sign in sign up
<<
Home , Nominalism, Richard Boyd

Realism, Anti- and the Enthusiasm for Natural Kinds Author(s): Richard Boyd Source: Philosophical Studies: An International Journal for in the Analytic Tradition, Vol. 61, No. 1/2, The Twenty-Ninth Oberlin Colloquium in Philosophy (Feb., 1991), pp. 127-148 Published by: Springer Stable URL: http://www.jstor.org/stable/4320174 . Accessed: 08/01/2015 09:38

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

.

Springer is collaborating with JSTOR to digitize, preserve and extend access to Philosophical Studies: An International Journal for Philosophy in the Analytic Tradition.

http://www.jstor.org

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions RICHARD BOYD

REALISM, ANTI-FOUNDATIONALISM AND THE ENTHUSIASM FOR NATURAL KINDS

(Received 14 March, 1990)

0. THE TRADITION OF NATURAL KINDS

0.1 Hacking'sChallenge

In his interestingand challengingpaper "A Traditionof NaturalKinds" Professor Hacking identifies a long-standingempiricist tradition of theorizingabout naturalkinds, he identifiesits roots as nominalistand its main concern as the characterizationof the sorts of kinds and kind definitionsappropriate for the formulationof reliableinductive gener- alizations.Hacking distinguishes a numberof sub-traditionswithin the tradition of natural kinds but what is central to all of them is the conception that a naturalkind is one such that it is (i) (at least on a certain idealization) defined by a of necessary and sufficient properties(relations, etc.) such that, (ii) the possession of these proper- ties is, as a matterof ratherthan of , indicativeof a very large numberof other methodologicallyinteresting properties and such that, (iii) these definingproperties are naturalrather than social properties. When the propertiesdefining a kind satisfycondition (ii) then it is a fact about nature that the kind in question exists and a fact about nature that reference to the kind in question is appropriatefor the formulationof causal generalizationsrelated to the correla- tions indicatedin (ii). Hackingpoints out that this conceptionof naturalkinds was modest in scope: althoughthe sub-traditionsdiffered on just which kinds are the naturalkinds all agreedthat they constitutedonly a smallfraction of all the kinds there were - a smallfraction even of those non-arbitrarily defined kinds which are naturalcategories in some importantsense of "natural". By contrast,recent work on naturalkinds - especiallywork under-

Philosophical Studies 61: 127-148, 1991. ? 1991 KluwerAcademic Publishers. Printed in the Netherlands.

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 128 RICHARD BOYD taken by realists in the - has reflected a much granderconception. Almost all sorts of kinds and kind terms except the most clearly arbitraryhave been treated as natural kinds and kind terms at least in the sense that the naturalisticconceptions of definitions and of reference developed for the case of more traditionalnatural kindsand naturalkind terms are understoodto applyto themas well. Hacking challenges the new enthusiasmfor the theory of natural kinds. He suggests,in particular,that kinds which fail to satisfythe first of the traditionalconditions on naturalkinds (as propertycluster kinds do) or which fail to satisfy the third condition (as social kinds do) should not be assimilatedto a conception of natural kinds however "natural"the categoriesthey correspondto may otherwisebe. In neither case apparentlyis the of the kinds in question a fact about naturein the relevantrespects. Against the assimilationof the first sort of kinds Hackingargues that kinds of that sort ". . . are constructed along lines of and what puts things into a familyis not naturebut people in concert." He also argues that the relevantfamily resemblanceswill characteris- tically include "plainlysocial" as well as "natural"properties, causing most propertycluster kinds to violate as well the stricturerepresented by the thirdcondition. In defense of that condition - which is to say against treating socially defined kinds as naturalkinds - Hackingoffers considerations which go to the heart of our conception of a naturalkind as a fact of nature.He suggeststhat the complex feedbackloops which connect our classificationof, and theorizingabout, people with changesin their self- and behaviorand back again make it inappropriateto think of socially defined kinds as kinds defined by nature rather than by us. Realistsoften talk about employingnatural kind terms to "carvenature at its joints." In the case of socially defined categories, the causal powers of the objects classified- and thus the location of the "joints" - depends on our own acts of classificationand theorizingrather than on nature. What I propose to do in the present paper is to defend the enthu- siasm for naturalkinds againstHacking's criticisms. I agree that there are important philosophical and methodologicaldifferences between kinds defined by property clusters and kinds defined by sets of

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 129 properties, and between natural properties and social properties. I believe however that there are extremely good for treating manyproperty-cluster kinds and social kinds, and the termswhich refer to them, on the model of natural kinds and natural kind terms. In particular I hold, about any property cluster kind or social kind referenceto which is importantin explanationor induction,that three claims be true of it which representthe centralcore of the current conceptionof naturalkinds. First, the definitionof the kind in question is determineda posteriorirather than by social convention.Second, this in turn is true because the use we make of reference to the kind in inductionand explanationrequires that it be defined by a set or cluster of propertieswhose membershipis determinedby the causal structure of the world and is thus, in a relevant sense (which I propose to specify),independent of our conventionsor our theorizing.Finally, the establishmentof a relationof referencebetween a term and the kind in question depends on the existence of an epistemicallyrelevant pattern of causal relationsbetween instantiationsof the kind and the use of the term in question and not on the acceptance by any or all of the membersof the relevantlinguistic community of the correctdefinition of the kind. In order to defend the assimilationof property cluster and social kinds to the model of naturalkinds I'll need to offer a brief and (only slightly) anachronisticaccount of the history of the traditionof theo- rizingabout natural kinds.

