Can a reductionist be a pluralist?

DANIEL STEELDepartment of Philosophy, Michigan State University, 503 S. Kedzie Hall, East Lansing, MI 48824-1032,USA (E-mail: steel@msu.edu).

Received 24 September 2002; accepted 18 June 2003

Key words: HIV, Molecular Biology, Multiple-Realizability, Pluralism, Reduction, Reductionism,Unification

Abstract. Pluralism is often put forth as a counter-position to reductionism. In this essay, I argue thatreductionism and pluralism are in fact consistent. I propose that there are several potential goals forreductions and that the proper form of a reduction should be considered in tandem with the goal that itaims to achieve. This insight provides a basis for clarifying what version(s) of reductionism are currentlydefended, for explicating the notion of a “fundamental” level of explanation, and for showing how onecan be both a reductionist and a pluralist.

Introduction

At the end of his classic statement of the anti-reductionist position in biology, PhilipKitcher wrote, “despite the immense value of the molecular biology that Watson andCrick launched in 1953, molecular studies cannot cannibalize the rest of biology”(1994 [1984]: 398). Kitcher’s use of the word “cannibalize” gives an indication ofthe consequences that anti-reductionists fear would ensue were reductionism correct.Apparently, if reductionism were right, then all provinces of biology other thanmolecular biology would be, in principle at least, superfluous – of practical use onlybecause of human limitations. To one strongly attached to these allegedly superfluousdisciplines, this would be a dire consequence indeed. Not surprisingly, then,philosophers who advocate pluralism – roughly, the thesis that there is a plurality oflegitimate and autonomous levels of description and explanation of a given phenomena– often see themselves as staking out a position that stands in direct opposition toreductionism. The primary purpose of this paper is to argue that the matter is far morecomplex than it appears, and that, surprisingly, one can consistently maintain bothpluralism and reductionism.

I propose that reduction is an explanatory strategy that can be pursued in order toachieve a variety of different goals, and what form the reduction should take dependson its purpose.1 I argue, therefore, that there are no uniformly correct constraints onthe form of reductions, since different types of reductive explanation might be suitedto different objectives. I introduce four possible goals of reduction: ontologicalparsimony, unification, decomposition, and correction. I show how these four potential

© 2004 Kluwer Academic Publishers. Printed in the Netherlands.Biology and Philosophy 19: 55–73, 2004.

reductive goals can be used as the basis for a categorization of distinct varieties ofreductionism and to specify what reductionisms are appropriately associated with the“reductionist anti-consensus.”2 Since this reductionist position is significantly weakerthan some commonly critiqued versions of reductionism, I consider whether it has alegitimate claim to the title. My strategy is to connect reductionism to the notion ofexplanations drawn from a level of description that is more fundamental than others,to explicate the relevant sense of “fundamental” in terms of what I call correctiveasymmetry, and to show that the reductionism in question satisfies this condition.

Next, I turn to pluralism. There are three principles that I associate with this doctrine:that there are multiple legitimate strategies for representing nature (principle of multipleperspectives); that there is no ideal representation that is sufficient for all explanatorypurposes (non-completeness); finally, that distinct levels of explanation are autonomous(autonomy of levels). I argue that the principle of multiple perspectives is consistentwith even the most extreme version of reductionism, while non-completeness andautonomy of levels are consistent with the reductionist anti-consensus, but not withmore extreme forms of reductionism.

A motivating example: HIV replication

Rather than beginning with a detailed examination of reduction and pluralism in theabstract, I shall commence by briefly describing a scientific example – namely, HIVreplication – that will serve to illustrate and motivate the issues addressed in theremainder of the paper.

HIV are retroviruses, which are so-called because they reverse the normal flow ofinformation from DNA to RNA. The infection of a cell by HIV begins when the HIVvirus attaches to the cell at the CD4 receptor site and injects its genetic material intothe host cell. The genetic material of HIV consists of RNA, and when a retrovirusinfects a cell, its RNA serves as a template for the transcription of viral DNA, whichis then insinuated into the cell’s nuclear DNA. Once this occurs, the cell becomes afactory that produces HIV. The viral DNA integrated into the cell’s genetic materialcodes for new viral RNA and proteins necessary for the functioning of the retrovirus.These materials are then assembled in the cytoplasm and new retroviruses bud fromthe cell membrane, ultimately destroying the cell if they do so in sufficiently largenumbers. This mechanism is often depicted by diagrams like the following (cf.Kalichman 1998: 16; Stine 2000: 64). Notice that, aside from a few references toDNA and RNA, resources from the molecular level are hardly invoked at all in thisrepresentation. Moreover, at this level of description one can speak of the HIVreplication mechanism, whereas once one delves even slightly into the finer moleculardetails one is forced to distinguish between versions of the mechanism found in distinctstrains. The more molecular detail one gives, the more such distinctions must be made.Furthermore, these differences are often of real significance, since they may be relevantto whether the strain in question is resistant to treatment or to the host’s susceptibilityto infection.

56

Figure 1. HIV replication.

Let us consider an example that illustrates this point. There are some people whohave been multiply exposed to HIV without becoming infected, and hence withoutdeveloping AIDS. Such individuals are labeled “EU” for exposed but uninfected.However, a person might be EU for a variety of reasons, some involving no intrinsicresistance to HIV. For example, some EU individuals might have simply been thebeneficiaries of pure, dumb luck; others may have been exposed only to a very mildform of the virus. Thus, it is of interest to determine which EU individuals, if any,possess a genuine HIV resistance. In the later 1990s, a great deal of interest wasstimulated by the discovery of a genetic mutation found in a few EU individuals thatconferred a significant level of resistance to HIV.

Even a cursory recounting of this story requires a much more detailed, and lessunified, description of the HIV replication mechanism than that provided above.Distinct HIV strains sometimes have affinities for different types of cell. T-helpercells are the best-known targets of HIV, but a type of phagocyte, namely themacrophage, is also prone to HIV infection. Indeed, in the early and non-symptomaticstages, HIV infection is almost exclusively restricted to macrophage tropic (M-tropic)HIV, with the emergence of T-tropic HIV and the widespread infection of T-helpercells occurring later in the progression of the disease (Zhu et al. 1993). There isevidence that the switchover from M-tropic to T-tropic HIV, which is generally morevirulent, is a cause of the onset of AIDS (Glushakova et al. 1998; Maas et al. 2000).The differing affinities of M-tropic and T-tropic HIV result from the fact that the twostrains utilize distinct co-receptors to enter into the host cell.

