Critical Notice: Cycles of contingency – developmentalsystems and evolution

JAMES GRIESEMER1,*, MATTHEW H. HABER1,GRANT YAMASHITA2 and LISA GANNETT3

1Department of Philosophy and Center for Population Biology, University of California, Davis;2Population Biology Graduate Group and Center for Population Biology, University of California,Davis; 3Department of Philosophy, Saint Mary’s College, Hallifax; *Author for correspondence (e-mail: jrgriesemer@ucdavis.edu; phone: 530-752-1068; fax: 530-752-8964)

Received 25 May 2004; accepted in revised form 13 June 2004

Key words: Behavior modulation, Developmental systems theory, Extended heredity, Epigeneticinheritance, Hoyle criteria

Abstract. The themes, problems and challenges of developmental systems theory as described inCycles of Contingency are discussed. We argue in favor of a robust approach to philosophical andscientific problems of extended heredity and the integration of behavior, development, inheritance,and evolution. Problems with Sterelny’s proposal to evaluate inheritance systems using his ‘Hoylecriteria’ are discussed and critically evaluated. Additional support for a developmental systemsperspective is sought in evolutionary studies of performance and behavior modulation of fitness.

Introduction

Cycles of Contingency is one of the finest anthologies we have ever read – asuperb exemplar of an edited volume done well. The edited-volume genre canserve many di!erent kinds of purposes; this volume aims to articulate a pro-gram of research rather than to merely record and summarize a conference inproceedings. It comes close to the ideal of a multi-authored textbook withoutlosing the identity and individuality of its several contributors. These comprisenot only major proponents and historically significant precursors of develop-mental systems theory (DST), but also fellow-travelers from a variety of fieldsand friendly critics who take gentle and supportive issue with the moredramatic and radical interpretations of DST’s core project.

Had the editors chosen to write a textbook rather than to collect togetherand organize separate contributions, the result would have been quite di!erentand we believe less valuable, especially given the wide availability of Sterelnyand Gri"ths’ (1999) Sex and Death, Oyama’s (2002) Evolution’s Eye, and herreprinted Ontogeny of Information. Cycles captures well the larval but rapidlymetamorphosing state of DST. The current instar looks back to foundinginfluences from ethology and evolution (section I), at core critiques of gene-centered biology and contemporary problems of heredity (section II), and atempirical and theoretical strategies for analyzing the development of mor-phology and behavior (section III). In addition to articulating DST itself, the

Biology and Philosophy (2005) 20:517–544 ! Springer 2005DOI 10.1007/s10539-004-0836-4

volume presents other systematic accounts (section IV) and, most admirably,challenges, responses and criticisms (section V). Many of the contributionsaddress conceptual, theoretical, and empirical issues that have emerged in thewake of three core DST documents: Oyama’s seminal (1985) critique ofnativism and genetic determinism, The Ontogeny of Information: DevelopmentalSystems and Evolution, Gray’s 1992 developmentalist manifesto, ‘Death of theGene: Developmental Systems Strike Back,’ and Gri"ths and Gray’s (1994)philosophical analysis, ‘Developmental Systems and Evolutionary Explana-tion.’

The editors have higher aims than scolding biological traditionalists for their‘crude dichotomous accounts of development’ (2). They focus ‘on the kinds ofunderstandings of biology, development, and evolution that make poor practicelikely and render critiques of that practice less than optimally e!ective’ (6). Thebook fully and engagingly succeeds in its primary aim ‘to provide a forum forexploring DST’s alternative conceptions of development, heredity, and evolu-tion’ through select contributors ‘whose writing promised to extend, illustrate,challenge, parallel, or contrast with issues raised in DST’ (ibid.). The intro-duction to the volume, co-authored by the three editors, gives a succinctoverview of DST’s main theses, summarized in Table 1.1 (2): (1) joint deter-mination of traits by multiple causes, (2) context sensitivity and contingency ofeach cause, (3) inheritance extended ‘beyond the gene,’ (4) development asconstruction rather than representation or mere consequence of genetic trans-mission, (5) control of development distributed among many types of interac-tants, and (6) evolution as construction – change in organism-environmentsystems – rather than exogenous molding of organisms and populations byenvironments. Together, these themes outline an alternative vision of biologicalcausality that is theoretically fruitful and empirically heuristic in its criticisms ofneo-Darwinian forerunners. In at least a piecemeal fashion, these tenets areendorsed by an increasing number of philosophers of biology and have at-tracted the interest of biologists seeking to integrate new work in behavioraland developmental biology with evolution and an expanded theory of heredity.Our local experience at UC Davis indicates that biologists, particularly grad-uate students of behavioral biology, are reading the book with interest.

The five sections of Cycles form an integrated whole, laying out DST as ascientific and philosophical research program. ‘Influences’ describes three his-torically significant projects that constituted critical resources for DST: (1)Daniel Lehrman’s critique of Konrad Lorenz’s theory of instinct (Timothy D.Johnston essay 2, ‘Toward a System’s View of Development: An Appraisal ofLehrman’s Critique of Lorenz’; Daniel S. Lehrman essay 3, ‘A Critique ofKonrad Lorenz’s Theory of Instinctive Behavior’), (2) Gilbert Gottlieb’s de-velopmentalist psychobiology (Gottlieb essay 4, ‘A Developmental Psychobi-ological Systems View: Early Formulation and Current Status’), and (3)Richard Lewontin’s gene-organism-environment dialectic (Lewontin essay 5,‘Gene, Organism and Environment: A New Introduction,’ and essay 6, ‘Gene,Organism and Environment’). All three present sustained attacks on the nat-

518

ure-nurture dichotomy. Their mutually reinforcing perspectives frame thepresentation of DST in later essays.

Contributions to the section, ‘Rethinking Heredity,’ reconceive traditionalnotions of heredity in both developmental and evolutionary biology in wayscongenial to the DST program. Eva Neumann-Held (essay 7, ‘Let’s TalkAbout Genes: The Process Molecular Gene Concept and Its Context’) andLenny Moss (essay 8, ‘Deconstructing the Gene and Reconstructing MolecularDevelopmental Systems’) challenge traditional concepts of heredity by refor-mulating the concept of the gene. As Neumann-Held notes, given that the term‘gene’ is used in di!erent ways in di!erent biological contexts, the biologist orphilosopher of biology has several options: the term can be abandoned alto-gether, the context-dependent status quo can be accepted, or the term can beredefined. Opting for the last of these, Neumann-Held and Moss o!er defini-tions of ‘gene’ geared specifically to developmental biology. Wider concepts ofinheritance suitable to evolutionary biology contexts are defended in chapterswritten by Eva Jablonka (essay 9, ‘The Systems of Inheritance’) and KevinLaland, F. John Odling-Smee, and Marcus Feldman (essay 10, ‘Niche Con-struction, Ecological Inheritance, and Cycles of Contingency in Evolution’).Neo-Darwinian approaches to evolution are challenged in these essays; theorganism versus environment dichotomy as well as Dawkins’ replicator-vehicledistinction and ‘extended phenotype’ concept come in for particular criticism.

In ‘The Development of Phenotypes and Behavior,’ a variety of targets ofexplanation are considered which must, according to the authors, be accom-modated by taking developmental perspectives on evolution. These essayschallenge or extend simplistic developmental mapping relations implied by theclassical genotype/phenotype distinction to include: developmentally plasticgenotype-phenotype relations (H. Frederik Nijhout essay 11, ‘The Ontogeny ofPhenotypes’); aggregate behavior that causes group-level development, exem-plified by studies of the consequences of ant behavior for colony-level prop-erties (Deborah M. Gordon essay 12, ‘The Development of Ant ColonyBehavior’); and maternal provisioning and other developmental performancesthat confound static conceptions of genotype-phenotype relations (PatrickBateson essay 13, ‘Behavioral Development and Darwinian Evolution’; PeterH. Klopfer essay 14, ‘Parental Care and Development’).

The essays of this section further the goals of the book as a whole by pro-viding empirical evidence of the themes of DST in action. As such, they providedata showing that DST treatments of research questions in biology are bothviable and fruitful. By o!ering data that cuts across biological disciplines, theeditors demonstrate that DST approaches are flexible and broad enough to berelevant to a variety of biologists, yet are still capable of generating and for-mulating specific kinds of research problems/questions. These essays showcaseparticular research programs as Kuhnian exemplars of how to practice DSTempirically, and thus reveal how the main DST theses are manifested inpractice. For example, Gordon’s thorough account of the development of redharvester ant (Pogonomyrmex barbatus) colonies provides examples of

519

extended inheritance (the colony), joint determination, context-sensitivity, anddevelopment as construction (worker ant development in newly founded versusolder colonies), distributed control and evolution as construction (ant behavioras a function of the relation of ants to each other and to their colonyenvironment – an environment they actively construct and maintain).