0.1. Locke on Projectability,or the TensionBetween Nominalism and RationalInduction Hackingmaintains, correctly, that the traditionof researchinto natural kinds is largely nominalistic. What he does not say but what is uncontroversialis that the tradition is nominalistic because it is basicallyverificationist: the nominalismin question is an outgrowthof the traditionalempiricist skepticism about the possibilityof real , hidden mechanisms,causal powers or other "unobserv- ables". Hacking also rightly emphasizes the fact that the tradition is concernedwith identifyingthose kinds appropriatefor induction(and, I wouldadd, ).

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 130 RICHARD BOYD

There is something else which is also uncontroversialbut which Hackingdoes not emphasize:that the traditionof naturalkinds reflects (at least in its empiriciststages) a deep tension between the nominalistic project of eliminating"" and the project (equallycentral to ) of characterizingrational principles for induction. That there is some such tension can be seen by reflectingon the fact that the central component of the traditionof naturalkinds is not compatible withtraditional nominalism. I take traditionalnominalism to be reflectedin the Lockeanposition that while naturemakes things similarand differentthe classificationof them into kinds is solely "the workmanshipof the understanding." According to this central empiricistconception of nominal definitions such definitionsare almost entirely arbitrary.For Locke, for example, the only permissibleconstraints on the of a mixed mode is logical consistencyof the united to form it; for an idea of substancethere is the additionalconstraint that the qualitiesand powers corresponding its constituentideas must be conjointlyrealizable in nature.Subject, at most, to these constraints "... each abstract idea, with a name to it, makesa distinctSpecies." Accordingto the traditionof naturalkinds, in both its empiricistand its realist versions, much deeper non-conventional constraints are appropriate:it is a fact about nature, rather than about the under- standing, that natural kinds exist and have the definitions they do. Likewise it is a natural fact about a particular natural kind that reference to it is appropriatefor certain sorts of inductive generaliza- tion and this naturalfact is explained by property correlationswhich obtainindependently of the understanding.If some kinds are featuresof natureitself, and if the principlesof classificationwhich define them fit the demandsset by our projectsand by natureitself better than others, then those kinds are at least as much the workmanshipof natureas of the understanding.The traditionalnominalist doctrine that kinds are merelythe workmanshipof the understandingis deeplycompromised. I see the tension between empiricist nominalism and the task of accounting for induction first arising in Locke's Essay. As Locke acknowledgesat several points (see, e.g., IV, iii, 13, 14, 25, 29; IV, xii. 9, 10) his nominalistic conception of kind definitions dictates a skeptical conclusion regardingthe possibility of general knowledge of

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 131 substances.If the observableproperties of substancesare in fact deter- mined by the primary qualities of their insensible parts, as Locke supposes, then inductive generalizationswill be unjustifiedunless we are justified in believing that the categories in which they are framed correspond to uniformities in corpuscular structure:If all observed samples of kind K have exhibited causal power P in of their underlyingmicrostructure, then we will therebyobtain to believe that K's generallyhave P only if we have reason to believe that other samples of K will have relevantlysimilar microstructure.If, as Locke insists, we must classify substances according to arbitrarynominal essences instead of according to microstructuralreal essences, we never have any such reason and general knowledge of substances is impossible. It is difficultto be certainjust how deeply Locke takes the skeptical difficulties concerning general knowledge of substances which his theory of ideas of substance dictates;in some places he seems to be genuinely pessimistic about chemical knowledge;in others he seems oddly optimisticabout the fruits of guessingand reasoningby analogy. What is importantis that the skeptical force of his argumentremains even when we're only guessing:if we have no reason to guess that K's all have relevantlysimilar microstructure (as we have not if the nominal defining the kind K is arbitrary)then we have no reason to guess that propertiesobserved K's have displayedin the past they will continueto displayupon furtherexamination. Even for the purposesof guessingwe need categoriesof substancewhose boundariesare not (or not just)"the workmanship of the understanding." Locke recognizesthis and severalrelated embarrassing consequences of his nominalismand at several places he suggests as a remedy that substantialkinds should ideally be defined so that their boundariesare set, not solely by the workmanshipof Man, but in accordance with observed correlations of properties -". . . That Men, versed in physical Enquiries, and acquainted with the several sorts of natural Bodies, would set down those simple Ideas, wherein they observe the Individualsof each sort constantlyto agree"(III, xi, 25; see also IV, xii, 14; II, xxix, 7). Officially in these passages Locke is addressingthe problem of the "inadequacy"of kinds of substances rather than the problem of inductivecategories, but (for reasons which lie beyond the

