Dawkins writes in his introduction (1976, p. ix): “We are survival machines — robot vehicles blindly programmed to preserve the selfish molecules known as genes. This is a truth, which still fills me with astonishment. Though I have known it for years, I never seem to get fully used to it.” I can only regard this honest admission as a striking example of the triumph of false consistency over legitimate intuition.
Sieves, plurifiers, and the nature of selection: the rejection of
replication as a criterion of agency
The linkage of selective agency to faithful replication has been urged with such force and frequency that the argument now functions as a virtual mantra for many evolutionary biologists. But when we consider the character of natural selection as a causal process, we can only wonder why so many people confused a need for measuring the results of natural selection by counting the differential increase of some hereditary attribute (bookkeeping) with the mechanism that produces relative reproductive success (causality). Replicators cannot be equated with causal agents (unless they also happen to be interactors, for only interactors can be agents). Units of selection must be actors within the guts of the mechanism, not items in a calculus of results.
Genes struck many people as promising units for a twofold reason that does record something of vital evolutionary importance, but bears little relationship to the issue of selective agency. Persistence and replication do lie among the necessary (but not sufficient) criteria for calling any biological entity an evolutionary individual. Since evolution requires hereditary passage, and since genes transmit faithful copies of themselves, and also represent the smallest functional unit of physical continuity between generations of sexual organisms (the kind of individuals we know best for obvious parochial reasons), many biologists assumed that genes must therefore act as the basic (or even the only) unit of selection.
This interesting error arises from two common fallacies in human reasoning:
The confusion of necessary with sufficient conditions. We all agree that units of selection must be evolutionary individuals in Darwinian theory — and that status as an evolutionary individual depends upon a set of criteria discussed on pages 602–613. These criteria do include hereditary [Page 620] passage and sufficient persistence — the properties most strikingly exhibited by genes. But evolutionary individuals, to act as units of selection, must also display other properties that genes do not generally possess. In particular, a unit of selection must interact “directly ... as a cohesive whole with its environment in such a way that replication is differential” — to quote Hull's definition once again (1980, p. 318).
But in sexual organisms, and in other higher-level individuals, genes do not usually interact directly with the environment. Rather, they operate via the organisms that function as true agents in the “struggle for existence.” Organisms live, die, compete and reproduce; as a result, genes move differentially to the next generation.
Of course genes influence organisms; one might even say, metaphorically to be sure, that genes act as blueprints to build organisms. But such statements do not substantiate the critically necessary claim that, therefore, genes interact directly with the environment when organisms struggle for existence. The issue before us — the venerable problem of “emergence” — is largely philosophical and logical, and only partly empirical. Genes would interact directly only if organisms developed no emergent properties — that is, if genes built organisms in an entirely additive fashion, with no nonlinear interaction among genes at all. In such a situation, organisms would be passive repositories, and genes could be construed as units of selection — for anything done by organisms could then be causally reduced to the properties of individual genes.
This aspect of the question must be decided empirically. But the issue is also quite settled (and was never really controversial): organisms are stuffed full of emergent properties; our sense of organismic functionality and intentionality largely arises from our appreciation of these emergent features. Thus, since genes interact with the environment only indirectly through selection upon organisms, and since selection on organisms operates largely upon emergent characters, genes cannot be units of selection when they function in their customary manner as faithful and differential replicators in the process of ordinary natural selection among organisms. Dawkins's metaphors of selfish genes and manipulated organisms may be colorful, but such images are also fatefully misleading because Dawkins has reversed nature's causality: organisms are active units of selection; genes, while lending a helping hand as architects, remain stuck within these genuine units.
The theory-bound nature of concepts and definitions. We are drawn to the faithfulness of gene replication, especially when compared with the contrasting transiency of sexual organisms, who must disaggregate to reach the next generation. We might therefore assume that genes become primary candidates for units of selection as a consequence of their potential immortality, while organisms fall from further consideration by the brevity of their coherent lives.
“Sufficient stability” surely ranks as an important criterion for the “evolutionary individuality” required of a “unit of selection.” But, in Darwinian theory and the search for units of selection, “sufficient” stability can only be defined as enough coherence to participate as an unchanged individual in the causal process of [Page 621] struggle for differential reproductive success. To be causal units under this criterion, organisms need only persist for the single generation of their lifetimes — as they do. This endurance may not strike us as a long time in some intuitively appealing psychological sense, or relative to the persistence of faithful gene replicates, or considered in comparison with geological scales — but these temporal frameworks are irrelevant to the question and theory at hand. Organisms last long enough to act as units of selection in a Darwinian process; they therefore possess the “sufficient stability” required of evolutionary individuals.