57

Although the presence of the CD4 receptor is generally sufficient for an HIV virusto attach to a cell, it is not sufficient for the entry of the viral core into the cytoplasm.In 1996, it was discovered that distinct co-receptors present on macrophage and onnon-circulating T-helper cells3 play an essential role in the entry of viral material intothe host cell (Deng et al. 1996; Dragic et al. 1996). The co-receptor in the case ofnon-circulating T-helper cells is called CXCR4, and its counterpart for macrophagesis known as CCR5. M-tropic HIV utilizes CCR5, while T-tropic HIV depends uponCXCR4, thereby accounting for the difference in affinities of the two strains. Normally,the CCR5 and CXCR4 co-receptors serve as binding sites for ß-chemokines, whichfunction to attract lymphocytes to areas of inflammation. ß-chemokines are classifiedaccording to their molecular structure, in particular, according to the position of theamino acid cysteine (Stine 2000: 140). Thus, “CC” indicates a peptide chain beginningwith two adjacent cysteines, while “CXC” indicates one beginning with a pair ofcysteines separated by any other amino acid. The co-receptors, then, draw their namefrom the chemokine that binds to them.

What tipped off researchers about the role of CCR5 as a co-receptor for HIV wasthe in vitro experimental results show that large doses of ß-chemokines could inhibitHIV entry into macrophages and T-cells (Deng et al. 1996: 662; Dragic et al. 1996:667–668). Given that the CXCR4 had already been identified as a co-receptor forT-tropic HIV, and since the genes for CCR5 are of the same family as those of theCXCR4, researchers inferred that M-tropic HIV utilized a CCR5 chemokine receptorto gain entry into host cells (Deng et al. 1996: 662; Dragic et al. 1996: 669–671).

About the same time, it was found that some individuals who had not becomeinfected with HIV despite repeated exposures possessed defective CCR5 co-receptors(Samson et al. 1996; Liu et al. 1996). Cells from these individuals proved highlyresistant to infection by M-tropic HIV in vitro (Liu et al. 1996: 367). The absence ofthe normal CCR5 co-receptor was linked to a homozygous mutation, in which 32base pairs in the ordinary gene coding for the co-receptor were deleted (Liu et al.1996). As noted in the foregoing section, M-tropic HIV predominates in the early andasymptomatic stages of infection. Thus, by interfering with the replication of M-tropicHIV, the 32 base pair deletion substantially inhibits the progression of HIV infection.Since the homozygous mutation affecting the CCR5 co-receptor appears to produceno other abnormal phenotypic effects, the discovery of its important role in theprogression of HIV infection provides an inspiration for the development of newtherapies. And, in fact, research is currently underway on several compounds designedto block HIV fusion to either the CCR5 or CXCR4 co-receptor.

Both defenders of reductionism as well as advocates of pluralism would find muchin the above example to support their positions. Reductionists would note the centralimportance of molecular details for the understanding of HIV replication, andespecially for explaining exceptions to the general description of this process, as thediscovery of the mutation affecting the CCR5 co-receptor illustrates. If we want toknow why a particular HIV succeeded or failed to infect a given cell, the reductionistwould assert, the molecular details are the place to look for the answer. In contrast,pluralists would point out the mixture of distinct levels of description in the above

58

account, and would observe that, owing to the heterogeneity of the underlyingmolecular details, higher-level descriptions cannot be readily translated into molecularones. Consequently, the pluralist would conclude that there is little plausibility to theproposal that a complete account of HIV replication could be provided at the molecularlevel, or any other level. To make any progress towards clarifying and perhapsresolving this disagreement, we need to examine the concepts of reduction andpluralism with care.

Four motives and three desiderata

Reduction is an explanatory strategy that can, in different contexts, be pursued toachieve distinct goals. The four potential motives for reduction that I shall considerare the following:

Ontological Parsimony: To show that where there appears to be two types ofentity there is in fact only one, more fundamental type.Decomposition: Given a feature possessed by a certain set of systems, to showthat the systems’ parts and their interactions, described at a specified level ofdetail, are sufficient to explain the feature.Unification: To demonstrate that a wide array of distinct generalizations andobservations can be explained by a small number of more fundamentalgeneralizations.Correction: To identify and explain exceptions to less fundamentalgeneralizations using generalizations and details drawn from a more fundamentalrealm.4

.

These motives are not intended to be exhaustive. Nor do I suggest that all of thesefour goals are always reasonable ones to pursue. In deciding whether a reductiveresearch strategy is appropriate in a particular scientific context, one should askwhether the intended goal of the reduction is worthwhile and, if it is, whether reductionis an effective way to achieve it. Clearly, the answers to these questions can vary fromcase to case.

The classic example of a reduction that achieves the goal of ontological parsimonyis the kinetic theory of heat, which is usually taken to show that heat is not a distinctsubstance, as Carnot’s caloric theory maintained. A stock example of a reduction thataccomplishes unification is the explanation by Newtonian Mechanics of a range ofless fundamental laws such as Galileo’s law of free fall and Kepler’s laws of planetarymotion. This case also illustrates correction, since Newtonian Mechanics can be usedto identify and explain exceptions to Kepler’s laws. Reductions that aim to attaindecomposition correspond closely to what Sahotra Sarkar terms “strong reduction”(1998: 43–45). Sarkar defends the controversial proposition that strong reductionsare in the process of being carried out in molecular genetics (1998: chapter 6).

It will be helpful to examine the four reductive motives listed above along withconditions often demanded of reductions. Consider the following three:

59

1. The distinct concepts of the more and less fundamental realms must be linkedby bi-conditional bridge laws or “synthetic identities.”

2. Reduction must be a relation that holds among theories, where theories areunderstood as consisting of, or at least being associated with, a relatively smallnumber of generalizations capable of accounting for a broad range ofphenomena.5

3. Aside from correspondence rules linking the two realms, a reduction mustexplain the aspect of the less fundamental realm solely in terms of concepts andprinciples drawn from the fundamental one. For instance, a molecularexplanation that presupposes various aspects of cellular context that are notdescribed in molecular terms is not a reduction.

.

The model of reduction that anti-reductionists often take to be the standard accountsatisfies all of these conditions. This is what might be called the “layer-cake model.”6

The classic presentation of the layer-cake model of reduction is found in PaulOppenheim and Hilary Putnam’s (1958) essay, “The Unity of Science as a WorkingHypothesis.” The layer-cake model presupposes that contemporary science justifiesdepicting nature in terms of a series of levels from most to least fundamental. InOppenheim and Putnam’s essay, these levels were: elementary particles, atoms,molecules, cells, multicellular orangisms, and social groups (1958: 9–10). The levelsare intended to be such that the entities at any level (except for the most fundamentalone) are fully decomposable into the entities at the level below. For example, moleculesare decomposable into atoms, and atoms into elementary particles. It was presumed,in addition, that there is (or would eventually be) a set of theories for each level andthat the terms in the theories at each level would refer exclusively to the entities atthat level. Given this framework, the layer-cake model asserts that reduction consistsof a deduction of the higher-level theory from the lower-level one with the aid ofbridge laws that equate any distinct higher-level kinds with complexes of lower-levelkinds. The layer-cake model, then, endorses all of 1 through 3; that is, according toit reduction is a relation between theories, requires synthetic identities, and onlyresources drawn from the fundamental realm may be used in the derivation.