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 132 RICHARD BOYD scope of this essay) I am inclinedto think that he means the solutionto apply to the latter problemas well. I thus believe that these and similar passages in Locke represent the earliest stages of the empiricist traditionof philosophizingabout naturalkinds to which Hackingrefers. At least they representthe earliestappearance within empiricism of the anti-nominalistconception that in defining kinds we should defer to naturallyoccurring patterns of propertycorrelations. The tension in Locke between his anti-metaphysicalnominalism and the need to account for ordinaryinductive methods itself has a struc- ture which is repeated again and again in the empiricist tradition. Locke's nominalismarises from his brodaer proto-verificationistskep- tical critiqueof "realessences": they are not possible objects of knowl- edge and are thus unavailableto us as the basis for classification.This skepticism was meant to be selective: it was not meant to spill over into a denial of the possibility of everyday knowledge of substances or even of something like general knowledge of their properties.What Locke recognizedis that if skepticismabout real essences is understood to entail the doctrine that substantialkinds are arbitrary- are just the workmanshipof the understanding- then the resultingskepticism fails to be selective. Locke's appeal to what we would now recognize as a conception of naturaldefinitions may have been - and later empiricist theories of natural kinds definitely were - attempts to restore the selectivity to the empiricist skeptical critique of real essences: if the verificationistcritique of real essences does not demandthat substantial kinds have entirely arbitrarynominal definitions- but only that their definitionsnot be framed in terms of insensible real essences - then the underlyingskeptical conception can perhaps be suitably selective afterall. We may characterizeLocke's philosophicalsituation in anachronistic but nonethelessuseful terms.He was concernedwhich the characteriza- tion of projectableproperties of substances,in the sense of Goodman 1973, and was faced with an argumentwhich suggestedthat judgments of projectabilityin chemistry would have to depend to the sort of theoreticalknowledge which his verificationismprecluded. The alterna- tive to embracingextreme skepticism about induction or accepting a realist conception that theoretical knowledge is possible which he

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 133 perhaps embraced (and which latter participantsin the tradition of naturalkinds certainlydid embrace)was to propose that the specifica- tion of projectableproperties proceed by second-orderinduction about induction in the sense of Quine 1969. It is thus proposed to preserve the verificationistcritique of metaphysicsat the expense of abandoning a strictlynominalist conception of kinds.

0.2. 20th CenturyDevelopments, or, The Theory-Dependenceof Almost Everything Except for suggestinga somewhatearlier date for its beginningI agree with Hacking'scharacterization of the developmentof the traditionup to the point at which the controversialenthusiasm for natural kinds emerged. That enthusiasm has arisen largely among committed to scientific realist rather than empiricist conceptions of science. To see why, we need to recognize that philosophicalexamina- tion of the methods of actualscience has led, in the last few decades, to the confirmationof what might have been Locke's worst nightmare.In the first place, of course, general and systematic knowledge of sub- stances is certainlypossible. Now the bad part: all of the fundamental methods by which that knowledge is obtained are profoundlytheory- dependent:principles of classification,methods for assessing project- ability and for assessing the quality and the evidential import of observations,standards for assessingexplanatory power, etc. has gained considerablecredibility as a result of the recognitionof these about scientificpractice. It seems possible to argue that inductive in science about observables are reliable only because they are guided by methodological principles which reflect previously acquired (approximate)knowledge of unob- servable real essences. The articulationof a naturalisticconception of the definition of naturalkinds and a related naturalisticconception of referencehas proven essentialto makingthis abductivecase for realism (Putnam 1975a, b; Boyd 1979, 1982) so it is hardly surprisingthat scientific realists have enthusiasticallyparticipated in articulatingthe traditionof naturalkinds. It remains to see whether the extension of that tradition in which

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 134 RICHARD BOYD

realists have played a large role is justified. I hope to show that, whether one accepts a realist or an empiricistdiagnosis of the theory- dependence of method, there are very good reasons for extendingthe conception of natural kind to include all property-clusteror social kinds referenceto which plays any significantrole in inductionor social explanation.

0.3. Life After Theory-dependence:Global Nominalism or Epistemic

Locke's theory of nominal essences emerged as 20th century opera- tionalism and failed for just the reasons Locke might have feared: scientific methods of classificationare profoundly theory-dependent and kind definitionsare thus revisablein the light of new knowledge. There is an ineliminablea posteriori element in the definition of any scientificallysignificant kind or category.Empiricist responses to this and other consequences of the theory-dependenceof methods fall broadlyinto two categories. Many empiricistsmoved from the local nominalismrepresented by the operationalistconception that individualscientific term definitions are purely conventionalto varietiesof global nominalismaccording to which various higher level propositionsare true by convention and in turn govern theory-dependent practices like the revision of kind definitions. When methodological practices depend irretrievablyon theoretical principles apparentlyreflecting a posteriori knowledge of unobservables,the global nominalist"rationally reconstructs" certain of the relevantprinciples as analyticstatements, L- for the relevant scientificlanguage L. This strategyfor rationalreconstruction is reflected,for example,in the view that, while particularkind definitionsare not a priori, certain "bridgelaws" or "correspondenceprinciples" linking theoretical and observationalterms are matters of linguisticconvention. It is likewise reflectedin the still more global conceptionof Carnap1950 according to which it is the most fundamentallaws within a scientific discipline which are to be thoughtof as L-truths,and in reconstructionsaccording to which various featuresor categoriesof scientificterms or sentences are treated as defined a priori: lawlikeness,for example, or the