‘Rethinking Development and Evolution’ puts side-by-side for comparisonfour parallel philosophical approaches to development and evolution, each ofwhich rejects, in some way, traditional neo-Darwinian evolution: DST (SusanOyama essay 15, ‘Terms in Tension: What do you do when all the good wordsare taken?’; Paul E. Gri"ths and Russell D. Gray essay 16, ‘Darwinism andDevelopmental Systems’), generative entrenchment (William C. Wimsatt essay17, ‘Generative Entrenchment and the Developmental Systems Approach toEvolutionary Processes’), autopoietic systems (Bruce H. Weber and David J.Depew essay 18, ‘Developmental Systems, Darwinian Evolution, and the Unityof Science’), and an anthropological ‘dwelling’ perspective (Tim Ingold essay19, ‘From Complementarity to Obviation: On Dissolving the Boundaries be-tween Social and Biological Anthropology, Archaeology, and Psychology’).All are critical of neo-Darwinian dichotomies and naıve interactionism. Theessays of this section reveal consilient lines along which many of DST’s fellowtravelers are pursuing developmentalism from distinct conceptual startingpoints. Broadly, ‘developmentalism’ names the general theoretical perspectivewhich interprets all biological phenomena in the light of the process ofdevelopment. Some of these essays also suggest ways to extend the researchprogram to the human sciences. As we argue below, it becomes criticallyimportant to assess the robustness of research findings, across varying per-spectives of the emerging developmentalist consensus, to assumptions thatdrive scientific modeling strategies beyond classical genetics (Griesemer, sub-mitted). The essays of this section provide a comparative basis upon which tobegin this line of work. Oyama considers the reconceptualization of centralterms and concepts. Gri"ths and Gray and Wimsatt interpret natural selectionand adaptation in terms of complementary approaches to extending Darwin-ism: DST focuses on the individuation of developmental systems and genera-tive entrenchment focuses on emergent organizational properties of suchsystems. Weber and Depew link DST with their physico-chemical perspectiveon self-maintaining and self-organizing systems, and Ingold broadens thediscussion in a social-scientific critique of dichotomies in the human sciences.The editors place their own chapters in this theory section, complementing themore empirical thrust of the previous section.

Finally, ‘Responses to Developmental Systems Theory’ considers five reac-tions to the work of the volume and the DST program as a whole, beginningwith its descriptive status and ending with its normative ethical status: (1)DST’s status as a descriptive philosophy of nature and as a scientific researchprogram (Peter Godfrey-Smith essay 20, ‘On the Status and ExplanatoryStructure of Developmental Systems Theory’), (2) defense of boundaries suchas skin and membrane that are denied, obviated or claimed irrelevant by DST

520

(Evelyn Fox Keller essay 21, ‘Beyond the Gene but Beneath the Skin’), (3)illustrations of the ‘politics’ of the distributed control/agency required byDST’s expansive view of causal contribution by many kinds of developmentalresources (Peter Taylor essay 22, ‘Distributed Agency within IntersectingEcological, Social, and Scientific Processes’), (4) categorization of extendedinheritance and assessment of how far DST can go in its denial of traditionaldistinctions among kinds of developmental resources (Kim Sterelny essay 23,‘Niche Construction, Developmental Systems, and the Extended Replicator’;we will have much more to say about Sterelny’s essay in section 3 below.), and(5) assessment of DST’s normative status as science criticism and ethics ofattention (Cor van der Weele essay 24, ‘Developmental Systems Theory andEthics: Di!erent Ways to Be Normative with Regard to Science’).

All of the responses in this section reflect tensions between DST’s critical andproductive programs. On the one hand, DST is committed to a critique ofnativism that denies a variety of distinctions: nature/nurture, classical geno-type/phenotype, organism/environment, and biology/culture. But at the sametime, DST pursues a scientific research program which cannot completely cutties with biological research activities dependent on the distinctions under-mined by DST’s critical program.

In the rest of this essay review, we address the themes, topics, critiques, andanalyses of the volume through consideration of three issues: (1) tensions thatarise in pursuit of a robust understanding of development between the plu-ralism of DST’s critique of neo-Darwinism and the monism of its positiveresearch program, (2) conflict among competing perspectives on extendedinheritance, and (3) the role of behavior or more broadly, performance, inevolutionary theory.

Robustness, explanatory pluralism and ontological monism

The promise of DST lies in the prospect of a scientific research program aswell as in its guiding ontology. A major challenge to configuring DST forscientific work will be to balance explanatory pluralism for many kinds ofdevelopmental resources, resulting from DST’s critique of gene-centrism,against the ontological monism implied by DST’s representation of every sortof causally relevant interaction as developmental. This ontological viewamounts to a sort of reductionism for interactions and processes rather thanstates and things. Clearly, from a developmental point of view, genes are noteverything. Should one be as sanguine about DST’s equally strong conclusionthat society and culture are nothing, at least in the special sense that there isno cultural evolution to be distinguished from the biological variety (Gri"thsand Gray 1994: 301; cf. Griesemer 2000a)? The ground for this conclusion isthe argument that there cannot be a separate channel of socio-culturalinheritance, but only an evolution-structured series of developmentalinteractions among ‘social,’ ‘cultural’ and ‘organism’ resources. Therefore,

521

there cannot be a process of socio-cultural evolution separate from the bio-logical evolution in which organisms participate. There are other argumentsby proponents of cultural inheritance against the likelihood of cultural evo-lution (Boyd and Richerson 1996), so the distinguishing mark of DST is notopposition to cultural evolution, but rather DST’s formal relation to theoriesof these supra-organism resources.

In one sense DST’s non-separability thesis is surely right. Unless it turnsout that machines lacking biology can have culture, then culture is inextri-cably tied causally to organisms. It does not follow, however, that the bestexplanations of culture will be biological – they could turn out to be socio-logical or cultural, just as the best explanations of biological phenomena areoften biological, rather than chemical or physical. But there is also a deeplyreductionistic sense in which the ontological monism implied by DST getsthings wrong. Social and cultural organization is emergent even if it is ofdevelopmental systems, although to date the DST ontology is rather too flatto describe this fact because it projects every resource and interaction into thedevelopmental plane. Explanatory success resulting from treating social andcultural phenomena as developmental resources in this way does not therebyconstitute evidence or support for the ontology. In general, althoughexplanatory pluralism and ontological monism are compatible (Sterelny1996), neither justifies the other. So even if every population with culture iscomposed of developmental systems, cultural states, cultural transmission,and cultural change may not be best or even well explained from a devel-opmental point of view.

In joining the DST bandwagon, it is tempting to think that developmentalinteractions are the sole units of (biological) process and that this develop-mental-ism o!ers a complete theoretical perspective to replace the gene-cent-rism that went before. However, part of the flaw in gene-centrism was preciselythe notion that it is a complete theoretical perspective. The lack of theoreticalattention in DST to the role of social organization and the establishment ofsocial and cultural institutions in structuring contingent developmental inter-actions is either a worrying lacuna in the ontology or rather too far down theresearch agenda for comfort, given DST’s disavowal of the reductionist ten-dencies of gene-centrism. Where is the ‘downward’ causal contribution of so-cio-culture to development that we have come to expect generally fromcomplex, emergent phenomena? Merely placing culture and society into adevelopmental context as sources of supra-organism resources runs reduc-tionist or interactionist risks equal to those of gene-centrism and its socio-biological excesses. Despite its ‘systemsy’ notions of interactions andorganisms, DST is a very organism-centered theory. Ironically, an overlystrong endorsement of developmentalism as the antidote to the fetishizedgenetics that gave us naıve human sociobiology could lead down the samepath, only with developmental legacies replacing genetic ones in a newreduction of the social sciences.

522

Because DST places organisms at the center of its developmentalist theory asunits of coordinated resource intersections,1 the social becomes a mere sur-round for development or even a part of it. At the end of the day, if DST weremerely to replace sociobiology’s gene-centrism with an organism-centered de-velopmentalism that makes no more room for genuinely social and culturalphenomena, then its ‘systems’ would look very much like classical organismswith a dash more environment than Dawkins’ extended phenotypes had. DSTmust halt the slide of the social into mere interaction among developmentalresources, else it cannot capture social phenomena at emergent levels ofdevelopmental organization. Developmental systems may be networks ofinteractions, but what one wants to know is how they are organized. Some ofthat organization is undoubtedly emergent and unit-like in its own right, socalling culture just part of the ‘developmental system’ seems to throw dirtybath-water into the tub along with the baby.

Development, like genes, is not everything. Creating an ontological unity ofpan-developmentalism out of an explanatory pluralism by projection into asingle conceptual dimension would be a mistake similar to the one gene-centrism made. If biological phenomena admit of descriptive and interactionalcomplexity (Wimsatt 1974), they cannot be adequately represented from anysingle perspective. Repeating gene-centrism’s mistake would obliterate the verycomplexities and textures that incline us toward DST’s critique of gene-cent-rism and neo-Darwinism in the first place.

A robust theory of developmental systems cannot be exclusively develop-mental. Gri"ths and Gray’s (1994) criteria for the individuation of develop-mental processes are evolutionary: resources count as parts of developmentalprocesses only if they have evolutionary explanations. So it turns out that DSTis not actually ‘Full Hoyle’ developmentalism, to co-opt Sterelny’s phrase (seebelow). If theoretical resources from evolution are required to individuatedevelopmental systems, then why insist on ontological monism when it comesto heredity or culture?

A robust theory will surely be one which transcends the false assumptions ofall its idealizing models, including those of its developmental models, and ifDST already requires evolutionary as well as developmental models, why notstill more kinds of models? Why, indeed, do with less when more are available?In these early stages of (re-)theorizing development, we see little point in anysort of ontological monism in the hope that a unified theory will follow.Reductionism works better as a tool of rational reconstruction than of theory-making, so treating developmentalism as a replacement for gene-centrism putsthe ontological cart before the explanatory horse: the philosophy may smell

1Look, for example, at the density of arrows representing developmental interactions in Gri"thsand Gray 1994, (Figure 1, 285) for subtle evidence that organisms are more central than ‘‘higher’’levels of resources.

523

better, but it cannot get to market.2 Robustness is the working principle ofparsimony for model-based sciences like biology. These seek theories in col-lections of models and their robust consequences that o!er ‘piecewiseapproximations to reality’ (Wimsatt in preparation), not in general principlesfor a unified ontology (Levins 1968; Wimsatt 1976).

