Over the years, Dawkins has developed a litany of similar admissions. Of course organisms must be regarded as the foci of selection, but since biased gene passage occurs as a result of this process, we may identify genes as agents of selection. (But results are not causes, although foci of action surely are): “Just as whole boats win or lose races, it is indeed individuals [organisms] who live or die, and the immediate manifestation of natural selection is nearly always the individual level. But the long-term consequences of non-random individual death and reproductive success are manifested in the form of changing gene frequencies in the gene pool” (1976, p. 48).
Dawkins then apologizes for framing his descriptions in terms of organisms as causal actors, excusing himself for succumbing to temptations of convenience. (But perhaps we find this mode “convenient” because we achieve the best description of a causal reality thereby — while the genic mode remains tortuous and uncomfortable because we sense the central error in such formulation): “In practice it is usually convenient, as an approximation, to regard the individual body as an agent 'trying' to increase the number of all its [Page 632] genes in future generations. I shall use the language of convenience. Unless otherwise stated, 'altruistic behavior' and 'selfish behavior' will mean behavior directed by one animal body towards another [p. 50] . . . We shall continue to treat the individual as a selfish machine, programmed to do whatever is best for his genes as a whole. This is the language of convenience [p. 71].”
In a later book (1982, p. 4) Dawkins admits that proxy — that is, must select gene replicators via organisms as causal actors: “The most important kind of replicator is the 'gene' or small genetic fragment. Replicators are not, of course, selected directly, but by proxy; they are judged by their phenotypic effects.”
This argument, I think, has truly become an inadaptive meme, destined for eventual extinction, but propagated wherever gene selectionism survives, whether in technical literature or popular presentation. A major popular book on this topic holds (Cronin, 1991, p. 289): “If organisms are not replicators, what are they? The answer is that they are vehicles of replicators... Groups, too, are vehicles, but far less distinct, less unified ... In this weak sense, then, 'group selection' could occur... [but it] would in no way undermine the status of genes as the only units of replicator selection. This does not mean that higher-level entities are unimportant in evolution. They are important, but in a different way: as vehicles.”
Bookkeeping and causality: the fundamental error of
gene selectionism
The error and the incoherence of gene selectionism, as documented above, can be summarized in a single statement illustrating the fruitful, “Pareto-like” character of the central fallacy: proponents of gene selectionism have confused bookkeeping with causality. This error achieves its Pareto status of substantial utility because changes recorded at the genetic level do play a fundamental part in characterizing evolution, and records of these changes (bookkeeping) do maintain an important role in evolutionary theory. But the error remains: bookkeeping* is not causality; natural selection is a causal process, and units or agents of selection must be defined as overt actors in the mechanism, not merely as preferred items for tabulating results.
No one has ever stated the issue more accurately or succinctly than George Williams himself (1992, p. 13), thus increasing my puzzlement at his failure to recognize how his own formulation invalidates the gene selectionism that still wins his lip service: “For natural selection to occur and be a factor in evolution, replicators must manifest themselves in interactors, the concrete realities that confront a biologist. The truth and usefulness of a biological theory [Page 633] must be evaluated on the basis of its success in explaining and predicting material phenomena. It is equally true that replicators (codices) are a concept of great interest and usefulness and must be considered with great care for any formal theory of evolution, either cultural or biological.” Williams's statement agrees completely with the position that I have advanced in this book — an attitude that, by general consensus, leads logically and directly to the hierarchical model of selection, and the invalidity of single-level, gene-based views. Williams allows that interactors represent the “concrete realities” confronting biologists (and chapter 4 of his 1992 book eloquently defends the concept of legitimate interactors at several hierarchical levels of increasing genealogical inclusion). He admits that both the “truth and usefulness” of a biological theory, natural selection in this case, depends upon the explanation of material phenomena — that is, interactors operating as agents. He does not include replicators — the basis of gene selectionism — in this category, for his last sentence grants them a separate but equal status in evolutionary theory: “It is equally true that replicators (codices) are a concept of great interest” needed “for any formal theory of evolution.” Now, if replicators are not causal agents, but are vital for any full account of evolution — then what are they? I suggest that we view gene-level replicators as basic units for keeping the books of evolutionary change — as “atoms” in the tables of recorded results.
Williams did not slip or misspeak in the quotation cited above. He repeats this separation of a causal agent from a basis of hereditary transmission — with interactors as agents and replicators as transmitters — in several other passages, including (1992, p. 38) “Natural selection must always act on physical entities (interactors) ... It is also necessary that there be what Darwin called 'the strong principle of inheritance'...”