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 135 notion of being a term or a law "of "as that notion is employed in traditionalreductionist reconstructions of . Global nominalism,if it works, eliminates from our conception of scientific practice an apparentreference to "metaphysical"knowledge of unobservablesand it does so in an essentiallyLockean way: where rational scientific practice would seem to require knowledge of un- observables the relevant judgments are recast as having a priori conventionaljustifications. The other empiriciststrategy for accommodatingthe deep theory- dependence of scientificmethod is reflected in the traditionof natural kinds. Where methodologicalpractices (definitionsof scientific terms for example) appear to depend upon a posteriori knowledge of unobservables, they are reconstructed as depending on a posteriori claims not about unobservablesbut about property correlations or other methodologicallyrelevant but non-metaphysicalmatters. Theory- dependent methods are thus diagnosed as a matter of second-order inductionabout induction (see Quine 1969). Note that the first of these strategiescannot be appliedto all cases of theory-dependenceof methodologicalpractices. This can be seen by considering the deepest strategic problem facing global nominalism. The global nominalistresponds to the theory-dependenceof methods by treatingcertain methodologicallyimportant laws or generalizations which are apparently a posteriori claims about unobservables as conventional and hence a priori instead. Since any law or generaliza- tion whatsoevermay play a role in determiningsome methodological practicesor other, she must assure herself, and her critics,that she will not be driven to treat as a priori all of the apparently a posteriori laws and generalizationsof science. The methodologicalinsights of the laws or generalizationswhich are not reconstructedas a priori must then be portrayedas reflections of inductions about inductive methods. Each particularversion of global nominalismwill embody a differentsolution to this problembut any version of global nominalismwill necessarilybe part of a mixed strategy for the rational reconstructionof theory- dependent methods. What is central to the new enthusiasmfor natural kinds is a certainestimate of the kind of mix which is dictatedby facts about the actual methods of science. It is to the justificationof that estimatethat we now turnour attention.

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 136 RICHARD BOYD

1. NATURAL KINDS AND THE NATURALISTIC OF SCIENCE

1.0. TheFailure of Foundationalism The recent enthusiasmfor naturalkinds arose from the realist insist- ence that a realist interpretationis required for all of the theory- dependentmethods of science includingthe methods of kind definition. Without accepting this claim - in particularwithout precludingthe possibilityof a sophisticatedempiricist treatment of those methods we can draw some important epistemological conclusions from an examinationof the one of the most central of the abductivearguments for realism.It goes like this: 1. In actual scientificpractice a theory,T, is substantiallysupported by a body D of data if and only if, (a) T itself is projectable,(b) the data in D confirmthe predictionsof T or are explicableon the basis of T, (c) for each significantlyprojectable theory T' which is an alternativeto T there are in D data which either refute predictions of T' or cast significantdoubt on its explanatoryresources, and (d) the experiments or observational situations in which the data in D were obtained embody adequatecontrols for those possible experimentalor observa- tional artifactswhich are themselves suggestedby projectabletheories aboutthe relevantcircumstances. 2. Projectabilityjudgments are almost alwaysjudgments of theoretical plausibility.A theory is projectablejust to the extent that what it says about both unobservableand observable phenomena coheres appro- priatelywith the best confirmedbackground theories: it is projectable just in case it is supportedby plausible inductive inferencesfrom the "observational"and "theoretical"claims embodied in previously well establishedtheories. 3. Projectabilityjudgments depend on the theoreticalcontent as well as the observationalcontent of backgroundtheories: In general a theory diagnosedas projectablewith respect to one of two empiricallyequiva- lent sets of backgroundtheories will not be empiricallyequivalent to any of the theoriesdiagnosed as projectableby the other.

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 137

4. Projectabilityjudgments are always empiricallyrevisable in the light of revisions to backgroundtheories necessitatedby new data or new theoreticalinsights.

5. Therefore projectability judgments are a posteriori judgments whose contributionto the reliabilityof scientificmethods depends on knowl- edge reflectedin the backgroundtheories. [From 1, 2, 4] 6. That knowledge goes beyond the empiricalcontent of the relevant backgroundtheories and is "extracted"by reasoningwhich has just the form of inductiveinference from partly"theoretical" premises. [From 2, 3951 7. Thereforethe best account of projectabilityjudgments is the realist one according to which those judgments reflect inductive inferences from previously acquired approximateknowledge of unobservableas well as observablephenomena. What is importantfor our purposes is that, through step six, this argumentis acceptable to most sophisticatedempiricists as well as to realists. The crucial idea in step 4 that important methodological practices are determinedby empiricallyrevisable features of accepted theories has been accepted by empiricistphilosophers of science for a long time:the same point about measurementprocedures did, after all, lead to their abandonmentof operationalism.Moreover the empiricist analysis of theory-dependentmethods as reflections of second-order inductionsabout inductionis ideally suited to an empiricistreconstruc- tion of the role of backgroundtheories in determiningprojectability judgments:the knowledgereflected in backgroundtheories which goes beyond their empirical adequacy is merely a matter of those back- ground theories also being repositoriesof empiricalknowledge about the reliabilityof inductivemethods. The premisesof the argumentwe are consideringare extremelywell established.It is thus reasonable to suppose that the most plausible versionsof empiricistphilosophy of science will accept steps 1-6 while adopting something like the reconstruction just mentioned as an alternativeto step seven. It is on this assumptionthat I will defend the claim that the recent extensionsby realists of the traditionalempiricist