Of course, evolutionary individuals must all be able to pass — differentially and in a heritable manner — their favorable properties into future generations. But no aspect of this requirement implies or requires that units of selection must pass copies of themselves, bodily and in their entirety, into the next generation. The criterion of heredity only demands that units of selection be able to bias the genetic makeup of the next generation towards features that secured the differential reproductive success of parental individuals. Units of selection only need to plurify their own representation in the next generation; they need not copy themselves. Sexual organisms happen to plurify by disaggregation and subsequent differential passage of genes and chromosomes. Other kinds of individuals, including genes, asexual organisms and species, plurify more coherently. This common confusion of plurifaction with faithful replication has erected a serious stumbling block to proper understanding of the hierarchical theory of selection.
We can best clarify this crucial issue of the relationship between selective agency and criteria of faithful replication vs. plurifaction if we drop, for a moment, the conventional framework of replication vs. interaction, and return instead to a different metaphor commonly invoked during 19th century debates about the nature of Darwinism and natural selection — namely sieves.
We may use the classical metaphor of sieving to illustrate the inappropriateness of faithful replication as a criterion for defining units of selection. The “goal” of a unit of selection is not unitary persistence (faithful replication) — and I can't quite figure out why so many late 20th century Darwinians ever tried to formulate the concept in this manner. The “goal” of a unit of selection is concentration by plurifaction — that is, the differential passage of “youness” into the next generation, an increase in relative representation of your heritable attributes (whether you pass yourself on as a whole, or in disaggregated form, into the future of your lineage).
In the favored metaphor of Darwin's day, selection works like a si
eve laden with all the individuals of one generation. Surrounding environments shake the sieve, and particles of a certain size become concentrated, while others pass through the webbing (lost by selection). Sieving represents the causal act of selection — the interaction of the environment (shaking the sieve) with varying individuals of a population (particles on the sieve). As a result of this interaction, some individuals live (remain on the sieve), while others die (pass through the sieve) — and survival depends causally upon variation in emergent properties of the particles (in this simplest case, large particles remain, and small particles pass through to oblivion). [Page 622]
The surviving particles need to reproduce in genealogical systems of evolutionary individuals. They may do so by fissioning (faithful passage) or by disaggregation and reconstitution of new individuals as mixtures of hereditary parts of previous individuals. The individuals of the old generation eventually die and evaporate. The individuals of the new generation now live on the sieve, waiting for the next shake.
But this specification of the varied modes for constituting new individuals does not represent what we mean by selection. An entity must be able to reproduce to be defined as an evolutionary individual, but this entity need not replicate faithful copies of itself. Rather, it needs to be able to plurify — that is, to increase, relative to other individuals, the representation of its hereditary contribution to the next generation. Integral “you” may be disaggregated in the process, but so long as the next generation contains a relative increase in your contributions, and so long as you operated as an active causal agent of the Darwinian struggle while you lived, then you qualify as a unit of selection (and a winning unit in this case).
An interesting episode in the history of Darwinism clarifies this concept in a striking manner. We all know that Darwin accepted the idea of “blending inheritance,” or the averaging of parental characteristics in the offspring of sexual reproduction. Now blending inheritance marks an ultimate denial of half in breeding degrades faithful replication — for the hereditary basis of any selected character with an average individual. A paradox therefore arises. If units of selection must be faithful replicators, and if Darwin both understood natural selection and believed in blending inheritance, then why did he ever imagine that selection could work as a mechanism?
We can only resolve this conundrum by recognizing that faithful replication is not — and never was — the defining characteristic (or even a necessary property) of a unit of selection. Darwin, even given his belief in blending inheritance, could view sexual organisms as primary units of selection because he understood agency in a different way that remains valid today: units of selection are evolutionary individuals that interact with the environment and plurify as a causal result. We may return to the metaphor of sieving. Natural selection can work under blending inheritance because shaking the sieve favors the possessors of advantageous traits in each generation — for any individual with a phenotype biased in the favored direction gains a better chance of remaining on the sieve. The offspring of the most favored individuals will blend substantially back to the mean, but this style of inheritance only slows the process of selection — for, as a result of differential survival and reproduction in each generation, the mean itself still gradually moves in the favored direction.
Interaction as the proper criterion for identifying units of selection
The aforementioned arguments about sieves, plurifaction, and the inappropriateness of faithful replication for designating units of selection lead to a simple conclusion: we can only understand the causal nature of selection when we recognize that units of selection must be defined as interactors, not as replicators. Hull's distinction has great merit, but he fell into an overgenerous [Page 623] pluralism in arguing that identification of causal agency must include statements about both the faithfulness of replicators and the potency of inter-actors. Individuals need not replicate themselves faithfully to be units of selection. Rather, they must contribute to the next generation by hereditary passage, and they must plurify their contributions relative to those of other individuals. But the contributions themselves can be wholes or parts; faithful replicates or disaggregated bits of functional heredity. Selection demands plurifaction, not faithful replication.
