reality?: Philosophical papers (Vol. 2, pp.

139152). Cambridge: Cambridge University Press.Putnam, H. (1975). The meaning of meaning. Mind, language and reality : Philosophical papers (Vol. 2,

pp. 215271). Cambridge: Cambridge University Press.Schwartz, S. (1980). Formal semantics and natural kind terms. Philosophical Studies, 38(2), 189198.Schwartz, S. (2002). Kinds, general terms, and rigidity: A reply to LaPorte. Philosophical Studies, 109(3),

265277.Schwitzgebel, E. (2008). Against metaphysics, especially the metaphysics of consciousness. The Splintered

Mind. http://schwitzsplinters.blogspot.com/2008/07/against-metaphysics-especially.html.Searle, J. R. (1969). Speech acts: An essay in the philosophy of language. London: Cambridge University

Press.Senor, T. D. (2013). Justified belief and demon worlds. Res Philosophica, 90(2), 203214.Shoemaker, S. (1981). Some varieties of functionalism. Philosophical Topics, 12(1), 93119.Shope, R. (1983). The analysis of knowing: A decade of research. Princeton: Princeton University Press.Shope, R. (2002). Conditions and analyses of knowledge. In P. Moser (Ed.), The Oxford Handbook of

Epistemology (pp. 2571). Oxford: Oxford University Press.Soames, S. (2002). Beyond rigidity: The unfinished semantic agenda of Naming and necessity. Oxford:

Oxford University Press.Soames, S. (2007). What are natural kinds? Philosophical Topics, 35(1), 329342.Sosa, E. (1999). How to defeat opposition to Moore. Nos, 33, 141153.Sosa, E. (2007). A virtue epistemology: Apt belief and reflective knowledge. Oxford: Oxford University

Press.Steup, M. (1996). An introduction to contemporary epistemology. Upper Saddle River, N.J.: Prentice Hall.Unger, P. (1968). An analysis of factual knowledge. The Journal of Philosophy, 65(6), 157170.Williamson, T. (1995). Is knowing a state of mind? Mind, 104(415), 533565.Williamson, T. (2000). Knowledge and its limits. Oxford: Oxford University Press.Williamson, T. (2007). The philosophy of philosophy. Malden, MA: Blackwell.Yoshimura, J. (1997). The evolutionary origins of periodical cicadas during ice ages. The American Natu-

ralist, 149(1), 112124.Zagzebski, L. (1994). The inescapability of Gettier problems. Philosophical Quarterly, 44(174), 6573.

123

Conceptual analysis and natural kinds: the case of knowledgeAbstract1 Introduction2 The argument from the natural kindhood of knowledge3 Knowledge as a natural kind?4 Natural kindhood and conceptual analysis5 Knowledge as a natural kind concept?6 ConclusionAcknowledgmentsReferences