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 138 RICHARD BOYD conceptions of naturalkinds are warrantedby empiricistas well as by realiststandards. Modern epistemology has been dominated by foundationalistcon- ceptions of knowledge.We may rhakeexplicit a fundamentalbut often tacit assumptionof foundationalistepistemology if we think of founda- tionalismas consistingof two parts. Premisefoundationalism, which is typically made explicit, holds that all knowledge is justifiablefrom a core of epistemicallyprivileged foundational beliefs. Inferencefounda- tionalism, which is less often made explicit, holds that justifiable principlesof inductiveinference are ultimatelyreducible to inferential principleswhich are justifiable a priori. Recent work in naturalisticepistemology (e.g., Goldman 1967, 1976; Quine 1969) strongly suggests that the foundationalismis fundamen- tally mistaken.In typical cases of perceptualknowledge, there seem to be neither premises nor inferences. Even where premises and infer- ences are important,it seems to be their contributionto the reliable productionof that constitutestheir contributionto knowledge.A variety of considerationssuggests that there are no beliefs which are epistemicallyprivileged in the way traditionalfoundationalism seems to require. The response to the theory-dependenceof projectabilityjudgments which is common to scientificrealism and to sophisticatedempiricism suggeststhat foundationalismis even more profoundlymistaken. If, as that response suggests, all inductive reasoning rest upon a posteriori projectabilityjudgments, then inferencefoundationalism is profoundly false. For the case of generalknowledge of naturalphenomena we must always rely on inferentialprinciples which lack a priori justification. This radical contingencyin epistemologyis, I shall suggest, the crucial fact - agreed upon both by realists and sophisticatedempiricists - whichdetermines the properscope of the theoryof naturalkinds.

1.1. RadicalContingency and NaturalKinds

Early theorists of natural kinds recognized that there were certain importantcircumstances in which it was necessary,in order to have the resourcesfor formulatingsuccessful inductive generalizations, to deploy

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 139

categories defined a posteriori in such a way as to reflect the actual causal structureof the world. As we have seen, in those cases in which reference to them is necessary, natural kinds reflect a strategy of deferringto naturein the makingof projectabilityjudgments: we define such kinds a posteriori in ways which reflect actual causal structure precisely because we are unable to identify or specify projectable generalizationswithout doing so. Our question in the present essay is how widespreadare the cases whichthus requiredeference to nature. On certain traditionaland plausible conceptions of induction the answer might be, "Only rarely",or "Only in immature sciences". It might, for example, seem that a kind of rough and ready deference to naturewith respect to projectabilityjudgments would be necessaryonly in the earlieststages of scientificinvestigation before a priori justifiable statisticaltechniques become availableto groundsuch judgments.What we have seen however is that the emergingconsensus in the theory of induction indicates that the answer is "Always."Projectability judg- ments are always a posteriori, so we should always require the sort of semanticmachinery indicated by the theory of naturalkinds when our aimis inductionor explanation. This is, if I am right, the basic considerationfavoring the recent extrapolationsof the traditionaltheory of naturalkinds. Note that the extrapolationis justified by both sophisticatedempiricist and sophis- ticated realist conceptions of induction. Realists and empiricistscan even agree on a realist-soundingformulation of their common insight about kinds and induction. Kinds useful for induction or explanation must always "cut the world at its joints" in this sense: successful induction and explanation always require that we accommodate our categoriesto the causal structureof the world. Of course the empiricist has a Humean conception of the referenceto causal structurewhereas the realist has a metaphysicalone. Neverthelessit is a non-trivialfact that the same formula can be used to describe each conception. The differences in interpretationof the formula in question correspond preciselyto the differencebetween the distinctlyrealist treatment of the epistemology of the theory-dependenceof method and the sophis- ticated empiricists' induction-about-inductionaccount of the same phenomenon.What the two conceptions have in common the formula captures precisely:that in induction and explanationwe must refer to

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 140 RICHARD BOYD

kinds whose definitionsare specified a posterioriin deferenceto nature ratherthan nominally.

1.2. TheScope of the ExpandedTheory of Kinds What should the lesson be for the questionof the scope of the theory of natural kinds? At least prima facie the realist and the sophisticated empiricist should be led to conclude that that scope should be quite broad: that in defining a kind we should be requiredto defer to the world just in case and to the extent that reference to the kind in questionis to be part of an inductiveor explanatoryproject. In cases in which our concerns are largely with the establishmentof workable conventionsfor non-inductivepractice, deference to the world should be largelyunnecessary. Thus, for example, the definitionof the kind "gene"should possess few conventionalfeatures whereas the definitionof the kind "fishfork" should be largely arbitrary.It likewise follows that there should be kinds and categories whose definitions combine naturalisticand con- ventionalfeatures in quite complex ways. Considerthe notion of having been legally married in New York State. Complex psychological notions like consent, deeply linked to induction and explanation,are involved in the legal definition of marriage,but so are notions with a high degree of arbitrariness- like being a properlyordained member of the clergy. In consequence the proposal we are consideringdictates that we should expect that the definitionof the legal notion in question should have, as it does, a mix of conventionalelements and of elements in which a deferenceto the world is reflected.It follows that extensions of the traditionalaccount of naturalkinds should be appropriatejust to the extent that the kinds in question are employed for induction and explanation- just as the expansionistrealism to which Hackingobjects maintains.