Although the layer-cake model has been called “the standard account of reduction”(Kincaid 1990: 576), it is rather different from that proposed by Ernest Nagel, whoseaccount of reduction is often cited as the classic source on the topic.7 Indeed, ofrequirements 1 through 3, Nagel’s model insists only upon 2. Nagel did not requirethat the correspondence rules connecting the reducing to the reduced theory besynthetic identities or bi-conditionals; he is, in general, rather flexible as to what mightqualify and explicitly allows one-way conditionals (cf. Nagel 1979: 105–107). Theinsistence on synthetic identities was a modification of Nagel’s model introduced bysubsequent commentators, particularly Schaffner (1967).8

Nagel’s model also differs from the layer-cake model in not requiring that the morefundamental theory exclusively refer to entities that are proper parts of those referredto by the less fundamental theory. This is most evident in the case of what Nagelrefers to as homogeneous reductions, that is, reductions in which the reduced theorycontains no terms not already present in the reducing theory (1961: 342). Nagel’s

60

stock examples of homogeneous reductions are the derivations of Galileo’s law offree fall and Kepler’s laws of planetary motion from Newtonian Mechanics (cf. 1979:98). Moreover, there is nothing in Nagel’s model to rule out the possibility that thereducing theory might span several levels of description, and thus violate requirement3.9

Consider how the differences between Nagel’s model and the layer-cake modelregarding requirements 1 through 3 translate into differences with respect to the fourreductive motives. A reduction that fits the form of the layer-cake model wouldsucceed in achieving the first three of the motives listed above. Synthetic identitieswould support the claim that any entity referred to by the reduced theory is identicalto a complex of entities, or features of such complexes, described at the morefundamental level. Hence, such a reduction would achieve ontological parsimony,since it would show that no entity referred to by the higher-level theory constitutesan independent type of substance. Likewise, since the explanation proceeds solelyfrom a more fundamental level of description, which is so characterized in virtue ofreferring to proper parts of the entities of the higher-level theory, the goal ofdecomposition would also be achieved. Since the layer-cake model assumes thatreduction is a relationship between theories, understood as a reasonably small numberof principles that encompass a broad range of phenomena, the goal of unification ismet as well. In contrast, a reduction that fulfilled the strictures of Nagel’s model wouldachieve the goal of unification – since what does the reducing is a theory – but notnecessarily ontological parsimony or decomposition. For instance, these two motivesare clearly irrelevant to the reduction of Kepler’s laws of planetary motion toNewtonian Mechanics.

In addition, a consideration of distinct motives for reduction illuminates the inherentimplausibility of treating the layer-cake model as the only model of reduction.10 Sincenot all attempts at reduction intend to achieve the goals of ontological parsimony,decomposition, and unification, it is unreasonable to insist that reductions in generalmust satisfy conditions relevant to the attainment of all three goals. For instance,requirement 1, which demands that the terms of the reduced and reducing theories beconnected by synthetic identities, is closely tied to the goal of ontological parsimony.Yet reductionists and anti-reductionists are united in their rejection of vitalism; thestandard anti-reductionist position is physicalist anti-reductionism. Synthetic identities,furthermore, have little relevance to the other three reductive goals. Hence, there isno reason to suppose that reductions in molecular biology in general must haveontological parsimony as one of their goals, and consequently it is unreasonable toinsist upon synthetic identities or bi-conditional bridge laws in biological discussionsof reduction.11

Reductionisms

By the term “reductionism,” I understand a substantive thesis about what sorts ofreductions are possible in which areas of science. Thus, reductionism should be

61

distinguished from models of reduction, which make claims about what characteristicsan explanation must have in order to qualify as a reduction. A model of reductionmight be judged correct or incorrect independently of any particular stance onreductionism. For example, the dispute between David Hull (1972, 1974) and KennethSchaffner (1969, 1993b: 437–445) concerning the alleged reduction of Mendeliangenetics to molecular genetics assumed Schaffner’s (1967) model but turned on itsapplication in this particular case. Given the four motives for reduction presentedabove, a variety of possible versions of reductionism can be distinguished. The mostextreme is what can be called hegemonic reductionism, according to which the capacityof an explanation given at a higher level to achieve any of the four goals is equaled,or surpassed, by an explanation provided at a more fundamental level. AlthoughOppenheim and Putnam’s classic (1958) paper is plausibly interpreted as anendorsement of hegemonic reductionism, it is doubtful that the position has any currentphilosophical defenders.

The standard objection to reductionism, the multiple-realizability argument, providesa good basis for rejecting hegemonic reductionism. Although this argument can beexpressed in various ways (cf. Fodor 1975; Kitcher 1994, 1999; Rosenberg 1985:93–96), I propose that it be interpreted as endeavoring to show that molecularexplanations are sometimes (perhaps, often) less unified than explanations providedat a higher level. Putting the matter in somewhat different language, the central claimis that heterogeneous collections of molecular mechanisms sometimes underlie whatare, from a higher-level perspective, single generalizations. Hence, an explanationthat replaced the higher-level generalization with its underlying molecular detailwould suffer a loss of unifying power. Hilary Putnam’s classic exposition of themultiple-realizability argument maintains that a geometrical explanation of why asquare peg won’t fit into a round hole is superior to one couched in terms of molecularstructure in virtue of being more general (1975: 296). Kincaid also puts themultiple-realizability argument in terms of unification, asserting that, “Attempts toexplain in purely biochemical terms will tend to see diversity where there is importantunity” (1990: 587).

The multiple-realizability argument has attracted a great deal of criticism, muchof which I think is legitimate. However, these criticisms do not show thatmultiple-realizability fails to pose a genuine objection to hegemonic reductionism;rather, they show that there are weaker forms of reductionism that the argument doesnot undermine. For example, consider a doctrine that one might call mitigated unifyingreductionism: molecular explanations of higher-level biological phenomena aresometimes, though not always, unified. Several authors have argued, in effect, thatthe multiple-realizability argument does not show that mitigated unifying reductionismis false; these critics charge that anti-reductionists have exaggerated the extent towhich multiple-realizability is a genuine problem, and thereby have overstated theheterogeneity of molecular explanations.