We favor theoretical perspectives that facilitate cooperation across empiricalresearch programs and lines of work. Science proceeds more by cooperationthan consensus and ‘our truth,’ Levins wrote, ‘is the intersection of indepen-dent lies’ (Levins 1966). Ontological monism of any kind cannot press hardenough for disciplined testing of the independence of empirical results fromfalse assumptions that were made for strategic reasons. The results of studiesfrom di!erent perspectives must be jointly articulated in order to address theirindependence from idealizing assumptions made by particular models orclasses of models (Griesemer, submitted). Robustness analysis at the level ofwhole families of models requires shifts of perspective across ontologies, soontological monism is likely to be hostile to theoretical pursuits involvingexplanatory pluralist strategies.

Thus, there is no point arguing on programmatic grounds that DNA cannotfunction as an inheritance ‘channel’ or that niche construction fails as a fun-damental representation of the evolutionary process even if such assumptions,according to Gri"ths and Gray (essay 16: 196, 206), turn out to be useful forproscribed tactical purposes. Modeling tactics of many varieties should begrounded in a general research strategy of robustness analysis to take advan-tage of strengths and weaknesses of each perspective, not to force a choiceamong them.

Extended inheritance

DST argues for an ‘extended’ view of heredity, one which acknowledges thatmany more resources than DNA are involved, including molecular, organismal,group, and ecological mechanisms. DNA does not replicate, save for the actionof enzymes, and DNA sequences do not code for amino acid sequences withoutthe mediation of inherited aminoacyl-tRNA-synthetase enzymes (Griesemer, inpreparation). Neither DNA nor enzymes would be inherited but for the trans-generational continuity of membranes, metabolism, and a significant portion ofthe ‘external’ environment. Thus, even if inherited enzymes are coded for byDNA, DNA only codes in virtue of pre-existing inherited enzymes. The sub-systems are locked together chemically, physiologically, and ecologically, soDST has a point in focusing attention on the ‘systemsy’ nature of heredity.Moreover, the chicken-and-egg problem of how such an interlocking systemcould evolve presents a causal and explanatory regress back to the origin of

2See Griesemer (2002) for an argument showing how to use reductionism as a heuristic tool informulating a theory of development compatible with genetic theory.

524

genetic systems, which neither developmentalist nor geneticist can win onpresent evidence. It would be a Pyrrhic victory, since few would be satisfied by aregress-appeal to the origin of life to explain why developmentalism or gene-centrism should be the better theoretical perspective to be going on with.

Additionally, gene expression is mediated by DNA methylation, histoneacetylation and other epigenetic mechanisms, which some argue constitute notonly inheritance systems in their own right (Jablonka and Lamb 1995, Jab-lonka essay 9) but perhaps even coding systems (Sharon and Lis 1989; Turner2000; Jenuwein and Allis 2001). Sexual organisms cannot reproduce bythemselves, so inheritance is necessarily a social and cultural process for them.Interdemic group selection involves group reproduction processes that generateheritabilities at the group level (Wade and Griesemer 1998; Griesemer andWade 2000). Many properties and parts of ‘environments’ must persist, recur,or reproduce in order for organisms to reproduce and DNA to replicate, whichsome call ecological inheritance (Laland et al. essay 10; Odling-Smee et al.2003). Thus, DST asserts that a plurality of kinds of things play a ‘hereditaryrole,’ as do many others endorsing extension of concepts of inheritance ‘be-yond the gene’ (see also Sterelny et al. 1996 and especially Wimsatt essay 17).

Moreover, DST sometimes presses its case for plurality in terms of symmetryor parity arguments (Gri"ths and Knight 1998; Gri"ths 2001; Gri"ths andGray essay 16). These address the particular qualities of nucleic acid polymersthought to privilege genes in heredity explanations, particularly their capacityto carry ‘genetic information,’ which other developmental resources areclaimed to lack (Maynard Smith 2000). But, argues Gri"ths, on any articulateaccount of information, whether syntactic (e.g. communication theoretic orDretskean) or semantic (e.g. teleo-semantic), a case can be made that if genescan carry information, so can other developmental resources. One need onlyperform the thought experiment of holding genes fixed to see that if geneticexplanation of heredity is a matter of di!erence-makers explaining phenotypicdi!erences, then holding genes fixed and allowing other resources to vary putsthe explanatory spotlight on those other resources, just as holding environmentsfixed and allowing genes to vary focuses attention on genes. The situations aresymmetrical, so genes are not intrinsically privileged, but rather pragmaticallyso (Gannett 1999). Explanatory and practical interests rather than intrinsicbiological properties lead to stories of genetic privilege.

On the other hand, DST argues against a particular class of views of extendedheredity that sort kinds of developmental resources which play hereditary rolesinto ‘channels’ of inheritance, including gene/culture dual-inheritance theories(e.g. Boyd and Richerson 1985), gene/environment dual-inheritance theories(Laland et al. essay 10, see also Odling-Smee et al. 2003), and multi-channel ormulti-level inheritance theories, including extended replicator theory (Sterelnyet al. 1996, Sterelny essay 23) and Jablonka’s fourfold theory of genetic, epi-genetic, behavioral, and symbolic inheritance systems (Avital and Jablonka2000; Jablonka 2002, Jablonka essay 9). This is particularly clear in Gri"thsand Gray’s essay, where they argue in response to the extended replicator theory

525

of Sterelny et al. (1996), that ‘it is both more biologically realistic and, in thelong run, more productive to think of the life cycle being reconstructed by asystem of resources’ than to suppose there are multiple channels of inheritance(196). They allow that the multi-channel assumption can be ‘tactically’ useful in‘modeling exercises,’ but their rhetoric clearly demotes that work in favor of themore biologically realistic and productive work DST promises. Indeed, Gri"thsand Gray even apply this reasoning to their niche-constructionist allies (206),suggesting that it is only for model tractability that one should interpretdevelopmental systems as ‘two things’ (organism + environment) rather than‘one thing’ (developmental system). Thus, they are at odds with multi-channeltheorists of many flavors over how best to shift attention away from gene-centrism and how to go on from there.

Gri"ths and Gray pose three challenges to multi-channelers. First, theyargue that some things that are clearly developmental resources are neverthe-less ‘not easily represented as ‘channels’ or ‘replicators’ ’ (197). They claim it ishard to imagine eucalypt germination is a character transmitted by the bushfire‘channel’ (ibid.), though we are not sure why, since Gray applauded Bateson’sretort to Dawkins that a bird is a nest’s way of making another nest (Gray1992). Nor do we see what’s so hard about thinking that gravity, sunlight orother ‘standing features’ are ‘transmitted’ by organism channels (ibid.), sinceDST teaches that it is the interactions of these features with organisms and notthe features themselves which are parts of developmental processes. After all,even gene transmission is di!erent than the flow of water through a pipe fromsource to receiver: the o!spring develops, it is not an empty receiver filled up bygenes, so perhaps the ‘transmission’ trope is taken too literally by defendersand detractors alike.

Second, they also argue that heredity holism of the sort DST endorses ismore heuristically valuable than multi-channelism (197) on grounds that itkeeps the context-dependencies always in view, even while tactically dividingheredity into separate component systems for modeling purposes or whileholding most factors fixed to identify the contribution of a single factor. Whilethis does seem a genuine heuristic advance over gene-centrism, it does not seemto us that multi-channelers have trouble remembering to track the context-dependencies, especially since it is generally explicit in multi-channel modelsand experiments that each channel is part of the context for the others, so thedependencies must be accounted for explicitly as well.

Third, and perhaps most importantly for heredity holists, the channels of themulti-channelers must be separate for the transmission of developmentalinformation (196). A core concern of DST is the individuation of develop-mental units and the separability of channels bears directly on this problem.The parity argument showed that developmental resources playing hereditaryroles have explanatory parity or equality when it comes to developmentalprocesses conceived as life cycles of developmental interaction sequences. Somulti-channelism sounds like a doctrine of separate but equal cycles. But DSTsays that is impossible, because in order to be part of a system a resource must

526

interact with others, hence it cannot constitute a separate channel. A resourcecan either be separate from a developmental system or an explanatory equalwithin a system, but not both. Thus, multi-channelism does not support ‘causaldemocracy’ of the sort Susan Oyama favors over bland interactionism, ofwhich she accuses DST critic Philip Kitcher in her essay (182).

We think DST raises a legitimate concern, but that the argument does notacknowledge an important distinction. Channels may be physically, chemicallyor biologically separate and parallel without being causally or statisticallyindependent. In general, two things in interaction are two things and so areseparate in some sense, on pain of violating the assumption that they are two.Separateness and (statistical) independence are two distinct properties. Cor-relation does not imply causation, but the reverse can also be true. Whether thestate of a receiver is correlated with the state of a source in the relevant respectsdepends on more than just whether the two are in physical interaction. As wehave learned from the debate over genetic information, whether (and what)information may be ‘carried’ by genes depends on the mechanisms of readinggenes as much as on the combinatorics of writing them. It takes sources andreceivers physically coupled in the right way for there to be a particular sys-tematic dependence of the states of the latter on the former. There is no nec-essary relation between separateness per se and statistical independence in aparticular respect. Rather, the connection between these relational properties ismodal, but in a di!erent, ceteris paribus way: given certain capacities of cou-pled systems and the right sorts of enabling conditions, the subsystems we call‘receivers’ will display statistical dependencies on those we call ‘sources.’