2. THE THEORY OF NATURAL KINDS EXTENDED

2.0. Property-Clustersand NaturalKinds Hackingmaintains, following the empiricisttradition, that naturalkinds

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 141 should be defined by a set of necessary and sufficientconditions. He holds that in the case of kinds for which this conditionfails - property clusterkinds - what unites the propertiesin the clusteris humansocial decision rather than nature, so that the extension of the theory of natural kinds to property cluster kinds is inappropriate.It is an interesting question whether or not there are property-clusterkinds such that the unity of the relevant propertiesis exclusivelyor mainly conventional.What is importantfor our purposes is that there are a great many property-clusterkinds for which the opposite is the case: kinds such that the unity of the property-clusterwhich defines them is causalrather than conceptual. What I have in is somethingwhich should be congenialto the general project of the traditional theory of natural kinds. On that conception a naturalkind is associated causally with a large family of methodologicallyimportant properties. Even if the kind is thoughtof as being defined by a set of necessary and sufficient conditions its naturalnessis a reflection of a wider sort of propertycorrelation. It is naturalto inquirewhether in definingsome kinds we might defer more fully to natureand take the kind in questionto be defined by the larger family of correlated properties rather than by any special sub-set singled out as providingthe necessaryand sufficientconditions. If this possibility is acknowledgedthen it is reasonable to inquire whether there may be kinds so defined except that the relevant property correlationsare not perfect, so that the set of correlated properties functionsas a property-cluster. I have arguedelsewhere (Boyd 1988, 1989, 1990) that this possible situationis actual for an importantfamily of naturalkinds. I argue that there are a of scientificallyimportant kinds (properties,rela- tions, etc.) whose naturaldefinitions are very much like the property- cluster definitionspostulated by ordinary-languagephilosophers except that the unity of the propertiesin the defining cluster is mainly causal ratherthan conceptual.The naturaldefinition of one of these homeo- staticproperty cluster kinds is determinedby the membersof a cluster of often co-occurring properties and by the ("homeostatic")mecha- nisms that bring about their co-occurrence. It is an a posteriori theoretical question which of these properties and which of the homeostatic mechanisms count, and to what extent they count, in

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 142 RICHARD BOYD determiningmembership in the kind. In cases of imperfecthomeostasis in which some of the propertiesin the clusterare absent or some of the mechanismsinoperative it will sometimes happen that neither theo- retical nor methodological considerations assign the being classifieddeterminately to the kind or to its complement,with the result that the homeostaticproperty-cluster definition fails to specify neces- sary and sufficientconditions for kind membership.Both the property- cluster form of such definitions and the associated indeterminacyare dictated by the fundamentalepistemic task of employing categories which correspond to inductively and explanatorily relevant causal structures.In particular,the indeterminacyin extensionof these natural definitionscould not be remediedwithout rendering the definitionsun- naturalin the sense of being scientificallymisleading. The paradigm cases of natural kinds - biological species - are homeostatic cluster kinds. The appropriatenessof any particularbio- logical species for induction and explanationin biology depends upon the imperfectly shared and homeostatically related morphological, physiologicaland behavioralfeatures which characterizeits members. The definitionalrole of mechanismsof homeostasisis reflected in the role of interbreedingin the modern ; for sexually reproducingspecies, the exchangeof genetic materialbetween popula- tions is thought to be essential to the homeostaticunity of the other propertiescharacteristic of the species and it is thus reflectedin species definitions.The necessaryindeterminacy in extension of species terms is a consequence of evolutionarytheory, as Darwin observed: specia- tion depends on the existence of populationswhich are intermediate between the parent species and the emergingone. Any "refinement"of classificationwhich artificiallyeliminated the resultingindeterminacy in classificationwould obscure the central fact about heritablevariations in phenotype upon which biological evolution depends and would be scientificallyinappropriate and misleading. Thus some paradigmaticcases of naturalkinds, playingthe paradig- matic role of specifyingcategories apt for inductionand explanationin science, are counterexamplesto the claim that - at least on an ideal rationalreconstruction - naturalkinds must be defined by necessary and sufficientconditions. I conclude that the requirementthat natural kinds have such definitions is to be diagnosed as a holdover from traditionalempiricist conceptions of linguisticprecision which must be

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 143 abandonedonce it is agreed that kind definitionsmust conform to the (sometimesmessy and complex)causal structure of the world.

2.1. Social Construction and the Metaphysics of Social Kinds: The Metaphysical Innocence of Social Practice

Should the theory of naturalkinds also, contraHacking, be extendedto cover social kinds? We have seen that there are considerationsaccept- able both to realistsand to sophisticatedempiricists which suggestthat such an extensionis necessaryfor social kinds involved in inductionor explanation. Against these considerationsthere are two plausible counterargu- ments, each suggested by Hacking's insistence that the adoption of definitionsfor social kinds has a real but inscrutableinfluence on the propertiesof the social objects under study. According to the first, the contemplatedextension of the theory of naturalkinds gets the meta- physicswrong: social kind definitionsmust indeed be accommodatedto actual causal structurebut the dependence of the causal structureof social phenomena on our social theories and (especially) on our definitions of social kinds makes the causal structure which those definitions must match theory-dependentor mind-dependenta way which neither the realist nor the empiricistaccounts of naturalkinds can handle. The second counterargumentis epistemologicalrather than meta- physical:to be sure social kinds appropriatefor inductionand explana- tion would requirenaturalistic definitions, but the dependenceof social structureson our conceptionsof them is so profoundand so far beyond our comprehensionthat in some appropriatelydeep sense induction and explanation about social matters are impossible. I shall be con- cernedwith the firstof these counterargumentsin the presentsection. Scientificrealism is often contrastedwith varioussocial constructivist conceptions by saying that realists affirm, while constructivistsdeny, that the realityscientists study is "mindindependent" in the sense that it is independent of their theories and conceptions. Certain phenomena scientists study, especially but not exclusivelysocial phenomena,have some properties which depend causally on aspects of social practice which include the constructionof theoriesabout those very phenomena. In this sense, they are not "theory-independent".

