The Structure of Evolutionary Theory

by Stephen Jay Gould

  Genetic change cannot, of itself, specify the causal level of sorting because selection at any higher level sorts individuals at all included lower levels as an automatic effect, and not necessarily for direct causal reasons at all. Two ba­sic considerations bar inferences of cause from the genetic account books alone. First, an observation of genetic sorting doesn't specify the relevant causal level. A gene associated with strong jaws may increase in frequency within the class Polychaeta because polychaete organisms with strong jaws out compete weaker-jawed conspecifics in organismic selection; because poly­chaete species with strong jaws also develop emergent populational charac­ters that defeat weak-jawed species in species selection; or because strong-jawed polychaetes do especially well in driving out jawless priapulids by clade selection.

  Second, even when we can identify the level of causality for an incident of genetic sorting, we cannot know (from the increase in frequency alone) whether the gene sorted positively has prevailed by a selected effect upon the phenotype, or for a set of possibly nonadaptive reasons. Does the gene associ­ated with strong jaws actually promote the construction of this phenotypic basis for organismic selection, or has this gene hitchhiked to greater fre­quency by close linkage with another gene that does build the selected pheno­type? Nonadaptive possibilities only increase for selection at higher levels. If polychaetes have increased by clade selection over priapulids, does the plurified polychaete gene big-A build part of the relevant priapulid-beating phenotype, or does big-A just count as one of the myriad polychaete genes that happen to specify, by homology and a few hundred million years of evo­lutionary separation, the historical uniqueness of the clade?

  The nature of hierarchies dictates a choice of genes as optimal units of bookkeeping. The nature of hierarchies also creates a possibility — then real­ized in nature for fascinating reasons largely unknown, and mostly beyond the scope of this book (but see Buss, 1987) — for structuring the world of biol­ogy as a hierarchy of individuals, each encompassing the ones below as new levels accrete in evolution, and each capable of acting as a unit of selection, a causal agent of Darwin's expanded theory.

  Gambits of reform and retreat by gene selectionists

  As I have emphasized throughout this section, gene selectionism can't be made to work as a general philosophy. The logic of the theory does not co­here, and the system cannot attain consistent completion. Yet the allure of the [Page 638] gene remains powerful, largely for reasons of general preference in our cul­ture, rather than for any observed power or intrinsic biological status pos­sessed by evolutionary individuals of this lowest level. When an incoherent argument remains intriguing, and supporters cannot bear the wrench of total abandonment, a favored theory must be festooned with compromises and “howevers,” or so changed in form that only lip service remains to cover a truly altered substance. Often, given human tendencies to paint a bright face on adversity, gene selectionists have made their necessary retreats, but pre­sented them as refinements or elaborations of the original theory. In this clos­ing section, I shall show that the two most prominent “revisions” of gene selectionism — Dawkins's extended phenotype (1982) and Williams's codical selection (1992) — represent defeats rather than improvements as advertised.

  DAWKINS ON THE “EXTENDED PHENOTYPE.” I always admired the chutzpah of Senator Aikens' brilliant solution to the morass of our involve­ment in the Vietnamese War. At the height of our reverses and misfortunes, he advised that we should simply declare victory and get out. Richard Dawkins got in with his 1976 book, The Selfish Gene. He declared victory with The Extended Phenotype in 1982 — although he had really, at least with respect to the needs and logic of his original argument, gotten out.

  With admirable clarity, and no ambivalence, Dawkins proclaimed the doc­trine of exclusive gene selectionism in 1976: “I must argue for my belief that the best way to look at evolution is in terms of selection occurring at the low­est level of all ... I shall argue that the fundamental unit of selection, and therefore of self-interest, is not the species, nor the group, nor even, strictly, the individual. It is the gene, the unit of heredity (1976, p. 12). So selection occurs at only one lowest level — the gene, labelled as 'the fundamental unit of selection.' Nothing more inclusive, not even an organism, can be called a unit of selection.”