The conditions required for statistical independence are the ones attacked inthe parity argument against privileging genes on grounds that they are the (soleor main) carriers of genetic information. But clearly, entities or processes thatinteract can be separate and persistent. The nucleotides that constitute a pieceof DNA are spatially separate, distinguishable molecules, even though they arechemically bonded to their neighbors and su!er shared (or at least correlated)fates because of this and other chemical facts. To identify persistent entities orprocesses as information channels is to identify a specific modal property ofthem – a capacity to function in a certain way under certain conditions. Theyneed not be statistically independent at all times or in all circumstances in orderto have the capacity to be so. And they need not carry information intrinsicallyor ‘ceteris absentibus’ (all other channels being absent) in order for the infor-mation to be carried in virtue of separate but interacting channels. Di!erentlyput, genetic information should be thought of as a dynamical property of thedevelopmental system while the channels of information flow are organiza-tional structures. We agree with DST that genetic information is carried in adistributed, even holistic way. But this is compatible with multi-channelism,which is a claim about the mechanical structure and distribution of develop-mental resources which collectively carry the information, not about explan-atory privilege with respect to system dynamics. Thus, we do not think thatDST’s successful parity argument against gene-centrism also weighs against

527

multi-channelism. By the same token, we will argue below that some challengesagainst heredity holism are also unsuccessful.

Critics of DST, even friendly ones (Sterelny et al. 1996; Sterelny essay 23),argue that the plurality of non-DNA developmental resources does not by itselfrequire equal representation in inheritance explanations because the geneticinheritance system is closer to ‘full Hoyle’ than are most other transmitted orpersistent developmental resources.3 Proponents Gri"ths and Gray (1997;essay 16) reject the full Hoyle argument on grounds that it only partiallyspecifies the targets of evolutionary explanation. In light of the above con-siderations, we want to split o! the controversy over how many channels ofinheritance there are from the dynamical issues concerning the propertiesnecessary or su"cient for inheritance systems to sustain an evolvable world. Inthe next section, we explore this latter problem further.

Full Hoyle specifications for an evolvable world

Kim Sterelny (essay 23, ‘Niche Construction, Developmental Systems, and theExtended Replicator‘’) describes a set of su"ciency conditions for an ‘evolv-able world,’ that is, for a world in which complex adaptations can be producedby cumulative natural selection. He argues that to the extent these conditionsare fully met by an inheritance system, it is the sort of system of interest to‘Hoyle & Co.,’ whose business is ‘supplying an empty planet with an indige-nous biota’ (338). In short, such a system is ‘full Hoyle.’ The argument goessomething like this: our world is evolvable and we have (at least) a geneticinheritance system that sustains it. There are three sorts of properties thissystem has in virtue of which it sustains an evolvable world, so these together inthe relevant context must be su"cient for it. But there are other (candidate)inheritance systems which may also satisfy at least some of these properties, sothey may sustain an evolvable world to some degree as well. Consider a list ofproperties of the three sorts and deem those inheritance systems ‘full Hoyle’which meet them all. Other systems can be judged evolvability-sustaining byhow close to full Hoyle they are.

Sterelny’s three general conditions for an inheritance system that will sustainan evolvable world are as follows: (1) outlaw replicators (cheaters, defectors)are blocked from reaping any fitness advantage due to sharing in benefitsa!orded by the systems in which they participate, but to which they contributeless toward the common good than do other, more cooperative components. In

3It is not clear in these essays what an inheritance system is. Is it a certain kind of developmentalsystem, a part of a developmental system, or something else? Griesemer (2000a, b, c) argues that aninheritance system is a special kind of evolved reproducing system and that reproducing systemsinvolve development in a way quite di!erent than that represented by genotype/phenotype maps.On the reproducer view, inheritance is a system property rather than a property of a subset ofsystem resources. But the mechanisms of inheritance are not ‘di!use’ either, so the reproduceraccount stands between multi-channelism and DST.

528

general, contribution is costly and free-riding is cheap, so variance in contri-bution to the common good relative to selfish reproductive interests sets up thecondition in which outlaw behavior can evolve and anti-outlaw conditions tendto prevent the origin, moderate the e!ects, or eliminate such variance. Since itseems that cooperation is necessary for complex living systems,4 outlaws wouldundermine them unless strong among-systems selection could overpowerwithin-system selection favoring cheating. (2) Phenotypes are stably trans-mitted over many generations, since there cannot be complex adaptationwithout accumulation of beneficial variants and accumulation requires stabletransmission of those variants generation after generation. (3) The systemgenerates a large range of possible phenotypes so selection has continuing widescope to explore the space of adaptive variations that can be accumulated.

Together, these three conditions are collectively something like su"cient foran inheritance system to sustain an evolvable world, though it is not clearwhether each is strictly necessary.5 For each of the three, Sterelny mentionsvarious and possibly alternative subconditions. As we will discuss, it is notclear whether or how the subconditions may or must combine to render eachmain condition part of a su"cient condition for evolvability. Sterelny suggeststhat, overall, evolvability is sustained by a degree property of inheritancesystems which we will call ‘Hoyleyness.’ We argue that satisfaction of thisproperty to a given degree by various combinations of the subconditions canonly be judged on the basis of a clear understanding of relationships among thesubconditions, since the judgment of Hoyleyness rests on a count of how manysubconditions are satisfied in each case as well as on judgments of theirapplicability. We think these relationships are not yet clear enough to supportthe judgments Sterelny makes, even though we agree with his judgment in anumber of the cases he presents.

More specifically, Sterelny identifies nine properties that fall under his threegeneral specifications. The anti-outlaw conditions that help insure cooperationby imposing a common evolutionary fate on components include: vertical (C1),simultaneous (C2), unbiased (C3) transmission of system components, with a‘ballistic’ or contingently irreversible commitment to participation in the system(C4) (339). The stability conditions that provide the basis for accumulation ofadaptive variants include: high copy-fidelity (C5) and a robust mapping fromreplicators to organizations (or as we used to say, from genotypes to pheno-types) due to built in redundancies and predictable contexts of expression (C6)(339–340). Finally, conditions on the generation of variation include: a verylarge array of possible variants, where ‘very large’ means much larger than canbe simultaneously realized in any actual population (C7), i.e. ‘unlimitedheredity’ (Szathmary 1995; see also Jablonka, essay 9: Table 9.1); a ‘smooth’

4Otherwise, there would be at most an aggregation of components rather than an organized whole.Wimsatt (1985, 1997) develops the notion of emergence in terms of an analysis of aggregativity.5Forber (in preparation) argues that there are important asymmetries between these three condi-tions in their significance for the role inheritance systems can play in evolution.

529

replicator/organization map such that small changes in replicators result insmall changes in organization (C8). If we suppose in addition a smooth map-ping between phenotype and fitness, then C8 can be related to Lewontin’scontinuity criterion for the genotype/phenotype map: ‘small changes in acharacteristic must result in only small changes in ecological relations’(Lewontin 1978: 230); and lastly organization should be modular (C9).

There are many di!erent notions of modularity (e.g. Wagner and Altenberg1996; Brandon 1999; Bolker 2000; Ra! and Ra! 2000; Gilbert and Bolker2001; Winther 2001). Sterelny’s is roughly that organization is modular if it isnot ‘holistic,’ i.e. that ‘the replicators as a whole should not generate thebiological organization of the organism as a whole; rather, replicators, or smallsets of replicators, should be designed so that they make a distinctive contri-bution to the generation of one or a few traits, and relatively little distinctivecontribution to others’ (Sterelny, essay 23: 340; see also Jablonka, essay 9:Table 9.1). If traits are described with respect to their contributions to fitness,then Sterelny’s characterization can be compared to Lewontin’s notion of‘quasi-independence’: ‘there is a great variety of alternative paths by which agiven characteristic may change, so that some of them will allow selection toact on the characteristic without altering other characteristics of the organismin a countervailing fashion; pleiotropy and allometric relations must bechangeable’ (Lewontin 1978: 230). Sterelny’s modularity concept is related toWagner and Altenberg’s (1996) and Brandon’s (1999) in defining modularity interms of genotype/phenotype mapping and can be related to Wimsatt’s conceptof emergence as failures of aggregativity as well (Wimsatt 1985, 1997).

Sterelny brings his entertaining and insightful description of ideally evolv-able inheritance systems to bear on DST’s notion of extended, and especially,ecological inheritance. According to DST’s version of extended inheritance,whole developmental systems are the units of inheritance – the true replicators– not genes (Gri"ths and Gray 1994). Moreover, if other extended notions ofinheritance involving multiple channels fail against the parity argument(Gri"ths and Knight 1998; Gri"ths 2001; Gri"ths and Gray essay 16), thereseem to be no other candidates besides DST’s heredity holism. Sterelny, bycontrast, favors a multi-level extension of the replicator concept (Sterelny et al.1996), but expresses skepticism in his Cycles essay that all levels of inheritanceare as ‘full Hoyle’ as the genetic inheritance system. If correct, it would followthat not all levels of organization are equally evolvable and thus that there maybe some sense to the explanatory privilege of genes rejected by DST. Sterelnyconsiders in particular whether ‘environmental engineering should count asenvironmental inheritance,’ i.e. whether organism-caused modifications toenvironments that persist and feed back to a!ect organism fitnesses function asunits of inheritance.