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 144 RICHARD BOYD

Proposed definitions of scientific realism are often designed to accommodatethis fact, which realistsdo not deny. Often realismis said to imply that subjectmatter of science is "largelytheory-independent". This terminologysuggests that if the causal theory-dependenceof the objects of investigationin some science is much greateror deeper than expected, then realismregarding that science is compromised.On such an interpretationof realism,the influenceof our classificatorypractices on the properties of the social phenomena we classify might refute realism regardingsocial and thus underminethe extension of the realistconception of naturalkinds to the case of social kinds. Since empiricists also treat as theory-independent,the extension of an empiricist version of the theory of natural kinds would also be excluded. I have discussed this interpretation of realism at some length elsewhere(Boyd 1990). What I conclude there, and urge here, is that it is a mistake to see realists (and empiricists) as differing with constructivistsabout the extent of the theory (or mind) dependence of reality.Instead, the disagreementis over the natureof that dependence. Realists and empiricistsneed not deny that the adoption of theories (schemes of classifications,etc.) may have a profound causal effect on the subject matter of those conceptions. What they deny is that the adoptionof such conceptionsmakes any non-causal,logical or concep- tual contributionto the causal powers of the objects of study.For both the Carnap of "Empiricism, and "(1950) and the Kuhn of The Structureof ScientificRevolutions (1970) the adoption of certain theoreticalconceptions (L-truths,paradigms) can result in the of certain high level theoreticalclaims "aboutatoms." For Kuhn and other social constructivistswhat is involved is the actual social constructionof the causal propertiesof atoms.The world chemistslive in (or at least the world they study)embodies certain quite general sorts of causalphenomena because chemistshave adopted the paradigmthey have. For Carnap, by contrast, as for other empiricistsand for realists, human social practices,like the adoption of theories and classificatory schemes, are metaphysicallyinnocent: they affect the causal structureof the world only via the operation of intermediarycausal mechanisms which superveneon the causal structuresstudied by the variousspecial

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 145

sciences and not also in some additionalway studied only by philoso- phers practicing conceptual analysis. The worlds which would be studied by chemistswho accepted differentparadigms would not differ in general causal structureand would differ specificallyonly in ways which supervene causally on the differences in the resulting social practices.Note that it is no part of a realist or an empiricistanalysis to deny (or affirm) the inscrutabilityof the details of that causal supervenience. It is beyond the scope of this paper to defend the metaphysical innocence of social practices. Suffice it to say here that (1) it is an assumption central to both the realist and the empiricist traditions within which the extension of the theory of naturalkinds is contem- plated, (2) it appearsto be underwrittenby fundamentalconceptions of causal relations common to all of the establishedsciences, and (3) it appears itself to be required to underwrite some pretty obviously correct methodologicalprinciples regarding the role of conventionsin science (Boyd 1990). Tentatively accepting metaphysicalinnocence principle,I conclude that the influenceof classificatorypractices on the actual propertiesof the objects of study in the social sciences does not compromise the realist (or empiricist)conception of propertiesupon whichthe proposedextension of the theoryof naturalkinds depends.

2.2. ClassificatoryCausation and the Possibilityof SocialKnowledge It remainsto ask whetherthe causalinfluence of classificatorypractices makes genuine induction and explanationimpossible in social inquiry. Might the influence of our classificatorypractices be so great and so inscrutablethat our very attempts to classify social phenomena will render them unstudiable? If the metaphysical innocence of social practicesis assumed then for an importantclass of social and generalizationsthe answermust clearly be "no." The metaphysicalinnocence principle entails that the influence of classificatorypractices on causal structurealways superveneson ordi- nary causal mechanisms.If this is assumedthen we can be certainthat our current classificatoryprinciples in history do not influence the properties which the objects of our historical studies exhibited when they were extant.To the social constructivistphilosopher it might seem

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 146 RICHARD BOYD that the propertiesexhibited by past events and persons are constituted by our current theorizing,but this possibility is precluded by meta- physical principlescommon to the two philosophicaltraditions within the extensionof the theory of nat4ralkinds is contemplated.I conclude that, barringa successfuldefense of social constructivism,the projectof extendingthe theoryof naturalkinds to historicalkinds is secure. It might still be arguedthat the effects of classificatorypractices are so great and so inscrutableas to preclude non-historicalsocial knowl- edge, or perhaps that they are so inscrutable that even historical knowledgeis precluded,not by the inscrutabilityof the effects of our current classificatorypractices, but by the inscrutabilityof the social effects of the classificatory practices current during the historical periods we study. It is beyond the scope of this essay to discuss this proposal at length but I suggest that the burden of proof would lie stronglyon someone who held, for example, that the unknown social effects of classificatorypractice render problematical all generalizations aboutthe effectsof unemploymenton wagelevels in marketeconomies. I do not mean to deny that the effects of classificatorypractices are socially or politically or methodologicallyimportant. I am sure the opposite is the case. What I doubt is that those effects are so pervasive and inscrutable as to make social knowledge impossible. If social knowledgeis possible then quite general ,acceptable to the realist and the sophisticatedempiricist alike, dictate that we extend the traditionalconception of naturalkinds to social kindsas well.