  Dawkins presented his later work, The Extended Phenotype, as an ex­tension and elaboration of gene selectionism: “This book,” he wrote, “is in some ways the sequel to my previous book, The Selfish Gene” (1982, p. v). Dawkins had admitted, in 1976, that genes work through phenotypes of the “lumbering robots” (organisms) serving as their passive homes. But if genes are nature's real actors, and phenotypes only their means of expression, then why limit phenotypes to bodies? Any consequence of a gene should be equally capable of carrying the gene's interest in a process of selection. Dawkins admitted of course that most aspects of this extended phenotype — the footprint of a shorebird in the sand, for example — will be too ephemeral, or too by the by, to be effective in the gene's interest. But other parts of the extended phenotype (with the beaver's dam as Dawkins's favorite example) do contribute to the success of beaver genes, and should be included within “the extended phenotype” that the gene — the ultimate and only unit of selec­tion (at least in 1976) — can manipulate in its full range of machinations for replicative success.

  Dawkins (1982, pp. iv-vii) therefore insisted that the viewpoint of The [Page 639] Extended Phenotype evolved gradually and progressively from The Selfish Gene. “The present book,” he tells us, “goes further,” presumably in the same direction:

  This belief — that if adaptations are to be treated as “for the good of” something, that something is the gene — was the fundamental assump­tion of my previous book. The present book goes further. To dramatize it a bit, it attempts to free the selfish gene from the individual organism, which has been its conceptional prison. The phenotypic effects of a gene are the tools by which it levers itself into the next generation, and these tools may “extend” far outside the body in which the gene sits, even reaching deep into the nervous system of other organisms. Since it is not a factual position I am advocating, but a way of seeing facts, I wanted to warn the reader not to expect “evidence” in the normal sense of the word.

  So genes have become even more fundamental, and bodies even more inconsequential: “Fundamentally, what it going on is that replicating molecules ensure their survival by means of phenotypic effects on the world. It is only incidentally true that these phenotypic effects happen to be packaged up into units called individual organisms” (Dawkins, 1982, pp. 4-5).*

  But now the argument begins to unravel. Just when the gene seems poised to swallow the organism entirely as just one incidental aspect of the gene's armamentarium (the fully extended phenotype), Dawkins turns around, and tells us that we may treat organisms as focal entitites after all, and describe evolution from the organism's point of view just as well: “I am not saying that the selfish organism view is necessarily wrong, but my argument, in its strong form, is that it is looking at the matter the wrong way up ... I am pretty con­fident that to look at life in terms of genetic replicators preserving themselves by means of their extended phenotypes is at least as satisfactory as to look at it in terms of selfish organisms maximizing their inclusive fitness” (1982, pp. 6-7).

  Shall we then favor the gene or the organism? Dawkins claims to prefer genes and to find greater insight in this formulation. But he allows that you or I might prefer organisms — and it really doesn't matter. In a telling analogy, Dawkins compares genes and organism to the two possible versions (different [Page 640] cerebral resolutions of the same visual reality) in the famous optical illusion known as the Necker Cube:

  After a few more seconds the mental image flips back and it continues to alternate as long as we look at the picture. The point is that neither of the two perceptions of the cube is the correct nor “true” one. They are equally correct. Similarly the vis
ion of life that I advocate, and label with the name of the extended phenotype is not probably more correct than the orthodox view. It is a different view and I suspect that, at least in some respects, it provides a deeper understanding. But I doubt that there is any experiment that could be done to prove my claim (1982, p. 1).