Sterelny argues that whether an extension does count as inheritance dependson how close to ‘full Hoyle’ a particular engineering process is. Most are notvery close according to Sterelny’s tally of the properties on his Hoyle list, al-though a few (vertically transmitted endosymbionts) are almost as close to full

530

Hoyle as the genetic inheritance system. Thus, it is a matter of degree how‘Hoyley’ an inheritance system is.Moreover, if the distribution of systems on theHoyle scale matches Sterelny’s sampling of exemplars, the genetic inheritancesystem should enjoy explanatory privilege to some extent, contrary to DST. Onecannot just claim ecological inheritance on grounds of ecological engineering ina given context, since not all engineered environments (or interacting symbionts)satisfy the Hoyle conditions to the same degree. Sterelny’s argument is a usefulantidote to excessive speculation, but it depends heavily on being able to tote upsatisfaction of properties on the list to produce a Hoyle score.

In this section, we probe Sterelny’s arguments further. Although his essayfocuses philosophical attention on many properties of current interest toevolutionary theorists, we identify several problems that limit the utility ofSterelny’s Hoyle list for drawing conclusions about extended heredity.

First, it is not obvious that the properties on Sterelny’s list hang together forall inheritance systems. Sometimes, the component properties might trade o!:satisfaction of one may help meet one of the general Hoyle conditions but atthe same time conflict with another subcondition’s satisfaction of the samecondition. If this is widespread now or was probable during evolutionarytransitions that generated levels of organization, it would confound Hoylescore as a measure of evolvability even if Sterelny is right that the three generalconditions he describes are su"cient for evolvability. This may indeed be thecase for the anti-outlaw conditions.

Alexey Kondrashov (1994) explored the evolutionary costs of several pos-sible modes of obligate vegetative reproduction, in which more than oneparental cell contributes to the formation of each o!spring (Bonner 1974;Dawkins 1982; Grosberg and Strathmann 1998). This paper is devoted toexplaining why even multi-cellular organisms generally pass through a one-cellbottleneck at some point in their life cycle. Asexual reproduction is a specialcase in which the number of founding cells equals one, i.e. a true bottleneck.Vegetative modes are near to or far from asexual reproduction depending onthe mechanism of the former. Sexual reproduction also involves an n=1bottleneck in the zygote that follows gamete fusion. A one-cell bottleneckeliminates within-organism among-cell variation (e.g. due to somatic mutationof cells during the life of the organism), suppressing any outlaw mutants arisingduring the life of the parent that would otherwise get into o!spring byout-competing non-mutant cells for the opportunity.

In obligate vegetative reproduction more than one cell forms the o!spring, sothere is scope for outlaws.6 Grosberg and Strathmann (1998) point out thatbecause it is very risky to go through a one-cell stage (e.g. due to exposure topredators), the benefits of blocking outlaws at the unicellular stage must belarge. Kondrashov’s question is whether there is su"cient scope for outlaws thatvegetative reproduction is generally at an evolutionary disadvantage compared

6‘Obligate vegetative reproduction, although relatively rare, is known in bryophytes, ferns, flow-ering plants, fungi, lichens and animals’ (Kondrashov 1994: 311).

531

to reproduction with a one-cell bottleneck. Our question is whether the inheri-tance systems of vegetative reproducers turn out not be as full Hoyle as those ofreproducers who regularly pass through a one-cell bottleneck. If the Hoylecriteria are a good way to assess capacity for evolvability, this question shouldhave a clear answer. We think it does not, so we do not think the Hoyle criteriaare good grounds (yet) for judging the evolvability of an inheritance system.

In the next few paragraphs, we describe the modes of vegetative reproduc-tion that Kondrashov analyzed in his model in order to address our questionrather than his. Kondrashov’s simulations show that the vegetative modeswhich diverge the farthest from true asexual reproduction have the highestfitness costs, so the ‘more vegetative’ forms of reproduction are not likely to getestablished or be maintained. Reproducers in the mode closest to a true bot-tleneck are not at much of a disadvantage, but then they can be expected tosuppress outlaws nearly as well as true asexual reproducers. Our point is thatthe conditions which open the door to outlaws in Kondrashov’s model mayalso close the door to a di!erent form of check on outlaws not discussed byKondrashov. If there are such trade-o!s in nature, the relevant Hoyle prop-erties cannot simply be added together to produce a Hoyle score reflectingevolvability.

For simplicity, Kondrashov assumed that each o!spring is formed fromexactly four initial cells of a single parental organism (see Figure 1). Heconsidered four di!erent ways these could be sampled from parental cell linesto initiate an o!spring. In the ‘false mode,’ all four founding cells are related

Figure 1. Kondrashov’s four modes of obligate vegetative reproduction: 1. ‘false,’ 2. ‘sectorial,’ 3.‘random,’ and 4. ‘structured.’ The top panel shows initial founding cells from the reproducingparent dividing to form cell lineages which variously contribute to the subsequent founding ofo!spring for another generation (after Kondrashov 1994, Figure 1: 312). The bottom panel stylizesthe contributions of initial cells under the four modes to show their di!erences.

532

as closely as possible, being drawn from a recently derived cell within theparent descending from a single initial cell that formed the parent.7 In the‘sectorial mode,’ all four founding cells of an o!spring descend from a singleinitial cell forming the parent, but are drawn at random from across thesector or clade of cells descended from that initial cell. Here the cells of thevegetative propagule are not as closely related as in the false mode, givingmore scope for somatic mutation to produce outlaws. In the ‘random mode,’cells are sampled at random from all terminal (developed) cells of the parent,so they could come from cell lines descending from any or all of the fourinitial cells that formed the parent. In the ‘structured mode,’ one cell issampled at random from each of the four cell lines that established theparental organism.

In Kondrashov’s models, all cell lines divide synchronously, each goingthrough the same number of cellular generations. Thus, transmission for allfour modes of vegetative reproduction is simultaneous (C2). However, as weshift from false toward structured mode, the sampled initial cells increase inmutational load due to somatic mutation. Thus, the potential for outlawbehavior increases in the progression from false to structured mode. Kon-drashov showed that the ‘structured mode’ sustains the highest mutationalload of all four modes of vegetative reproduction he considered, since thismode leads to o!spring that are on average the most chimaeric (see alsoGrosberg and Strathmann 1998). At the other extreme, Kondrashov’s ‘false’mode has a much lower mutational load, approximating that of true asexualreproduction with a one-cell bottleneck. False mode vegetative reproductionmeets Sterelny’s ‘vertical’ transmission criterion but at the cell lineage ratherthan single cell level (C1), similar to the way in which plants reproduce from atotipotent meristem.

Now consider the combination of satisfaction of C2 (simultaneity) and failureto satisfy C1 (verticality) in the structured mode. In this mode, cells of each ofthe four lineages are included in every o!spring. Sterelny emphasizes thatsimultaneity prevents outlaw cells that might reproduce faster (perhaps becausethey contribute less to somatic welfare) from getting into o!spring first and‘closing the door,’ excluding other, more cooperative cells joining in from othercell lineages. But by the same token, cells that might reproduce faster becausethey reap benefits from within-cell-line cooperation cannot get into o!springfaster than selfish cells which compromise their own cell line. In the structuredmode, even these outlaws are guaranteed a place at the o!spring’s tablewhereas, in random mode, there could be among-cell-lines selection favoringcooperation within lines.

Put simply, if reproduction is vegetative, with several to many varying cellsof di!erent initial lineages in the parent contributing to o!spring-formation,then simultaneous but non-vertical transmission (in the sense of Kondrashov’s

7‘False’ because this mode of vegetative reproduction resembles true asexual reproduction but isnot since the false mode involves a multicellular propagule.

533

structured vegetative mode) may increase the likelihood of outlaws. Simulta-neity may help control outlaws that put their e!ort into faster selfish replica-tion within cell lines, but at the cost of eliminating the possibility of closing thedoor on outlaws through asynchronous transmission or cell-lineage levelselection. Simultaneity only blocks outlaws if it is generally true that outlawswould ceteris paribus ‘get there first.’

In general, we should expect mutation/selection and simultaneity/selectionbalances within the organism that determine whether a particular anti-outlawcondition actually blocks outlaws. Sperm competition might be an analogouscase in which the simultaneous transmission of many variable sperm competeto fertilize an egg, since there is evidence that sperm can apparently cooperatein fertilization (Moore et al. 2002). The door to fertilization is closed once theegg is depolarized by a sperm, but it is the faster swimmer(s), not necessarilythe faster replicator, that wins, at least in those cases where fertilization occursin ‘ballistic’ episodes (C4). Simultaneity in fertilization would give the fastreplicator that is a slower swimmer an opportunity it otherwise would nothave. Thus, there are modes of reproduction for which C1 and C2 may notjointly constitute anti-outlaw conditions, so simply counting up properties onthe Hoyle list may not estimate how suitable a system is, given its particularnatural history, for sustaining an evolvable world.

The Hoyle criteria can only be as generally useful in evaluating inheritancesystems for evolvability as our understanding of the relations among the criteriaand their joint applicability to a wide variety of taxa. Most surveys of thesekinds of properties (including Kondrashov’s; see also Buss 1987) note that thereis much terra incognita. We do not, however, doubt that Sterelny has described avery important list of properties or that his judgments about the Hoyleyness ofthe examples he discusses are correct. But for Hoyleyness to be an operationalbasis for such judgments, more work is needed to understand interrelationshipsamong these very interesting criteria as well as fuller discussion of how they maytrade-o! in the taxa among which they have been studied.

Other potential dependencies among the properties on Sterelny’s list chal-lenge the evaluation strategies of Hoyle & Co. in more fundamental ways. It isnot clear to us to what extent criteria C9 (modularity) and C8 (smooth map)di!er since, as we noted above, Lewontin’s notion of quasi-continuity includesat least some features of modularity but also of smoothness, provided we areentitled to make assumptions about the phenotype/fitness map that are com-monly made by evolutionary geneticists. Perhaps this non-independence is dueto the fact that many have formulated notions of modularity in terms ofgenotype/phenotype mapping relations.