2.3. Social Construction, One Last Time One response to all of the forgoingargumentation might be to suggest that the theory of naturalkinds - as developed in recent realism or sophisticatedempiricism - deserves extension to all those areas of knowledgeto which empiricistor realist, as opposed to constructivist, interpretationsare appropriate,but to hold that it is characteristicof social inquiry that it, unlike inquiry in the physical or (some of) the biological sciences, requires a constructivistinterpretation. A selective constructivismreflecting such a view seems to be an emergingtheme in the thinkingof many who reflect on the ideological determination,as well as the ideological impact, of social inquiry (and some biological

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions ENTHUSIASM FOR NATURAL KINDS 147 inquiryas well). I thinkthat there are deep reasonswhy this approachis unpromising. I have argued elsewhere (Boyd 1990) that by "realismabout" a subject area we should mean the doctrine that the characteristic intellectualachievement in that area involves the acceptance of state- ments which reflect,when understoodliterally, approximate knowledge of a reality which is logically (conceptually,etc.) independent of the theories,conceptual schemes, research interests, etc. which are adopted. In other words, we should understandrealism about an area of inquiry as the doctrinethat it producesknowledge to which the principleof the metaphysicalinnocence of social practiceapplies. The proposal to depart from realism (or empiricism)in favor of constructivismin the special case of social inquirywould then involve denying the metaphysicalinnocence doctrine for the special case of social theorizingwhile retainingit with respect to the effects of other sorts of social activity. This is a deeply unpromisingprospect. The justificationfor the metaphysicalinnocence doctrinelies in very general and very deep features of our scientific conception of causation.To accept the doctrine for almost all cases of social practice but to hold, for instance, that historical theorizing can create causal relations between past historical events, would be to adopt a postion of very doubtfulcoherence. One last point should be made regardingconstructivism about social inquiry.Sometimes when philosophersor others hold that social inquiry is a matter of "the social constructionof reality"they are best under- stood as holding not that social inquiry typically produces genuine knowledgeof real but socially constructedcausal relations,but instead as holding that the most typical productof social inquiryis not knowl- edge at all but ideology ratifyingexisting social patterns.It goes with this conception of some area of social inquirythat the definitionsof certain of its terms must be purely nominal ("Jewishphysics", "the British national character","I. Q."). If the distinction between realism and empiricismon the one hand and constructivismon the other lies in the acceptance or rejection of the metaphysicalinnocence doctrine, then this critique of social inquiry as ideological, far from being construc- tivist in the philosophicalsense, must rest on an empiricistor realist foundation;it is furthermorecompatible with the claim that the theory

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions 148 RICHARD BOYD of natural kinds should be extended to whatever social kinds are implicatedin genuinesocial knowledge, however rare it maybe. It is likewise true that the contemplatedextension of the theory of natural kinds is compatible with the recognitionthat, in cases where social knowledge is possible and where, in consequence, kinds must have a posteriori natural definitions, ideological and other social factors will often be a source of error in our efforts to formulatesuch defini- tions. I conclude that nothing in the methodologicallyand politically very important social embeddedness of our classificatorypractices precludesthe extensionof the theory of naturalkinds to social kinds an extensionin whose favor there are, in any , deep and indepen- dent epistemologicalarguments.

BIBLIOGRAPHY

Boyd, R., 1979. "Metaphorand Theory Change"in A. Ortony (ed.) Metaphorand Thought.Cambridge: Cambridge University Press. ,1982. "ScientificRealism and NaturalisticEpistemology." in P. D. Asquith and R. N. Giere (eds.) PSA 1980. Volume Two. E. Lansing:Philosophy of Science Association. ,1988. "How to be a Moral Realist."in G. Sayre McCord (ed.) . Ithaca: Press. ,1989. "WhatRealism Implies and WhatIt Does Not" Dialectica. 1990. "Realism, Conventionalityand 'Realism About"' in G. Boolos (ed.) Festschniftfor HilaryPutnam. Cambridge: Cambridge University Press. Carnap, R., 1950. "Empiricism,Semantics and Ontology."Revue internationalede philosophie,4th year. Goldman, A., 1967. "A Causal Theory of Knowing"Journal of Philosophy (LXIV): 357-372. a 1976. "Discriminationand Perceptual Knowledge" Journal of Philosophy (LXXIII):771-791. Goodman, N., 1973. Fact Fiction and Forecast, 3rd edition. Indianapolisand New York:Bobbs-Merrill. Kuhn, T., 1970. The Structureof ScientificRevolutions, 2nd edition. Chicago:Univer- sity of ChicagoPress. Putnam,H., 1975a. "The Meaningof ''."in H. Putnam,Mind, Languageand Reality.Cambridge: Cambridge University Press. - 1975b. "Languageand Reality."in H. Putnam, Mind, Languageand Reality. Cambridge:Cambridge University Press. Quine, W. V. O., 1969a. "NaturalKinds." in W. V. 0. Quine, OntologicalRelativity and OtherEssays. New York:Columbia University Press.

SageSchool of Philosophy CornellUniversity Ithaca,New York14853 U.S.A.

This content downloaded from 192.122.237.41 on Thu, 8 Jan 2015 09:38:55 AM All use subject to JSTOR Terms and Conditions