  Moreover, we really needn't quarrel over our choices because the issue can achieve no empirical resolution in any case. I'll push my preference (and hope to persuade you of its greater capacity for mind stretching, its superior liter­ary charm, or its greater tickling of the fancy); and you can then advocate your opposite, and equally valid, version. Dawkins begins his book: “This is a work of unabashed advocacy. I want to argue in favor of a particular way of looking at animals and plants, and a particular way of wondering why they do the things that they do. What I am advocating is not a new theory, not a hypothesis which can be verified or falsified, not a model which can be judged by its predictions... I am not trying to convince anyone of the truth of any factual proposition” (1982, p. 1). This argument about equally valid, but quite inverse, perspectives on a common reality pervades the entire book, as in this late passage (1982, p. 232): “The whole story could have been told in ... the language of individual manipulation. The language of extended genet­ics is not demonstrably more correct. It is a different way of saying the same thing. The Necker Cube has flipped. Readers must decide for themselves whether they like the new view better than the old.”

  Among professional philosophers, such Necker-Cube thinking goes by the name of conventionalism, an argument that frameworks of explanation can­not be judged as true or false, or even more or less empirically adequate — but only as equally correct, and only as more or less preferable by such nonfactual criteria as depth of insight provided or satisfaction gained in understand­ing. Conventionalism may offer an interesting and fruitful approach, espe­cially for some scientific debates that seem especially refractory to empirical resolution — and also (more generally) for teaching people that ideas and atti­tudes influence science; and that “naive realism,” with its assumption that improved theories arise only from observation, represents a silly and bank­rupt approach to the natural world.

  But conventionalism cannot apply to this case because an empirical resolu­tion exists, and the apparent Necker-cube duality of gene or organism does not denote, as Dawkins mistakenly argues, two equally valid perspectives on the same issue, but rather expresses a correct vs. a false view of the nature of causality in Darwinian theory. Dawkins has misconstrued his categories in judging gene-based and organism-based viewpoints as alternative versions of [Page 641] the same kind of explanation. The gene-based view works best for bookkeep­ing, while the organism-based view represents one legitimate level of causal­ity — the one regarded as effectively ubiquitous and exclusive by Darwin him­self. In this sense, both views are valid; but they are not comparable — and genes vs. organisms do not represent alternatives on an identical playing field of common explanatory intent.

  Moreover, Dawkins's shift from the selfish gene to the extended phenotype does not reflect a simple extension or elaboration of a consistent and develop­ing viewpoint. He tries to save face with such a portrayal, but his strategy fails. The conventionalism of The Extended Phenotype negates and denies the explicit defense of gene selectionism as an empirical reality, as presented in The Selfish Gene. Dawkins's first book says, in no uncertain terms (see quotation on page 618), those genes are exclusive units of selection (or causal agents), and that bodies, as passive lumbering robots, cannot play such a role. The second book says that we can view evolution equally well from either the gene's or the organism's point of view, that Dawkins still prefers genes, but that others remain free to favor bodies with just as much claim to empirical adequacy. The disparate logic of these two formulations precludes their inter­pretation as developing versions of the same view of life, and one theory is not a subtler extension of the other. These two positions connote logically contrasting, and mutually exclusive, accounts of causality in evolution. I do not happen to regard either as correct, but I think we can all agree that Dawkins's later view of the extended phenotype derails and controverts his earlier defense of gene selectionism as nature's true way.

  I do not know why Dawkins altered his view so radically. But may I suggest that he simply could not — because no one can after a proper analysis of the basic logic of the case — maintain full allegiance to the fallacious argument of strict gene selectionism. Dawkins tried hard in 1976, but ultimately needed to make so many statements from the organism's point of view that he must have begun to wonder whether he could really continue to regard such organismal language as a mere convenience, while touting the genic formula­tion as a unique reality. Perhaps he finally decided that if organism-based lan­guage seemed so stubbornly ineluctable, then organism-based causality might be equally inevitable, at least as a legitimate option. With such an admission, the selfish gene becomes an impotent meme.