For example, Joseph Robinson (1992: 465) takes this line of argument in hisresponse to Kincaid (1990). The response is also pursued by Sahotra Sarkar withrespect to the relationship between molecular and Mendelian genetics (1998: 159–168),while William Bechtel and Jennifer Mundale (1999) make an analogous argument

62

for neuroscience and psychology. Similarly, Schaffner (1993a, 1993b: 463–466)argues that, although molecular biology lacks an all-encompassing theory, it possesses“middle-range theories.”12 Generalizations in molecular biology, often described asmechanisms (cf. Machamer, Darden and Craver 2000), differ from laws as traditionallyconceived in virtue of not taking the form of universally quantified generalizations,being historically contingent on the course evolution, and being of spatiotemporallylimited scope. Nevertheless, some molecular mechanisms – for instance, genereplication in eukaryotes – are quite prevalent across a broad range of biological taxa.Thus, it is arguably the case that some mechanisms provided by molecular biologycan serve as the basis for explanations that achieve a significant measure of unification.

A second line of criticism of the multiple-realizability argument is that reduction,and explanation in general, can serve purposes other than unification.13 I take this tobe Waters’ point when he asserts that, “the unificationist criterion for explanation isimplausible when invoked within the nitty-gritty details of genetics” (1990: 136).More recently, Elliott Sober has argued in a similar vein (1999: 549–551), maintainingthat while the lower-level account may not be unified it nevertheless explains. Soberasserts, moreover, that it is “a matter of taste” as to whether the unified but shallowor the heterogeneous but deep explanation is preferable (1999: 550–551). Sober alsoargues that there are sometimes sound reasons for preferring an explanation thatincludes molecular detail at the expense of unity, especially if this detail makes itpossible to correct the higher-level generalization (1999: 555–556). Sober’s argumentis nicely illustrated by the HIV replication example described in section 2, in whicha more heterogeneous, less unified molecular explanation was clearly preferable fromthe perspective of explaining why some individuals possess a heightened resistanceto HIV infection.

The point of these arguments is that one can have reduction without unification,and that there is sometimes good reason to desire such reductions. The type ofreduction being defended here can be characterized as follows: for any feature of acomplex system, there is a reduction of that feature which achieves decomposition,though not necessarily unification. This is similar to what Jerry Fodor calls“token-token reductionism.” Token-token reductionism only asserts that, in eachparticular case, the parts and their interactions can, in principle, explain the featuresof the whole. One important reason why token-token reductionism is of interest is,as Sober observes, that it entails that knowledge of the underlying details will allowfor corrections to higher-level generalizations. This point brings up a forth species ofreductionism, namely, corrective reductionism, which states that molecularexplanations of higher-level generalizations achieve the goal of correction. Token-tokenreductionism entails corrective reductionism, since if every instance can be explainedat the molecular level, then any exception to any higher-level generalization can alsobe thus explained. The entailment does not go in the opposite direction, however,since correction might be had without decomposition – as the example of NewtonianMechanics and Kepler’s laws of planetary motion illustrates.

In sum, the reductionist anti-consensus can be interpreted as maintaining theconjunction of mitigated unifying reductionism and token-token reductionism, along

63

with the observation that token-token reductionism entails corrective reductionism.But it might be objected that the only reductionism that should count as such ishegemonic reductionism. For example, according to Kincaid, the only legitimateinterpretation of reduction is one that asserts that:

One theory reduces another when it can do all the explanatory work of the reducedtheory … If there were good reasons to think that molecular biology and statisticalmechanics could not do all the explanatory work of their higher-level counterparts(and there is), then whatever they have achieved, it is not reduction. To claimreduction while admitting explanatory incompleteness is to make the issue atrivial semantic one. It is not. (1997: 5)

.

Although Kincaid is right that it would be pointless to defend reductionism merelythrough a redefinition of terms, it is equally the case that one ought not to criticizereductionism by knocking down a straw man. As we saw above, the layer-cake modelof reduction, which Kincaid assumes as the standard account (1990: 576; 1997: 50),is more stringent than Nagel’s model, has few if any current defenders, and isimplausible considered on its own merits. Once one distinguishes betweencausal-mechanical explanations that aim to elucidate underlying processes andtheoretical explanations designed to achieve unification, it is plausible that a reductionmight accomplish one of these explanatory purposes but not the other.

But if one can be a reductionist without advocating hegemonic reductionism, it isfair to ask what distinguishes reduction from other sorts of explanatory relationshipsthat might hold between distinct levels of inquiry. Surely, examinations of aphenomenon from diverse perspectives can be mutually illuminating in a variety ofways, but not all of this can count as reduction if the term “reduction” is to meananything. On what basis, then, can a non-hegemonic reductionism lay a genuine claimto the title?

Corrective asymmetry

Reduction rests on the idea that some levels of explanation are more fundamentalthan others; for example, hegemonic reductionism asserts that there is a level ofexplanation that is most fundamental in virtue of being able to achieve any goalattainable by any explanation provided at any other level. Clearly, one who rejectshegemonic reductionism, but who nevertheless claims to defend a reductionism ofsome sort, requires a different interpretation of “fundamental.” I propose that the keynotion here is corrective asymmetry: resources from the fundamental level arenecessary to correct explanations provided at other levels, but not vice versa.

If token-token reductionism is true, then molecular biology is correctivelyasymmetric with respect to higher levels of biological description. In the case of HIVreplication, for example, the molecular details correct higher level descriptions of theprocess. As we saw, the ability of molecular biology to correct and refinegeneralizations stated at higher-levels, even when no unified molecular explanation

64

is in the offing, was the basis of one criticism of the multiple-realizability argument.But in contrast, token-token reductionism entails that resources drawn from higherlevels are never necessary to correct molecular explanations of particular biologicalevents. Notice that this does not mean that such corrections could never be stated inhigher-level terms; rather, the claim is that any correction stated in such terms couldbe replaced by one drawing solely upon molecular resources. Let us consider thismore carefully.

Consider how one might explain why a certain strain of HIV is resistant to aparticular anti-retroviral drug, for example, a class known as non-nucleoside reversetranscriptase inhibitors (cf. Stine 2000: 85–87). These drugs interfere with HIVreplication by binding to the enzyme reverse transcriptase, which catalyzes the reversetranscription of the viral RNA to viral DNA that is then incorporated into the DNAof the host cell. By binding to reverse transcriptase, non-nucleoside reversetranscriptase inhibitors prevent it from carrying out its normal function. However,there are strains of HIV that are resistant to such drugs as a result of possessingmutations that alter the molecular structure of reverse transcriptase, and in some casesthe relevant changes in the base pairs of the viral RNA are known. It is clear that thereis a straightforward selective explanation of the prevalence of such mutant strains inindividuals treated with non-nucleoside reverse transcriptase inhibitors. Does thisconstitute a non-molecular explanation of an exception to the usual molecular accountof HIV replication and hence a counter example to corrective asymmetry?