Whereas Williams makes valid separations and defines proper roles, but then seems unwilling to own the theoretical consequences, Dawkins, on the other hand, seems merely confused. In discussing group selection (1982, p. 115), for example, Dawkins writes: “The end result of the selection discussed is a change in gene frequencies, for example, an increase of 'altruistic genes' at the expense of 'selfish genes.' It is still the genes that are regarded as the replicators which actually survive (or fail to survive) as a consequence of the (vehicle) selection process.”
By putting the word “vehicle” in parentheses, as a reminder of selection's intrinsic nature rather than a mere modifying adjective, Dawkins admits that interactors (vehicles in his terminology), not replicators, operate as agents of selection. He describes the differential survival of replicators as a consequential result of this causal process — therefore as units for bookkeeping rather than agents of causality — but he then fails to disentangle these two different aspects of evolution, while continuing to grant favored status to genes.
We may indeed, and legitimately as a practical measure, decide to keep track of an organism's success in selection by counting the relative representation of its genes in future generations. (In large part, we count at the genic level for the reason always emphasized by Williams and Dawkins — because [Page 634] sexual organisms do not replicate faithfully and therefore cannot be traced as discrete entities across generations.) But this practical decision for counting does not deprive the organism of status as a causal agent, nor does such a procedure grant causality to the objects counted. The listing of accounts is bookkeeping — a vitally important subject in evolutionary biology, but not a form of causality.
If, as I have argued (see also Wilson and Sober, 1994), the incoherence of gene selectionism denotes a rare case in science of an influential theory felled by a logical error — in this case the confusion of bookkeeping with causality — rather than a fallacious proposal about the empirical world, then we must ask why so many people fell into this error so readily, and why the fallacy did not become more quickly apparent. I suspect that three major reasons underlie not only the error of gene selectionism, but also the strong willingness, even the fervor, expressed by so many evolutionists in embracing the concept. The first two reasons may claim a social basis in traditions of scientific inquiry. But the third reason, and surely the most intriguing from both a scientific and philosophical perspective, emanates directly from th
e logical structure of hierarchies, the conceptual framework that must replace gene selectionism.
The two reasons rooted in traditions of scientific procedure include the most general of statements and a preference peculiar to traditions of Anglo-phonic evolutionary biology. For the generality, I state nothing profound or original in pointing out that a decision to privilege the level of genes plays into the strongest of all preferences in Western science: our traditions of reductionism, or the desire to explain larger-scale phenomena by properties of the smallest constituent particles.
The allure of reductionism encourages the following kind of error, or sloppy thinking: we correctly note that genes play a fundamental role in evolution (as preferred items for a calculus of change — the bookkeeping function); we also recognize that genes lie at the base of a causal cascade in the development of organisms; finally, and most generally, we view genes as the closest biological approach to an “atom” of basic structure, and therefore as the cardinal entity of a reductionistic research program. From these statements, we easily slip into the unwarranted inference that genes must also be fundamental units or agents in natural selection, the primary causal theory of our profession — all the while forgetting the criteria of individuality and interaction that define units or agents within the logic of the theory itself.
The second, and more particular, reason flows from explicit traditions of the Modern Synthesis, especially from the approach favored by Fisherian population genetics (see Chapter 7). The heuristics of this field prospered greatly with models that kept track of gene frequencies without worrying much about the locus of selective action. A common fallacy in science then conflates a practical basis for success with the causal structure of nature. Jim Crow (1994, p. 616), one of the world's most thoughtful geneticists, expressed this point particularly well, but then also failed to distinguish bookkeeping from causality. Writing “In praise of replicators” — and well should we praise them, but, I would argue, as excellent agents for accounting! — [Page 635] Crow explained why our traditions have favored the genetic level of analysis (1994, p. 616):
The reason, I think, is that these pioneers [Fisher and other founders of the Synthesis] and their intellectual heirs have been concerned, not with selection as an end in itself, but with selection as a way of changing gene frequencies. Selection acts in many ways: it can be stabilizing; it can be diversifying; it can be directional; it can be between organelles; it can be between individuals; it can be between groups ... But the bottom line has always been how much selection changes allele frequencies and through these, how much it changes phenotypes. This suggests that we should judge the effectiveness of selection at different levels by its effects on gene frequencies.