Moreover, it seems to us that C8 may entail C7, given a very reasonablefurther assumption. C8 says that the replicator/organization map should besmooth (quasi-continuous), since otherwise a series of mutations cannot beselected in a way that ‘tracks’ changing environments (i.e. the conditionscausing changes on fitness landscapes). C7 says that there should be a verylarge, possibly unbounded, set of possible variants, i.e. that hereditary systems

534

should be ‘unlimited’ sensu Szathmary (1995, 1999). The argument for C7 isthat if the number of possible variants is limited, then all possible variants canbe simultaneously realized in an actual population such that selection willrapidly ‘find’ the globally optimal variant and eliminate all the rest, bringingevolution e!ectively to a halt since no mutation can then invade against theglobal optimum. But suppose in addition that the array of phenotypes is acontinuum or quasi-continuum, e.g. height, running speed, enzymatic reactionrate, or prey size. Then the assumption of a smooth mapping (C8) entails thatthere is at least a quasi-continuum of genotypes/replicator sets. This is becausenearby points in genotype space have to map to nearby points in phenotypespace for the mapping to be smooth, but for there to be enough points, thegenotype space has to be continuous as well unless C9 (modularity) is violated.However, actual population sizes cannot be on the same order of magnitude asthe continuum (i.e. infinite), hence the possible number of variants must exceedthe actual number realizable. Thus, to the degree that phenotype distributionsare (quasi-)continuous, that fact together with C8 entails C7.

Why is this logical dependency a problem for Sterelny’s Hoyle strategy forevaluating the evolvability of inheritance systems? It’s because the strategyinvolves counting how many of the properties on the list are satisfied. The moreproperties on Sterelny’s list are satisfied, the closer to full Hoyle the inheritancesystem is. However, if some properties are logically dependent on others, thenthe count will overestimate the Hoyleyness of those systems satisfying theentailing properties. Incompatibility of properties, such as a trade-o! of C1and C2 due to between-cell-lines selection for structured mode vegetativereproduction, underestimates Hoyleyness. Even if we look only for partialorderings, where one system satisfies a subset of Hoyle properties satisfied byanother system, we still may not be justified in concluding the former is lessHoyley than the latter. This is because satisfaction of each of these subcon-ditions also comes in degrees and because the more Hoyley system may satisfyproperties that entail the ones the less Hoyley system satisfies. Thus, if theproperties are not additive, so that satisfaction of two of them may not becloser to a su"cient anti-outlaw block (say) than one of them, then partialordering will not estimate proximity to full Hoyle.

A second major problem with the Hoyle strategy, also discussed by Gri"thsand Gray (essay 16), is that Sterelny characterizes evolvability or adaptabilityin terms of modern living systems that do indeed have a fairly full Hoylegenetic system. Because of the focus on modern, evolved systems, the problemSterelny says Hoyle & Co. faces is artificially posed. The problem is not tosupply an empty planet with an inheritance system su"cient, in the ways thisone is, to sustain the evolvable modern world, but rather how to engineer achemical or proto-biological world in which such an inheritance system (orsystems, see below) could itself evolve. We need to explain the evolution ofevolvability, not just the state of evolvability. Sterelny, along with Gri"ths andGray (essay 16), acknowledge this problem of ‘evolutionary transition’ – theevolution of new levels of organization (Maynard Smith and Szathmary 1995)

535

– as a separate problem from that of ‘evolvability’ – the capacity to evolve new,complex adaptations at a given level.

We think, however, that the problem of evolutionary transition presents asti! challenge to Sterelny’s approach to conceptualizing extended inheritanceand that, in general, adaptationists have underestimated the importance of atheory of origins to answer their questions about adaptation. Sterelny supposesthe problem is to measure various systems against the Hoyle scale in order tojudge which can (or did) play the right sort of evolutionary role to meritexplanatory notice, though he does admit that some evolutionary transitions(e.g. symbiogenesis of the eukaryote cell) might have been lucky accidents(342). If we ask instead about the evolution of inheritance itself, as a keyfeature of both evolutionary transition and evolvability, we surely need notsuppose that inheritance systems typically rise by macromutation or luckyaccident to whatever position on the Hoyle scale they now have. But if positionon the Hoyle scale is itself the product of evolution, one must wonder whetherthe conditions su"cient at evolvable worlds played any necessary role in theevolutionary transition to such worlds.8 More fundamentally, it is not clearwhether evolutionary transitions require the concentration of Hoyle propertiesin a single ‘system’ of inheritance or instead could be distributed over several.Certainly, many theories assume that the origins of life involved at least asuccession of inheritance processes or systems, if not several emerging inparallel (e.g. Dyson 1985; de Duve 1991; Maynard Smith and Szathmary 1999;Ganti et al. 2003). The possibility that Hoyleyness is distributed would suggestthat there could be an evolvable world in which only the distributed system is(close to) full Hoyle, while the several inheritance systems linked into a givendevelopmental system are each very far from it.9 Moreover, it might be the casethat the ‘centralized control’ evident in the modern genetic inheritance system(Maynard Smith and Szathmary 1995: 9–10) is an evolved and fairly lateproperty of inheritance, on a scale of billions of years, so that the coincidenceof the Hoyle properties in a single inheritance system is itself the product of anevolvable world and not a precondition for one.

The general problem, then, is that su"cient conditions for a modernevolvable world only give partial clues to the necessary conditions for evolu-tionary transitions. We may be able to recognize the problems that have to beovercome if a newly emergent level of reproductive organization is to sustainan evolvable world, but the properties in virtue of which these problems havetended historically to be overcome is the question at issue for assessing theexplanatory place and privilege of inheritance systems of varying degrees ofHoyleyness. Thus, it may be the case that germ/soma di!erentiation has beenthe historical process through which some lineages achieved some of the higher

8See Forber (op. cit., note 5) for related arguments.9As noted previously (see note 3), it becomes critical to delimit inheritance systems and any solutionrisks begging the question at issue between developmental systems and multi-channel extendedinheritance.

536

evolutionary transitions, even though it appears that a one-cell bottleneck maybe su"cient to block outlaws despite the lack of early germ/soma di!erentia-tion (Grosberg and Strathmann 1998).

Suppose we idealize a system of developmental resources into three com-ponent interacting subsystems with a crude division of labor: one serves as a‘metabolism’ whose function is to harvest energy and matter and direct it intosystem growth and reproduction, one serves as a bounding membrane whosefunction is to insure strong interactions within and weak interactions without,and one serves as a regulator-controller-modulator of the others.10 Supposefurther that each subsystem is autocatalytic (in the right sort of nutrientenvironment of course), and that reproduction of the supersystem requires theparallel propagation of all. Now let us suppose the Hoyle properties are dis-tributed at random among the subsystems. Is there any reason to think that thesystem as a whole could not sustain an evolvable world even if none of thecomponents could? But if it could, then this would mean that we must choosethe right level of analysis in order to evaluate Hoyleyness: is there one inher-itance system because the component subsystems are coupled together or arethere three systems because they are transmitted in parallel? On this score, DSTcan easily claim that any rejection of a case of ecological engineering as too farfrom full Hoyle to merit the name ‘ecological inheritance system’ merelymistakes the ecological subsystem for the whole system of interacting devel-opmental resources. And does not this lead to heredity holism, as DST argued?Of course, this strategy encumbers DST with the obligation to explain theevolutionary transition that led to the ecological inheritance system in questionbecause that will be the only way to establish individuation criteria whichdistinguish the whole from its subsystem parts. As Gri"ths and Gray (1994)argued, individuation of developmental systems ultimately depends on anontology of processes in which an interaction counts as part of the process ifand only if it has an evolutionary explanation. Since here we are concernedwith the evolution of inheritance systems themselves, the evolutionary expla-nation will have to be a transition explanation of how the inheritance systemcame to be full Hoyle enough to sustain an evolvable world.

Perhaps, though, we should not focus on the general evolutionary conditionsfor the origin of evolvable inheritance systems. Consider instead the distribu-tion of actual modes of development among extant taxa. What one noticesabout tabulations of this kind (e.g. Buss 1983, 1985, 1987) is that, in virtue ofthe mode of development of a germ-line (reproductive cells that pass throughthe one-cell bottleneck), significant parts of the evolved world must fall fairlyshort of full Hoyle, despite having inheritance systems that clearly do sustainan evolvable world exhibiting complex adaptations which resulted from nat-ural selection. Plants, for example, have a totipotent meristem in which there isfull scope for outlaws. Although it is true that plants ‘block’ outlaws in somatic

10This is, in essence, Ganti’s chemoton model of the minimal organization of a living system (Gantiet al. 2003).