It is not difficult to argue that the answer to this question is no. Consider theparticular biological events of which one would claim a molecular explanation in thiscase, for instance, the infection, or failure of infection, of a given cell by a particularHIV. The evolutionary explanation described above does not claim to correct suchexplanations, since, if token-token reductionism is true, then the ability or inabilityof the HIV to infect the cell depends on the molecular facts of the case. Moreover,the prevalence of one strain of HIV over another in an HIV+ person results from thesummation of a large number of such events, each molecularly explicable. Of course,the explanation that appeals to natural selection is more unified than the one thatprovides the gory molecular details of each case, but that does not conflict withtoken-token reductionism or corrective asymmetry.

Another concern is that molecular explanations presume a context that ischaracterized in higher level (e.g., cytological) terms. For example, a description ofHIV replication presupposes the existence of cells of several types and their organelles,as well as the larger organ systems (e.g., lymphatic) in which they occur. But variationsin these contextual features might account for exceptions to the usual molecularprocesses. However, there is a simple response to this concern. If token-tokenreductionism is true, then the relevant contextual features can be described in molecularterms in each individual case; hence, the higher level description is not necessary toexplain the exception.

Corrective asymmetry is also exhibited by reductions that achieve explanatoryunification. For example, Newtonian Mechanics identifies and explains exceptionsto Kepler’s laws of planetary motion, but Kepler’s laws do not return the favor. Thus,corrective asymmetry explicates a shared sense of “fundamental” operative in

65

reductions that achieve unification and those that accomplish decomposition. In boththe Newtonian and the HIV replication examples, the fundamental level is the onethat has, as it were, the last word about what happens. Consequently, using correctiveasymmetry as a criterion for what distinguishes reduction from other sorts ofrelationships between distinct explanatory levels has the appealing feature of beingable to account for how genuine reductions may come in several distinct forms. Noticethat such a plurality of forms of reductive explanations does not fit comfortably withhegemonic reductionism.

Hence, an advocate of token-token reductionism would qualify as a reductionist ifcorrective asymmetry is taken as a mark of reductionism. However, the same cannotbe said of mitigated unifying reductionism. The fact that some molecular explanationsachieve a significant measure of unification does not entail that there is a correctiveasymmetry between molecular biology and higher levels of biological description.Mitigated unifying reductionism, then, is of interest primarily as an appendage oftoken-token reductionism, serving to blunt the effect of the multiple-realizabilityargument. In what follows, then, token-token reductionism is the reductionism I willargue is consistent with pluralism.

Pluralism

Pluralism is often thought of as being directly opposed to reductionism. Whilereductionism asserts that there is a privileged, fundamental level of description,pluralism maintains that there are multiple autonomous perspectives from which thesame phenomena can be studied. However, I will argue that token-token reductionismis in fact consistent with pluralism. In order to make that case, it is necessary to saymore about what pluralism is. Several authors have defended pluralism (cf. Dupré1993; Cartwright 1999; Longino 2000, 2002a, 2002b; Kitcher 2001; Mitchell 2002),and although there are some differences of detail and emphasis among them, I thinkthat the following three principles are a good characterization of their core position.

Principle of Multiple Perspectives: There are multiple legitimate strategies forrepresenting nature.Non-Completeness: There is no ideal representation that is sufficient for allexplanatory purposes.Autonomy of Levels: Distinct levels of explanation are autonomous.

.

The first two of these principles are succinctly articulated by Kitcher, who writes:

The pluralism I propose consists of the following claims: (1) there are manydifferent systems of representation for scientific use in understanding nature; (2)there is no coherent ideal of a complete account of nature …14 (2002: 570)

.

Clearly, (1) is a statement of the principle of multiple perspectives, while (2) isnon-completeness. For his part, John Dupré encapsulates the pluralism as follows:

66

The most general positive doctrine I shall advocate is pluralism: first, inopposition to an essentialist doctrine of natural kinds, pluralism as the claim thatthere are many equally legitimate ways of dividing the world into kinds, a doctrineI refer to as “promiscuous realism”; and second, in opposition to reductionism,pluralism as the insistence on the equal reality and causal efficacy of objectsboth large and small. (1993: 6–7)

.

Dupré’s “promiscuous realism” is a version of the principle of multiple perspectives,while I interpret the second of his two claims as a statement of autonomy of levels.15

There are, I believe, two primary motivations for pluralism: one pragmatic and theother ontological. The pragmatic motivation rests upon the sensible notion that, owingto differences in goals and interests, one phenomenon can be legitimately studiedfrom a variety of perspectives. Given that there is no one objectively correct set ofgoals and interests to have (cf. Kitcher 2001: chapters 4, 5, and 6), it follows thatthere is no one objectively correct perspective from which to pursue one’s inquiries,and that it is doubtful that a complete representation of the world is possible. A“complete representation” in the sense at issue in non-completeness would be onethat would suffice (in principle) for the achievement of any purpose that any otherrepresentation could accomplish. Kitcher supports non-completeness via an analogywith maps (2001: 55–63). Maps are representations designed to serve certain purposes,and although there may be a map that is sufficient for a given set of aims, it isimplausible that there could be a map of (say) the earth that would suffice for all goalsthat an earthbound traveler might conceivably have (2001: 60). By analogy, scientificrepresentations are devised for particular ends, so we should be likewise skepticalthat there is, even as an ideal, a single representation that would serve all ends.

The ontological motivation for pluralism appeals to features of the world, especiallyits complexity. For example, according to Mitchell, “The complexity of nature andthe idealized character of our causal models to explain that complexity conspire toentail an integrated pluralistic picture of scientific practice” (2002: 67). The idealizedcharacter of representations is presumably related to both the complexity of thephenomena (it is not possible to include all relevant factors) as well as pragmaticconcerns (one might be interested in only one aspect of the phenomenon).

So, can one consistently be both a reductionist and a pluralist? The answer dependson the type of reductionism one has in mind. Let us begin with hegemonicreductionism; clearly, this ought to be inconsistent with pluralism. And it is easy tosee that it comes into conflict with non-completeness, the proposition that there is noone complete representation of nature. For if hegemonic reduction is correct, thenany legitimate explanatory purpose that one wishes to achieve can (in principle) beattained via the specified fundamental level. Hence, the representation provided atthe fundamental level would constitute the complete account whose existence (andcoherence) is denied by non-completeness.