I could not ask for a better statement of (unconscious) support for the position here maintained. Again, as I noted in several other quotations from gene selectionists, Crow allows that selection, as a causal force (“selection as an end in itself,” in his words), operates on interactors at several hierarchical levels of individuality, including groups. He also admits that changes in gene frequencies arise as a result of such selection. He then states — and again I don't object — that these alterations in allelic frequencies should be read as a “bottom line” in judgments about selection's effect. Nicely said, but a bottom line for what? Crow then gives his crucial answer — for keeping the books of evolutionary change: “we should judge the effectiveness of selection at different levels by its effects on gene frequencies.” Altered gene frequencies are therefore results (for bookkeeping), while selection (the cause of the changes) operates upon interactors “at different levels” of individuality.
This notion of a “bottom line” also provides our best entree into the third and most important reason for choosing genes as units of bookkeeping: the intrinsically asymmetrical nature of causal flow in hierarchies of inclusion. I particularly appreciate the doubly amplified utility of hierarchy theory in this example — for the hierarchical view, as I shall show, both serves as a replacement for gene selectionism, but also (in a situation not devoid of irony)* provides the rationale for why many biologists chose, albeit for fallacious reasons, to focus on genes in the first place!
We do need to keep the books of evolutionary change, and bookkeeping does require a basic unit of accounting. Candidates for this status must obey the primary criterion always stressed by gene selectionists: faithful replication. But genes do not exhaust the range of faithful replicators. Asexual organisms and species also rank as sufficiently faithful. Reductionistic preferences in general, and claims for relatively greater faithfulness of genic vs. higher-level replication, might set a preference for genes — but another crucial argument, usually unrecognized or unmentioned, seals the case. [Page 636]
Because bookkeeping is not the same enterprise as causality, and because we are not, in simply counting, trying to establish the causes of differential success, we want to make sure, above all, that we choose a unit better suited than any other to record all evolutionary changes, whatever their causal basis. No single unit of bookkeeping can monitor every conceivable change, but the gene becomes our unit of choice because the nature of hierarchies dictates that genes inevitably provide the most comprehensive record of changes at all levels. (Even so, gene records will miss certain kinds of changes that we generally call evolutionary. For example, as Wilson and Sober (1994) point out, assortative mating of organisms within a population may greatly increase the ratio of homozygotes to heterozygotes at many loci, but need not change gene frequencies in the population.)
Hierarchies are allometric, not fractal (see Gould and Lloyd, 1999), and various levels translate a common set of causes to strikingly different results and frequencies. Moreover, hierarchies are directional, and therefore not indifferent to the nature of the flow of influence. As the most important of all such asymmetries, change at a low level may or may not produce an effect at higher levels — “upward causation” in the standard terminology (see Campbell, 1974; Vrba, 1989). But change at a higher level must always sort the included units of all lower levels — by the analogous process of “downward causation.”
If a gene increases in copy number within a genome by duplication and lateral spread (gene selection in the genuine sense), phenotypes of organisms may or may not be affected. But selection on higher-level individuals always sorts the lower-level individuals included within. If ugly organisms out compete beautiful conspecifics, then genes for ugliness increase in the population. If stenotypic species prevail over eurytopes in species selection, then genes associated with stenotypy increase within the lineage. If species of polychaetes eliminate species of priapulids in competition over geological time, then polychaete genes increase in the marine biota.
Given this intrinsic asymmetry, what single unit would a good bookkeeper choose? Obviously not the organism, or any high-level individual, because we would then miss many changes at lower levels — and a good bookkeeper wants, as the chief desideratum of his profession, to record all changes. As noted above, low-level selection need not impose any effect upon higher levels at all. Equally obviously, our optimal bookkeeper will choose genes — not because genes are intrinsically more basic (the reductionist fallacy); not because genes are primary causal agents, or causal agents at all (the gene selectionist fallacy); and not because genes replicate faithfully (for other kinds of individuals do so as well); but, rather, because genes, as the lowest-level individuals in a hierarchy, manifest the unique property of recording all changes. Thus, the intrinsic nature of hierarchies sets our preference for genes as units of bookkeeping — for only genes act as nearly ubiquitous recorders of all evolutionary alterations, whatever their level or cause of occurrence.
Finally, we must note one other property that, while strongly favoring genes as units of bookkeeping, shows even more clearly why genes cannot be [Page 637] exclusive units of selection, or causal agents. Bookkeepers must, above all, be objectivists, not
sleuths or storytellers. A good bookkeeper wants an unimpeachable record, not a causal hypothesis (that can always be wrong). Books kept in terms of gene frequencies become the best objective records of “descent with modification” because they do not make causal attributions, but only count changes (“just the facts, ma'am,” to cite a famous detective from the early days of television). The hierarchical nature of evolutionary mechanics, and the simultaneous action of selection on individuals at several levels, implies our inability to know the causal basis of change from records of altered gene frequencies alone.
The Structure of Evolutionary Theory Page 101