537

support tissues because they have rigid cell walls which prevent cells frommoving to the sites where reproductive tissue develops, they also must have a‘somatic embryogenesis mode of development, since one tissue type is toti-potent’ (Buss 1985: 495). Buss concludes that ‘Hence plants are largely pro-tected from exploitation by variants in their supporting tissues, but at theexpense of allowing production, proliferation, and propagation of potentiallydebilitating variants within the meristematic cell line’ (ibid., italics added). Inother words, there is in fact a trade-o! between one outlaw blocking mecha-nism and another in plants: stamping it out in somatic tissues with rigid cellwalls but leaving the meristem totipotent a!ords such an opportunity. Bussargues that there is a fairly broad cellular trade-o! between ability to move andability to divide, due to a mechanical constraint of the microtubule apparatus.This sort of trade-o! is far from necessary, of course, and many have arguedthat the key constraint that ‘solves’ the trade-o! is the one-cell bottleneck(Grosberg and Strathmann 1998). Whatever the generalities may turn out tobe, while we agree with Sterelny and the biologists that outlaw-blocking is akey feature of evolvable systems, the fact that mechanisms can trade-o! in thisregard and blocking criteria may be non-independent make it very di"cult toevaluate Hoyleyness at present. We conclude that Hoyle scoring is not yet acompelling reason to reject DST’s holistic attitude toward heredity, eventhough we favor a multi-level, multi-channel approach to extending inheri-tance, as does Sterelny.

Behavior modulation

Much of the DST critique of contemporary biology focuses on the relationbetween heredity and development, emphasizing the importance of develop-mental systems in understanding the role of resources besides genes in thedetermination and transgenerational stability of traits. As is well documentedin Cycles, motivation for the critique grew out of dissatisfaction with the waygene-centrism dealt with behavior, culminating in the language of ‘genes forbehaviors,’ which helped fuel the sociobiology debate. However, the relationbetween behavior and development is also of concern to DST. Moreover,attempts to integrate behavior with evolution from a developmental processperspective have led to further tensions, particularly over the value of Dar-winian approaches to behavior, since neo-Darwinism is a refuge of the gene-centrism DST opposes. Of particular note in many of the essays in Cycles is theview that behavior may play an important role as modulator of the relationbetween development and evolution. If so, behavior should be integral to anytheoretical perspective on the life cycle, not a poor cousin to be let in by theback door on holidays but never on workdays. The question, though, is how toconceive the relation between behavior and development. Should behavior betreated, for example as Weber and Depew (essay 18) do, as a further resource

538

which may interact with others in the developmental process (249), or as aprocess or aspect of process in its own right which somehow integrates withother processes or aspects such as development, evolution, and inheritance? Ina word, the integration of behavior presents as many challenges as doesdevelopment. DST’s dual system/process ontology for development does notyet clearly resolve these fundamental matters.

In this section, we consider behavior modulation – the way in whichbehavior alters the performance of a developing system so as to modulate theimpact of variable contexts and thus a!ect fitness. We discuss an appositeresearch program in biology – evolutionary physiology – which DST mighttake on board as Cycles has done with niche construction.

A number of the essays inCycles comment on the potential for behavior to actas a modulator of the e!ects of variation (both genetic and environmental) so asto yield a stable outcome in the process of development. Active modulation orself-modulation of developmental performance may be achieved throughbehavioral feedback. Bateson (essay 13) points out that ‘The individual may beable, through its behavior, to match its environment to suit its own character-istics’ (156). Klopfer (essay 14) describes how the behavior of a mother goatmodulates the variation in vigor of her kids, so that ‘initial di!erences in strengthand weight are muted, and the weaker animals are not necessarily relegated tothe status of subordinate’ (170). Gordon (essay 12) points out that older coloniesof red harvester ant ‘respond in a more stable way to environmental perturba-tions than do younger ones’ (141), tracing this result to the di!erent possibilitiesfor the development of colony-level behavior in older colonies, which are large,than in younger colonies, which are smaller. In consequence, the ability of acolony to modulate e!ects of variation must itself develop. Nijhout (essay 11)argues that complexity in the genotype/phenotype relation leads to bu!ering orcompensation of the phenotype so as to render the organism insensitive tovariation, which will therefore be rendered selectively neutral (contingent neu-trality). Weber andDepew (essay 18) note that ‘there will exist selection pressurefor behavioral innovations that help organisms respond to contingencies in theirenvironments,’ and argue that the reason behaviors, specifically learnedbehaviors, ‘are extremely e!ective ways of assuring matches between organismsand otherwise shifting environments’ is that the agency of organisms stabilizesenvironments and so renders organismal traits heritable (249).

Each of these observations about a potentially significant role of behavior inmodulating the e!ect of selection on developmental processes (by renderingphenotypes insensitive to variational inputs of genomes and environments)inspires insights into the specific developmental processes under discussion inthe various essays. But they are paralleled by a research program that to ourknowledge has not been discussed in the DST literature, which seeks to oper-ationalize the impact of the modulating role of behavior in evolution: ‘evolu-tionary physiology’ (Feder et al. 2000; Kingsolver and Huey 2003). Much ofthis work can be traced back to Stevan Arnold’s paper, ‘Morphology, Per-formance and Fitness’ (1983; see also Arnold 2003). The research program

539

Arnold proposed and began 20 years ago sought to bring together functionalmorphology and phenotypic selection theory. Arnold used insights fromquantitative genetics and selection theory (primarily Lande’s selection gradientapproach, e.g. Lande 1979; Lande and Arnold 1983) to argue that fitnessconsequences of morphology could be empirically studied by decomposing theproblem into two parts: the causal path from morphology to ‘performance’ andthe path from performance to fitness (or fitness components). Although theearly conception of these paths was in terms of linear partial regression orWright’s path analysis, more sophisticated approaches are possible (Arnold2003). The payo! is as much methodological as conceptual: since the firstcomponent path can sometimes be studied e!ectively in the laboratory whilethe second can be pursued in the field, selection gradient theory a!orded thetechnical means to isolate subproblems best studied in these di!erent contexts.At the time Arnold presented his ideas, functional morphologists were asskeptical of the possibility of measuring fitness as evolutionary biologists wereincredulous of morphologists’ disbelief in individual variation in morphologywithin populations.

Of great additional benefit conceptually and of most relevance to DST is theidea that performance is in general a causal mediator between morphology andfitness. We can think of performance as a functional notion of behavior –working a morphological mechanism or set of parts so as to produce an e!ect.In the context of evolution, the interesting e!ects are those that influence fitnessor its components so that the linkages can be interpreted in terms of their rolesin the evolution of complex adaptive traits, like feeding mechanisms. Mor-phology comprises the mechanisms which, given a range of environments,a!ord behavioral capacities. These in turn (and again in a given range ofenvironments) a!ord an array of possible fitness consequences. The perfor-mance or working of these capacities throughout the developmental processresult in a lifetime fitness outcome. Whether through stimulus-response, trial-and-error, or social learning mechanisms, behavior may modulate a perfor-mance with respect to a particular array of environments. A snake that comesupon a particularly large egg may work its jaw muscles in a di!erent way tomanipulate the prey in its mouth than if the egg were smaller. The behaviormodulates the environmental variation in prey size, so that a snake whose jawmechanism is adapted to smaller prey sizes may nevertheless accommodatelarger ones. Thus, despite environmental variation or, analogously, geneticvariation in the morphological inputs, performances may achieve a stablephenotypic e!ect through the modulating e!ects of behavior. Many familiarphenomena may be brought under this perspective, e.g. habitat selection, matechoice, and thermoregulation. Moreover, many of the behaviorally plastictraits discussed by those essays in Cycles concerned with behavior discussexamples like these. Bateson, e.g. (152) considers color switching between greenand black morphs in African grasshoppers. It is interesting to note thatBateson does not see a conflict between Dawkins’ gene-centrism and devel-opmental systems theory (162) because he views the former as ‘teleological’ and

540

the latter as ‘causal’ explanation. Evolutionary physiology explicitly bringsthese two explanatory forms together. Functional morphology is a mode ofcausal explanation which for a long time eschewed teleological or evolutionaryissues (transformed, however, into the evolutionary morphology of biologistslike David Wake and Stevan Arnold).11 Since selection gradient theory isexpressly pitched at analyzing the force of selection on suites of traits linked bygenetic covariation, it makes many of the assumptions that DST criticizes, andyet it successfully models a phenomenon which is of direct relevance to DST asa scientific research program.

Evolutionary physiologists, interested in phenotypically and behaviorallyplastic traits in an evolutionary context, address issues that are already ofdirect interest to developmental systems theorists. But phenotypic selectiongradient theory is controversial (Turelli 1984; Barton and Turelli 1989) pre-cisely because it must be harnessed to a genetic theory in order to have evo-lutionary implications. And here the di!erences between ‘causal’ and‘teleological’ (Bateson essay 13), or ‘Gene-D’ and ‘Gene-P’ (Moss essay 8), ormolecular process and evolutionary (Neumann-Held essay 7), conceptions ofgenetic units matter. Posed as an issue of methodology or research strategy, thecore problem facing DST, as it tries to take on the function of a theoreticalperspective to foster scientific research, is that evolutionary physiology, likeniche construction, makes assumptions philosophically at odds with its de-velopmentalism. In order to take on board these and other scientific researchprograms (epigenetic inheritance, gene-culture dual inheritance, symbiogenesis,evolutionary transition), DST must learn to live with conflict among its coreassumptions. In our view, scientists are adept at this and the reason is that theytend to care more for the robustness of their results than they do the purity orconsistency of their theories.

We close by calling again for a strategy of robustness analysis to assessempirical and theoretical results in light of the false, but heuristic idealizingassumptions that are inevitable in empirical inquiry. Successfully trans-forming DST from descriptive critique and philosophy of nature into a sci-entific research perspective may require backing o! developmental-ism inorder to embrace and evaluate a variety of modeling strategies. Theoreticalpurity spurs creativity primarily in theoreticians while empirical biologiststend to flourish best in a heterogeneous conceptual environment. And sincescientists must be concerned, eventually and ultimately, with the worldaround them, any biological research program worth its salt must not onlylive with the consequences of conflicting false idealizations among its diversemodels but also embrace them. As DST matures – Oyama’s Ontogeny ofInformation will be 20 next year – we look forward to a flourishing ofempirical work inspired by the discovery of robust results on the behavior ofdevelopmental systems.