Notice, however, that hegemonic reductionism is consistent with the principle ofmultiple perspectives. The hegemonist could happily agree that, for obvious practicalreasons, a variety of representational strategies are expedient and hence legitimate:

67

given human limitations it is often impossible to study a phenomenon from theperspective of fundamental theory. Consequently, the hegemonic reductionist canaccept that there are many distinct, legitimate ways of conceptualizing nature. Ofcourse, hegemonic reductionism maintains that there is one most fundamental levelof description, but there is no apparent reason why only the most fundamentaldescription should be regarded as legitimate. In sum, hegemonic reductionism isinconsistent with pluralism, but not in virtue of conflicting with the principle ofmultiple perspectives. Since hegemonic reductionism is logically stronger thantoken-token reductionism, this is a useful conclusion, for if P entails R, and P isconsistent with Q, then R is also consistent with Q. As a result, token-tokenreductionism is also consistent with the principle of multiple perspectives.

The issue, then, is whether token-token reductionism is consistent withnon-completeness and autonomy of levels. There is a straightforward argument thattoken-token reductionism is consistent with non-completeness. Token-tokenreductionism asserts, in effect, that it is always possible (in principle) to provide acausal explanation of a particular biological event at the molecular level. This doesimply that representations at this level are sufficient for a particular set of explanatorypurposes, namely, causal explanations of particular biological events and such things(like correction) that might ensue from these explanations. However, it does not entailthat all explanatory aims can be thus achieved. In fact, that is the point illustrated bythe multiple-realizability argument. For example, an explanation of the prevalenceof a drug resistant strain of HIV provided in terms of natural selection is more unifiedthan one provided exclusively in terms of the molecular details of the replication, orfailure of replication, of each particular HIV. Hence, token-token reductionism iscompatible with non-completeness. To return to Kitcher’s “many maps” analogy, theclaim that there is no ideal map that suffices for all purposes does not rule out thepossibility that there is a map that (ideally) suffices for some particular purpose.

Moreover, there is no good reason to dismiss the virtues of the more unifiedexplanation as desirable merely in virtue of being linked to human interests andcognitive limitations in a way that case-by-case causal explanations are not.16 Causalexplanations in general are closely tied to human interests owing to the usefulinformation they provide us about the results of interventions in our environment. Inaddition, what information is regarded as relevant in a causal explanation dependscrucially on what, in a given context, are regarded as stable background conditionsand as variable causal factors. The more unified explanation, meanwhile, can claimto identify patterns that are actually present and, as well, that the knowledge of thesepatterns is of interest not only from a theoretical perspective but also from the pointof view of practical interventions. For example, the explanation based on naturalselection enables one to predict that resistant strains are likely to arise, and evensuggests some approaches for dealing with the problem, such as therapies involvingthe combination of several distinct anti-retroviral drugs.

Let us turn, then, to the autonomy of levels. In order to decide whether this principleis consistent with token-token reductionism, it will be necessary to clarify just what“autonomy” amounts to in this context. I suggest that it means the following two

68

things. First, there is no level of description to which all others are reducible; second,the properties described at the various levels are causally efficacious, or causally real.I take the second of these two points to be what Dupré intends when he says thathigher-level properties are not “causally inert” (1993: 101). Let us consider these twoaspects of the autonomy of levels more carefully.

There are, as we have seen, several distinct senses of “reduction”; hence, whetherthe claim that there is no level of description to which others are reducible conflictswith token-token reductionism depends on which sense of “reduction” is intended.For example, if the relevant sense of “reduction” in autonomy of levels is token-tokenreductionism, then there is clearly a conflict. However, I think that there is goodreason to think that one could legitimately claim to be a pluralist while eschewingthis interpretation. The reason for this is that the primary motivation provided for theautonomy of levels is the multiple realizability argument, which (as we saw in section4) is an argument against the possibility of reductions that achieve unification, notagainst token-token reductionism. This provides reason to think that the relevant senseof “reduction” in autonomy of levels is unifying reduction. Rejecting token-tokenreductionism, in contrast, requires maintaining that there are strongly emergentproperties, that is, properties of a whole that cannot be explained by its parts and theirinteractions even on a case-by-case basis. There are well-known puzzles associatedwith the proposition that there are strongly emergent properties,17 and at least onepluralist, namely Kitcher, is clearly uncomfortable with such metaphysicalexcrescences (cf. 2002: 571). A pluralist, then, can interpret the claim that there is nolevel to which all others are reducible to be a claim about unifying rather thantoken-token reduction.

Let us turn, then, to the second component of autonomy of levels, namely, theproposition that higher-level properties are causally real. Clearly, we need to beginby saying what “causally real” means in this context. I propose the following simpledefinition: the property P is causally real if and only if there are correct generalizationsasserting that P is a cause of other properties. Given this definition, token-tokenreductionism, and even hegemonic reductionism, is consistent with the causal realityof higher-level properties. For example, there may be true causal generalizationsconcerning properties of a brick (its hardness, mass, etc.) even though those propertiesare explicable at the level of its molecular structure. No reductionist need deny thatthat dropping a 1 kilogram brick on your foot is a cause of pain. The same pointapplies to entities with more complex internal structures: to explain a higher-levelproperty of a system is not to deny its causal reality.

One who advocates token-token while rejecting hegemonic reductionism couldtake the above reasoning a step farther. Owing to multiple-realizability, there may betrue causal generalizations stated in higher-level terms for which no manageablecorresponding molecular generalization exists, which means that there are causallyreal properties for which there is no manageable, general definition in molecularterms. Hence, an advocate of token-token reductionism could agree that somethingof explanatory value would be lost if the higher-level generalization were dispensedwith, as was illustrated by the example of selective explanation of the emergence of

69

the resistant strain of HIV. Nevertheless, token-token reductionism entails that everyinstance of the generalization can be given a molecular explanation, and moreoverthat any exceptions to it can also be accounted for in molecular terms.