The simple observation of plurifaction — the relative increase of an individual's representation in the heredity of subsequent generations — does not suffice to identify the operation of natural selection, for plurifaction can occur by nonselective means, and phenotypes can increase in frequency but then be unable to plurify. Consider the primary example of each phenomenon. First, individuals may plurify by accidents of genetic drift. Suppose that individuals fall through the sieve of selection at random, but survivors show increased frequency of certain heritable traits by accident. These surviving individuals will plurify, but they have not operated as active units of selection. Second, individuals may increase in frequency for phenotypic reasons unrelated to heredity. Suppose that large individuals remain differentially on the sieve, but that individuals grow larger than average for purely ecophenotypic reasons uncorrelated with any aspect of heredity that can pass to subsequent generations. Large phenotypes have increased in frequency for causal reasons — but they will not be able to plurify because they cannot bias the heredity of subsequent generations.
So selection demands plurifaction because evolutionary individuals must maintain lineages by hereditary passage, and selection occurs by increase in relative representation. But plurifaction can only represent a necessary condition, not a cause. We define selection as occurring when plurifaction results from a causal interaction between traits of an evolutionary individual (a unit of selection) and the environment in a manner that enhances the differential reproductive success of the individual. Thus, and finally, units of selection must, above all, be interactors. Selection is a causal process, not a calculus of results — and the causality of selection resides in interaction between evolutionary individuals and surrounding environments. The study and documentation of group and higher-level selection has been stymied and thrown into disfavor by our confusion over these issues — and especially by the blind alley of a logically false argument that identified replicators rather than interactors as units of selection, and then constructed a fallacious, reductionistic theory, precisely opposite in structure to the hierarchical model, by specifying genes (because they replicate faithfully) as ultimate or exclusive units of selection. In this context, I note with delight that group selection has risen from the ashes to receive a vigorous rehearing (Sober and Wilson, 1998, for a full treatment; Lewin, 1996, for a popular account under the title “Evolution's new heretics”; and Gould and Lloyd, 1999, for resolution of a final logical problem). This potent revival rests upon two proposals that, as centerpieces of this book, could not gain my stronger assent: the identification of evolutionary [Page 624] individuals as interactors, causal agents, and units of selection; and the validation of a hierarchical theory of natural selection based upon a principled understanding that evolutionary individuals exist at several levels of organization — including genes, cell lineages, organisms, demes, species, and clades.
D. S. Wilson has most vigorously championed this revival (Wilson, 1980, 1983), while his collaboration with philosopher E. Sober has produced a particularly important paper and a subsequent book on the subject (Wilson and Sober, 1994, with 33 accompanying commentaries and the authors' response; Sober and Wilson, 1998). Wilson and Sober anchor their argument by insisting that units of selection must be defined as interactors, not replicators.
I must raise only one mild quarrel with Wilson and Sober. I agree entirely that units of selection must be denned as interactors, but I prefer a “looser” or “broader” concept of interaction that fosters the proper identification of highest-level individuals in species and clade selection. Wilson and Sober stress the “organism-like” properties of interactors, and therefore make the confus
ing and regrettable linguistic decision to use “individual” for conventional bodies, and “organism” as the general name for a unit of selection at any hierarchical level; whereas I and most biologists (see Gould and Lloyd, 1999) advocate a reversed terminology. In characterizing the evolutionary principle of interaction, I would stress the potential for rich panoply of emergent fitnesses, and for the consequent capacity of plurifaction.
Their chosen stress on “organism-like” properties leads Wilson and Sober to emphasize direct modes of interaction based on actual contact of sympatric individuals — the old vision of two gladiators duking it out to the finish. But interaction does not require physical contact. Interaction occurs between individuals and environments, not necessarily between individual and individual. The interaction must be able to yield plurifaction for causal reasons based on properties that enhance differential reproductive success — but, again, competing individuals need not interact directly with each other. Rather, to speak of selection, competing individuals only need to plurify at different relative rates based on similar causal interactions with environments. But the environments may be spatially separate and broadly defined. This issue does not often arise at the traditional level of Darwin's chosen evolutionary individuals — that is, organisms. But higher-level individuals, particularly species and clades, do often compete without contact — and our notion of units of selection must include this important mode of interaction.
Several thoughtful biologists have stressed this point, and I have compiled a small file of such statements. I shall present here only the forceful argument of Williams (1992, p. 25), who has changed his view substantially since formulating the theory of gene selectionism in 1966:
The Structure of Evolutionary Theory Page 99