11For a recent reflection on evolutionary morphology, see Love (2003).

541

Acknowledgements

The authors wish to thank Susan Oyama, Paul Gri"ths, Russell Gray, theother authors in Cycles, and Kim Sterelny for their patience. The first authorthanks the Dean of Social Sciences, University of California, Davis, forfinancial support. We all thank Michael Turelli and EVE for providing labspace.

References

Arnold S.J. 1983. Morphology, performance and fitness. Am. Zool. 23: 347–361.Arnold S.J. 2003. Performance surfaces and adaptive landscapes. Int. Compar. Biol. 43: 367–375.Avital E. and Jablonka E. 2000. Animal Traditions: Behavioural Inheritance in Evolution. Cam-

bridge University Press, New York.Barton N.H. and Turelli M. 1989. Evolutionary Quantitative Genetics: How Little Do We Know?

Ann. Rev. Genet. 23: 337–370.Bolker J.A. 2000. Modularity in development and why it matters to Evo-devo. Am. Zool. 40(5):

770–776.Bonner J.T. 1974. On Development; The Biology of Form. Harvard University Press, Cambridge.Boyd R. and Richerson P.J. 1985. Culture and the Evolutionary Process. University of Chicago

Press, Chicago.Boyd R. and Richerson P.J. 1996. Why culture is common, but cultural evolution is rare. Proc. Br.

Acad. 88: 77–93.Brandon R. 1999. The units of selection revisited: the modules of selection. Biol. Philos. 14:

167–180.Buss L.W. 1983. Evolution, development, and the units of selection. Proc. Natl. Acad. Sci. United

States Am. 80(5): 1387–1391.Buss L.W. 1985. The uniqueness of the individual revisited. In: Jackson J.B.C., Buss L.W. and

Cook R.E. (eds), Population Biology and Evolution of Clonal Organisms. Yale University Press,New Haven, pp. 467–505.

Buss L.W. 1987. The Evolution of Individuality. Princeton University Press, Princeton.Dawkins R. 1982. The Extended Phenotype: The Gene as the Unit of Selection. Oxford University

Press, New York.deDuve C. 1991. Blueprint for a Cell: The Nature and Origin of Life. Neil Patterson Publishers,

Burlington.Dyson F. 1985. Origins of Life. Cambridge University Press, Cambridge.Feder M.E., Bennett A.F. and Huey R.B. 2000. Evolutionary Physiology. Ann. Rev. Ecol. Syst. 31:

315–341.Gannett L. 1999. What’s in a cause? The pragmatic dimensions of genetic explanations Biol. Philos.

14: 349–374.Ganti T., Griesemer J. and Szathmary E. 2003. The Principles of Life, with a Commentary by

James Griesemer and Eors Szathmary. Oxford University Press, Oxford.Gilbert S. and Bolker J.A. 2001. Homologies of process and modular elements of embryonic

construction. J. Exp. Zool. (Mol. Dev. Evol.) 291: 1–12.Gray R. 1992. Death of the Gene: Developmental Systems Strike Back. In: Gri"ths Paul E. (ed.),

Trees of Life. Kluwer Academic Publishers, Dordrecht, pp. 165–209.Griesemer J. 2000a. Development, culture and the units of inheritance. Philos. Sci. 67: S348–S368.Griesemer J. 2000b. Reproduction and the Reduction of Genetics. In: Beurton P., Falk R. and

Rheinberger H.-J. (eds), in Development and Evolution, Historical and Epistemological Per-spectives. Cambridge University Press, Cambridge, pp. 240–285.

Griesemer J. 2000c. The units of evolutionary transition. Selection 1: 67–80.

542

Griesemer J. 2002. Limits of reproduction: a reductionistic research strategy in evolutionarybiology. In: van Regenmortel M.H.V. and Hull D.L. (eds), Promises and Limits of Reductionismin the Biomedical Sciences. Wiley, Chichester, pp. 211–231.

Griesemer J. and Wade M.J. 2000. Populational Heritability: Extending Punnett Square Conceptsto Evolution at the Metapopulation Level. Biol. Philos. 15(1): 1–17.

Gri"ths P.E. 2001. Genetic information: a metaphor in search of a theory. Philos. Sci. 68(3):394–412.

Gri"ths P.E. and Gray R.D. 1994. Developmental systems and evolutionary explanation. J. Philos.91(6): 277–304.

Gri"ths P.E. and Knight R.D. 1998. What is the developmentalist challenge? Philos. Sci. 65(2):253–258.

Gri"ths P.E. and Gray R. 1997. Replicator II – judgement day. Biol. Philos. 12: 471–492.Grosberg R.K. and Strathmann R.R. 1998. One cell, two cell, red cell, blue cell: the persistence of a

unicellular stage in multicellular life histories. TREE 13(3): 112–116.Jablonka E. 2002. Information: its interpretation, its inheritance, and its sharing. Philos. Sci. 69(4):

578–605.Jablonka E. and Lamb M. 1995. Epigenetic Inheritance and Evolution. Oxford University Press,

Oxford.Jenuwein T. and Allis C.D. 2001. Translating the histone code. Science 293(5532): 1074–1080.Kingsolver J.G. and Huey R.B. 2003. Introduction: the evolution of morphology, performance,

and fitness. Int. Compar. Biol. 43: 361–366.Kondrashov A.S. 1994. Mutation load under vegetative reproduction and cytoplasmic inheritance.

Genetics 137: 311–318.Lande R. 1979. Quantitative genetic analysis of multivariate evolution, applied to brain-body size

allometry. Evolution 33: 402–416.Lande R. and Arnold S.J. 1983. The measurement of selection on correlated characters. Evolution

37: 1210–1226.Levins R. 1966. The strategy of model building in population biology. Am. Sci. 54: 421–431.Levins R. 1968. Evolution in Changing Environments, Some Theoretical Explorations. Princeton

University Press, Princeton, NJ.Lewontin R.C. 1978. Adaptation. Sci. Am. 239(3): 213–230.Love A.C. 2003. Evolutionary morphology, innovation, and the synthesis of evolutionary and

developmental biology. Biol. Philos. 18: 309–345.Maynard Smith J. 2000. Concept of information in biology. Philos. Sci. 67(2): 177–194.Maynard Smith J. and Szathmary E. 1995. Major Transitions in Evolution. Freeman Spektrum,

Oxford.Maynard Smith J. and Szathmary E. 1999. The Origins of Life: From the Birth of Life to the Origin

of Language. Oxford University Press, New York.Moore H., Dvorakova K., Jenkins N. and Breed W. 2002. Exceptional sperm cooperation in the

wood mouse. Nature 418: 174–177.Odling-Smee F.J., Laland K.N. and Feldman M. 2003. Niche Construction: the Neglected Process

in Evolution. Princeton University Press, Princeton.Oyama S. 1985. The Ontogeny of Information: Developmental Systems and Evolution. Cambridge

University Press, New York. (2nd ed. 2000, Duke University Press, Durham).Oyama S. 2000. Evolution’s Eye: A Systems View of the Biology-Culture Divide. Duke University

Press, Durham.Ra! E.C. and Ra! R.A. 2000. Dissociability, modularity, evolvability. Evol. Develop. 2(5):

235–237.Sharon N. and Lis H. 1989. Lectins as cell recognition molecules. Science 246(4927): 227–234.Sterelny K. 1996. Explanatory pluralism in evolutionary biology. Biol. Philos. 11: 193–214.Sterelny K., Smith K. and Dickison M. 1996. The Extended Replicator. Biol. Philos. 11: 377–403.Szathmary E. 1995. A Classification of replicators and Lambda–Calculus models of biological

organization. Proc. Roy. Soc. London (Ser. B Biol. Sci.) 260(1359): 279–286.

543

Szathmary E. 1999. Chemes, Genes, Memes: A Revised Classification of Replicators. Lect. Math.Life Sci. 26: 1–10.

Turelli M. 1984. Heritable genetic variation via mutation-selection balance: Lerch’s Zeta Meets theAbdominal Bristle. Theor. Popul. Biol. 25(2): 138–193.

Turner B.M. 2000. Histone acetylation and an epigenetic code. BioEssays 22(9): 836–845.Wade M.J. and Griesemer J. 1998. Populational heritability: empirical studies of evolution in

metapopulations. Am. Nat. 151(2): 135–147.Wagner G.P. and Altenberg L. 1996. Perspective – complex adaptations and the evolution of

evolvability. Evolution 50(3): 967–976.Wimsatt W.C. 1974. Complexity and organization. In: Scha!ner K.F. and Cohen R.S. (eds), PSA

1972. D. Reidel, Boston pp. 67–86.Wimsatt W.C. 1976. Reductive explanation: a functional account. In: Michalos A.C., Hooker C.A.,

Pearce G. and Cohen R.S. (eds), PSA 1974. D. Reidel, Boston pp. 671–710.Wimsatt W.C. 1985. Forms of aggregativity. In: Donagan A., Perovich A. and Wedin M. (eds),

Human Nature and Natural Knowledge. D. Reidel, Dordrecht, pp. 259–293.Wimsatt W.C. 1997. Aggregativity: reductive heuristics for finding emergence. Philos. Sci. 64(4

SUPPS): S372–S384.Winther R.G. 2001. Varieties of modules: kinds, levels, origins, and behaviors. J. Exp. Zool. (Mol.

Dev. Evol.) 291: 116–129.

544