An objection to the position just described is that it entails a disturbingoverabundance of causes. The effect has causes at multiple levels, but if thehigher-level causes are not merely superfluous, then the effect must, in some obscureway, result from the combined action of these higher and lower-level causes; but sincetoken-token reductionism claims that the causes at the molecular level are sufficient,this must mean that the effect is over-determined.18 I think that this objection ismistaken. What is being claimed is that the cause (or causes) of any given event canbe described in various ways and that several such descriptions may be correct. Forexample, a detailed molecular account of the prevalence of a resistant strain of HIVand a more abstract explanation given in terms of natural selection represent the sameprocess in different ways, and both might be accurate. Which account is preferablemay depend on what goals are pressing in the context in question. To return toKitcher’s map analogy, there are many different ways to map the earth, correspondingto the different uses to which a map may be put, but that does not entail that there aremany different earths, one for each map. Thus, it is not being asserted or implied that,for each level of description, there is a separate realm of ontologically distinct causes.19

In sum, the principle of multiple perspectives, non-completeness, and autonomyof levels can be interpreted in such a way as to be consistent with token-tokenreductionism. Moreover, the interpretation of these three principles was not a strainedone, but was motivated by standard pluralist themes, such as the multiple realizabilityargument and the “many maps” analogy.20 Hence, token-token reductionism cancomfortably and reasonably coexist with pluralism. In addition, I argued in theforegoing section that there is a basis for claiming that token-token reductionismdeserves the name. Consequently, I claim that an affirmative answer to the questionposed in the title of this essay is justified.

Conclusion

Let us return to the HIV example presented in section 2. Both reductionists andpluralists would see this example as illustrating some of their main contentions.Pluralists would point to the complex interaction of distinct levels of description inthe representation of HIV replication; reductionists would focus on the special roleplayed by the molecular level in identifying and explaining exceptions to the usualgeneralization. The main point of this essay is that both the pluralist and reductionistinterpretations of the HIV example, and other similar examples, may be right.

I have argued that reductionism need not be an anti-pluralistic cannibal that wouldhave molecular biology swallow up all other areas of biology. More specifically, Ihave proposed that reductionism involves the belief in a level of description that ismore fundamental that others, where “fundamental” is understood in terms of correctiveasymmetry. Resources drawn from the fundamental level are necessary to correct

70

higher-level generalizations, but not vice versa. If token-token reductionism is true,then the molecular level would qualify as the fundamental one in the context ofbiology. Thus, the reductionist anti-consensus, which I interpret as consisting of theconjunction of corrective, token-token, and mitigated unifying reductionism, issufficiently robust to be counted as reductionism despite being weaker than hegemonicreductionism. Nevertheless, this version of reductionism is sufficiently modest to beconsistent with pluralism.

Acknowledgments

I would like to thank Megan Delehanty, Elliott Sober, Paul Griffiths, Kim Sterelny,and an anonymous referee for helpful comments on earlier drafts of this essay.

Notes

1 This insight was expressed by William Wimsatt (1976).2 This delightful turn of phrase is borrowed from Sterelny and Griffiths (1999: 149).3 That is to say, T-helper cells in lymphoid tissue in contrast to those circulating in peripheral blood.Circulating T-helper cells express both the CCR5 and CXCR4 co-receptor and thus can be infected byboth M-tropic and T-tropic HIV (cf. Stine 2000: 141). However, the vast majority of lymphocytes arenon-circulating (ibid. 129).4 Nagel (1979: 99), Wimsatt (1979: 352), and Rosenberg (2001: 157) each cite correction as a goal ofreduction.5 My use of the term “theory” is intended to presuppose no specific analysis of what theories are (cf. Suppe1974).6 See Jerry Fodor (1975: 10–12), Kincaid (1990: 576), and John Dupré (1993: 88) for succinct presentationsof the layer-cake model of reduction.7 For example, see Rosenberg (2001: 136).8 See Schaffner (1993a: 328) for an acknowledgment of this point. Sarkar also emphasizes that syntheticidentities were never a requirement in Nagel’s model of reduction (cf. 1998: 25).9 Schaffner also rejects requirement 3 as a desideratum of reductions (cf. 1993a: 340).10 Not even Oppenheim and Putnam claimed that the layer-cake model represented the only possible typeof reduction (cf. 1958: 8). Rather, they advanced the layer-cake model as an account of reduction capableof explicating the intuitive notion of the unity of science.11 Sarkar rejects synthetic identities as a necessary condition for biological reductions on these grounds(1998: 36, 62).12 The notion of a middle-range theory stems from Robert Merton (1968).13 See (Weber and Van Dyck 2002) for a defense of a pluralist approach to explanation of this sort.14 Kitcher includes two additional claims on this list: (4) states a further aspect of the principle of multipleperspectives (that the representations accepted by science at any given time may be inconsistent), while(3) asserts that any such inconsistencies are the result of the imperfections of the current state of science(2002: 570). There is a minor controversy between Kitcher and Longino regarding (3) (cf. Longino 2002a:184; 2002b: 575–576; Kitcher 2002: 571), but that disagreement is immaterial to our concerns here.15 Kitcher also endorses the autonomy of levels (cf. 1994: 379).16 Some statements of Alex Rosenberg invite the attribution of such a viewpoint (cf. 2001: 161; 1997:467). However, I think that Rosenberg’s claim is not that higher-level explanations are superfluous, but

71

rather that causal-mechanical explanations of biological events are incomplete unless molecular detail isprovided (cf. 2001: 160).17 See Kim (1999) for a good discussion of this issue.18 For an exposition of this style of argument, see Kim (1998).19 Downward causation could be also understood from this perspective. Just as a causal process might beaccurately represented at higher and lower levels of description, it might also be accurately represented bya description that mixed one or more of these levels.20 There are several versions of pluralism just as there are several versions of reductionism. For example,Cartwright’s pluralism is more overtly ontological than Kitcher’s, and she makes a point of criticizingtoken-token reductionism (cf. 1999: 32–33). However, the existence of some versions of pluralism thatare incompatible with reductionism of any sort would not undermine the conclusion that I wish to drawhere, namely, that one can be a pluralist and a reductionist too. For a critique of Cartwright’s pluralismfrom a perspective that is sympathetic to Kitcher’s position, see (Ruphy 2003).

References

Bechtel W. and Mundale J.: 1999, ‘Multiple Realizability Revisited: Linking Cognitive and Neural States’,Philosophy of Science 66, 175–207.

Cartwright N.: 1999, The Dappled World: A Study of the Boundaries of Science, Cambridge UniversityPress, Cambridge.

Deng H.: 1996, ‘Identification of a Major Co-receptor for Primary Isolates of HIV-1’, Nature 381,661–666.

Dragic T.: 1996, ‘HIV-1 Entry into CD4+ Cells is Mediated by the Chemokine Receptor CC-CKCCR5’,Nature 381, 667–673.

Dupré J.: 1993, The Disorder of Things: Metaphysical Foundations of the Disunity of Science, HarvardUniversity Press, Cambridge, MA.

