Gardner refrains from using the term ‘ultimate causation’, and instead speaks of ‘ultimate explanations’, presumably because he recognizes that this use of ‘causation’ would be dubious. Can any problems with the proximate-ultimate dichotomy be resolved merely by changing terms? Dickins and Barton maintain that since contemporary evolutionists know what they mean by proximate and ultimate explanations, there is no problem. We, of course, agree that much good work is done within the evolutionary sciences by recognizing that separate answers to how and why questions are needed, but the same good work could be done, and often is, using alternative terminology, without any of the negative ramifications of Mayr’s framework. The primary problem with the proximate-ultimate dichotomy is not that it is ambiguous, but that it channels thinking, leading to the neglect of alternative hypotheses.
Gardner points to both a logical and a formal connection between function and a history of selection. We accept his argument, based on Grafen’s (2002) mathematical analysis, that natural selection leads organisms to act as if maximizing their inclusive fitness. However, where strong adaptationism is coupled with genetic determinism and unidirectional models of causation it becomes problematical. It would be a mistake to assume that, because natural selection of genetic variation leads to design, that all semblance of design arises solely from natural selection of genetic variation. We foresee many characters that exhibit design features but are not biological adaptations (e.g. spandrels, exaptations, products of cultural evolution, as well as the appearance of design brought about through niche construction). Gould and Lewontin (1979) pointed out over three decades ago that features (e.g. spandrels) can possess design yet not be biological adaptations, and in spite of the progress made in the intervening period recent reviews have concluded “even in the presence of both functional data and evidence for selection from DNA sequence data, it is still difficult to construct strong arguments in favor of adaptation” (Nielsen 2009, p. 2487), and “the ‘adaptive’ designation may be premature and may lead to incorrect conclusions about the relationships between gene function and fitness” (Barrett and Hoekstra 2011, p. 767). Contrary to Gardner’s claim, we do see value in adaptationist thinking within biology, for instance, as a useful vehicle for generating hypotheses; however, it becomes a problem if it leads researchers to fail to consider viable alternative hypotheses. For instance, we don’t disagree with Gardner’s defense of Fisher’s modeling, but our point about Fisher’s emphasis on adaptive genetic variation with small effects is that Fisher, like most mainstream evolutionary biology from the 1930s to the present, prematurely dismissed saltationist and developmental explanations for evolutionary change. Fisher did this entirely by assumption. For instance, he wrote of the “logical case for rejecting the assumption that the direction of evolutionary change is governed by the direction in which mutations are taking place” (Fisher 1930, p. 17, our italics). We likewise worry that, by wrapping together function and evolutionary history, the ultimate causation conception leads researchers to think about evolutionary causality in linear terms, and to neglect alternative explanations for the appearance of design. We will illustrate our concern that the proximate-ultimate dichotomy blinkers thinking with two examples, one Gardner’s and the other from Dickins and Barton’s writings.
Gardner expresses apprehension over the “tangled mess of causation” that he envisages follows from our emphasis on reciprocal causation. This graphic phrasing is very instructive. Naturally, we agree that there is infinite regress of interlinked causal influences for any current event—if so inclined, researchers could trace causation back in time all the way to the big bang—but the suggestion that it is “not conducive to successful biological science” is a little overdramatic. The reciprocal causation stance is perfectly operational. It merely places the onus on researchers to make sensible judgments as to how far to trace back causality for the problem in hand. Evolutionary biologists do this all the time (as do scientists in many other fields). If reciprocal causation were genuinely crippling then there would be no theory of sexual selection or coevolution. In practice, researchers merely require a different kind of formulation, such as coupled equations, to accurately describe the evolutionary dynamics. The fact that these bodies of theory are progressive establishes that cases with reciprocal causation are not inherently problematic to study.