  WILLIAMS'S CODICAL HIERARCHY. Williams's epochal book of 1966 set the intellectual basis for gene selectionism, and may justly be called the founding document for this ultimate version of Darwinian reductionism. But by 1992, Williams had realized that interactors, and not replicators, consti­tute units of selection, or causal agents in the usual sense of the term — and that hierarchy must hold because no level of interaction can be deemed exclu­sive, or even primary. Williams, however, did not wish to abandon his old ap­paratus for viewing genes as fundamental and preferred units of selection. But que faire? Genes are replicators in their only universal role (they can also [Page 642] be interactors in the much more restricted status of one legitimate level in an extensive hierarchy, as discussed on pp. 689–695) — and interactors, not repli­cators, are units of selection in the causal sense.

  Williams therefore tried an interesting gambit. He admitted that interactors form a hierarchy of evolutionary individuals at several levels, and that these interactors are units of selection in our usual sense of material entities partici­pating in a causal process. These interactors build a material hierarchy — and gene selectionism cannot apply to this legitimate domain. Williams therefore established a different and parallel hierarchy* for abstract units of informa­tion (as opposed to material entities) — and he construed genes as basic “units of selection” in this alternative and parallel domain, which he called codical (the adjectival form of codices, the plural of codex, his term for a single unit [Page 643] of information). If genes can't claim exclusivity (or even causal status at all) as units of selection in the usual domain of material objects, then Williams would establish a new and separate hierarchy for nonmaterial units of infor­mation — and here the gene could continue to reign.

  Williams therefore proposed a fundamental distinction between entities and information, speaking of “two mutually exclusive domains of selection, one that deals with material entities and another that deals with information and might be termed the codical domain” (1992, p. 10). But I do not think that the codical domain can claim either meaning or existence as a locus for causal units of selection, for two reasons:

  Odd mapping upon legitimate intuitions. Williams continues his allegiance to the nemesis of gene selectionism, the false criterion that has always doomed the theory to incoherence: faithful replication as the defining property for a “unit of selection” — now reformatted as a unit that only exists in the newly formulated codical domain, for Williams has now admitted that replicators are not causal agents in the usual realm of material entities. Wil­liams promotes his old standard — faithful replication — as the primary crite­rion for “unithood” in his codical domain, thus leading to the following pe­culiar position: genes are units of selection (as the replicating consequence in the codical domain of selection upon organisms in the material domain); gene pools are also units of selection (as replicating consequences of higher-level selection upon groups to clades); but genotypes, in an intermediate category, are not units of selection (exc
ept in asexual organisms, where replication is faithful). Thus the codical domain skips a space in the hierarchy, and contains no organismic level of selection (except for asexual creatures) because the corresponding codex is impersistent.

  The old error of confusing bookkeeping with causality. Williams's complex move in devising a separate hierarchy for nonmaterial units of information (and then juxtaposing this new sequence against the conventional hierarchy of evidently material and admittedly causal units), amounts to little beyond an elaborate and superfluous effort to rescue the un-salvageable theory of gene selectionism by granting both primacy and causal status (but only linguistically) to genes as replicators. But nothing new has been added beyond some terminology. The old error remains in full force — if anything even heightened by the counterintuitive complexities and mental manipulations required operationalizing the scheme of dual hierarchies. A parallel hierarchy for nonmaterial entities of information? What can such a claim mean? Take the idea apart; pull the codical clothing off this new em­peror, and whom do we find naked underneath? our old friend, the book­keeper. Why must he continually try to play on the field of material objects engaged in nature's grand game of causality? Why should he be ashamed of his vital but different role? Bookkeeping is also a necessary, and entirely hon­orable, activity. The results of causal processes must be tabulated, and we rightly treasure the lists. We continue to stand in awe before “60” in Babe Ruth's home run column for 1927, and “70” in Mark McGwire's for 1998. But 70 is a record, not a cause — a summary of a great achievement, not the [Page 644] bat itself, or the muscles in a pair of strong arms. As nonmaterial objects suited for recording, codices are units of bookkeeping.