Fodor J.: 1975, The Language of Thought, Harvard University Press, Cambridge, MA.Glushakova S.: 1998, ‘Evidence for the HIV-1 Phenotype Switch as a Causal Factor in Acquired

Immunodeficiency’, Nature Medicine 4, 346–349.Hull D.: 1972, ‘Reduction in Genetics – Biology or Philosophy?’, Philosophy of Science 38, 491–499.Hull D.: 1974, Philosophy of Biological Science, Prentice-Hall, Englewood Cliffs, NJ.Kalichman S.C.: 1998, Understanding AIDS, 2nd. ed., American Psychological Association, Washington,

DC.Kim J.: 1998, Mind in a Physical World: An Essay on the Mind-body Problem and Mental Causation,

MIT Press, Cambridge, MA.Kim J.: 1999, ‘Making Sense of Emergence’, Philosophical Studies 95, 3–36.Kincaid H.: 1990, ‘Molecular Biology and the Unity of Science’, Philosophy of Science 57, 575–593.Kincaid H.: 1997, Individualism and the Unity of Science: Essays on Reduction, Explanation, and the

Special Sciences, Rowman & Littlefield, Lanham, MD.Kitcher P.: 1994, ‘1953 and All That: A Tale of Two Sciences’, in E. Sober (ed.), Conceptual Issues in

Evolutionary Biology, 2nd. ed., MIT Press, Cambridge, MA, pp. 380–399. Originally published in 1984,Philosophical Review 93, 335–373.

Kitcher P.: 1999, ‘The Hegemony of Molecular Biology’, Biology and Philosophy 14, 196–210.Kitcher P.: 2001, Science, Truth, and Democracy, Oxford University Press, Oxford.Kitcher P.: 2002, ‘Reply to Helen Longino’, Philosophy of Science 69, 569–572.Liu R.: 1996, ‘Homozygous Defect in HIV-1 Coreceptor Accounts for Resistance of Some

Multiply-Exposed Individuals to HIV-1 Infection’, Cell 86, 367–377.Longino H.: 2000, ‘Toward an Epistemology for Biological Pluralism’, in Creath R. and Maienschein J.

(eds.), Biology and Epistemology, Cambridge University Press, Cambridge, pp. 261–286.Longino H.: 2002a, The Fate of Knowledge, Princeton University Press, Princeton.Longino H.: 2002b, ‘Reply to Philip Kitcher’, Philosophy of Science 69, 573–577.Maas J.: 2000, ‘Strong Association between Failure of T Cell Homeostasis and the Syncitium-Inducing

Phenotype among HIV-1-Infected Men in the Amsterdam Cohort Study’, AIDS 14, 1155–1161.

72

Machamer P., Darden L. and Craver C.: 2000, ‘Thinking about Mechanisms’, Philosophy of Science 67,1–25.

Merton R.: 1968, Social Theory and Social Structure, The Free Press, New York.Mitchell S.: 2002, ‘Integrative Pluralism’, Biology and Philosophy 17, 55–70.Nagel E.: 1961, The Structure of Science: Problems in the Logic of Scientific Discovery, Routledge and

Kegan Paul, London.Nagel E.: 1979, Teleology Revisited and Other Essays in the Philosophy and History of Science, Columbia

University Press, New York.Oppenheim P. and Putnam H.: 1958, ‘The Unity of Science as a Working Hypothesis’, in Herbert F. (ed.),

Minnesota Studies in the Philosophy of Science, Vol. 2, University of Minnesota Press, Minneapolis,MN, pp. 3–36.

Putnam H.: 1975, Mind, Language and Reality: Philosophical Papers, Vol. 2, Cambridge UniversityPress, New York.

Robinson J.: 1992, ‘Aims and Achievements of the Reductionist Approach in Biochemistry/MolecularBiology/Cell Biology: A Response to Kincaid’, Philosophy of Science 59, 465–470.

Rosenberg A.: 1985, The Structure of Biological Science, Cambridge University Press, New York.Rosenberg A.: 1997, ‘Reductionism Redux: Computing the Embryo’, Biology and Philosophy 68,

445–470.Rosenberg A.: 2001, ‘Reductionism in a Historical Science’, Philosophy of Science 68, 135–163.Ruphy S.: 2003, ‘Is the World Really “Dappled”?: A Response to Cartwright’s Charge against Cross-Wise

Reduction’, Philosophy of Science 70, 57–67.Sarkar S.: 1998, Genetics and Reductionism, Cambridge University Press, Cambridge.Schaffner K.: 1967, ‘Approaches to Reduction’, Philosophy of Science 34, 137–147.Schaffner K.: 1969, ‘The Watson-Crick Model and Reductionism’, British Journal for the Philosophy of

Science 20, 325–348.Schaffner K.: 1993a, ‘Theory Structure, Reduction, and Disciplinary Integration in Biology’, Biology

and Philosophy 8, 319–347.Schaffner K.: 1993b, Discovery and Explanation in Biology and Medicine, University of Chicago Press,

Chicago.Sober E.: 1999, ‘The Multiple Realizability Argument Against Reductionism’, Philosophy of Science

66, 542–564.Sterelny K. and Griffiths P.: 1999, Sex and Death: An Introduction to Philosophy of Biology, University

of Chicago Press, Chicago.Stine G.: 2000, AIDS Update 2000, Prentice Hall, Upper Saddle River, New Jersey.Suppe F.: 1974, The Structure of Scientific Theories, University of Illinois Press, Chicago.Waters K.: 1990, ‘Why the Anti-Reductionist Consensus Won’t Survive: The Case of Classical Mendelian

Genetics’, in Fine A., Forbes M. and Wessels L. (eds.), PSA 1990: Proceedings of the 1990 BiennialMeeting of the Philosophy of Science Association, Vol. 1, Philosophy of Science Association, EastLansing, pp. 125–139.

Weber E. and Van Dyck M.: 2002, ‘Unification and Explanation’, Synthese 131, 145–154.Wimsatt W.: 1976, ‘Reductionism, Levels of Organization, and the Mind-Body Problem’, in Globus G.,

Maxwell G. and Savodnik I. (eds.), Consciousness and the Brain: A Scientific and Philosophical Inquiry,Plenum Press, New York, pp. 199–268.

Wimsatt W.: 1979, ‘Reduction and Reductionism’, in Asquith P.D. and Kyburg H. (eds.), Current Researchin the Philosophy of Science, Philosophy of Science Association, East Lansing, pp. 352–377.

Wimsatt W.: 2000, ‘Emergence as Non-Aggregativity and the Biases of Reductionisms’, Foundations ofScience 5, 269–297.

Zhu T.: 1993, ‘Genotypic and Phenotypic Characterization of HIV-1 in Patients with Primary Infection’,Science 261, 1179–1181.

73