It is nonetheless fascinating and instructive that Gardner should view reciprocal causation as a problematical “tangled mess”. This seems to betray a fear that if evolutionary biologists were to let go of their unidirectional, externalist model of causation then chaos would ensue. But there is nothing to fear here. The ‘reciprocal causation’ stance requires researchers to make a judgment about where it is sensible to view causation starting. However, we regard this as a virtue, since it leaves researchers open to considering the full range of possibilities. For instance, a bout of sexual selection could start with a mutation generating variation in the trait, a novel environment eliciting variation in preference, with a pre-existing sensory bias, and so forth. Sorting between these alternative accounts is an empirically tractable issue, and researchers consider all these possibilities precisely because it is clear that causation is reciprocal.
Compare this open stance with Gardner’s (e.g. West et al. 2011) critique of an account of the evolution of cooperation through social norms (e.g. Boyd and Richerson 1985; Fehr and Fischbacher 2003; Gintis 2003; Henrich 2004). Gardner’s theoretical position constrains him to view this body of work as muddling proximate and ultimate causation, as from his standpoint social norms (for instance, to reward cooperators or punish defectors) can only be proximate causes. However, those criticised regularly make the distinction between proximate and ultimate causes in their own work (see, for instance, Fehr and Fischbacher 2003). We do not think either of the camps, both consisting of highly respected evolutionary thinkers, are making a schoolboy error; rather, they are separated by different models of causation. Gardner’s resistance to conceiving of the social norms explanation as a case of reciprocal causation left him and his coauthors unable to consider (or too quick to dismiss) the possibility that in this instance causation starts with plasticity in human cooperators, which generates selective feedback at either cultural or genetic levels. West et al. (2011) assert that imitation, punishment, and the satisfaction derived from these, are proximate answers to the question of why people cooperate, and complain that:
This does not solve the ultimate problem, because it does not answer why evolution should have produced a psychology or nervous system that mechanistically encourages (rewards) such punishment.
But where researchers are thinking in terms of reciprocal causation, it may do. Society-based institutions to reward cooperators and punish norm-violators will generate (natural or cultural) selection for tendencies to cooperate, and may modify (natural or cultural) selection on the pre-existing tendencies to reward/punish, and the satisfaction so derived. Of course, one can ask how the society-based cooperative norms, and/or predispositions to punish norm violators, came into existence in the first place, and answers to these questions push back the causal account further. Plausibly, cooperative norms piggy back on the prior existence of a sense of fairness that evolved through reciprocity or group selection, or an ancient tendency to retribution (Richerson and Boyd 2005; Richerson and Henrich 2012). However, the prior existence of such tendencies cannot be regarded as a complete “ultimate” explanation, since without the social norms to stabilize cooperation it does not evolve in this instance. Previous work by cultural evolution researchers has established that culture takes human populations down evolutionary pathways not available to non-cultural species, either by creating conditions that elicit established mechanisms (e.g. kinship, reciprocity) or via mechanisms not seen in other taxa (e.g. Henrich 2009), to generate an evolved psychology (tribal instincts, docility, shame applied to norms) that is entirely different from what can evolve through genes alone (Boyd and Richerson 1985; Henrich 2009; Chudek and Henrich 2011). Fehr, Boyd, Richerson and Henrich (personal communications) have confirmed that our characterization of their thinking about causality (in essence, that this is an instance of reciprocal causation) was correct, and that Gardner and his colleagues had misconceived their position.
The second illustration of how the proximate-ultimate dichotomy makes communication difficult can be found in Dickins and Barton’s treatment of developmental bias. In our articles we discussed cases like the repeated rapid adaptation of sticklebacks to post-glacial lakes through the loss of a pelvic girdle (Chan et al. 2010) and rapid adaptation and range expansion of the house finch in North America (Badyaev 2009), arguing that these examples, alongside many others, illustrate how developmental processes construct evolutionary pathways. In the first case, fast phenotypic change results from the elevated mutability of a major regulatory control gene, which, in a single step, generates an adaptive phenotype, and does so repeatedly in isolated populations, which are then subject to parallel (rather than convergent) evolution. In the second example, developmental processes respond to environmental challenges to generate functional, directional and coordinated suites of morphological and behavioural traits, in both parent and offspring (the latter through parental effects), which expose genetic variation to strong selection. This is perhaps the best-documented example of the Baldwin effect, a clear case in which (to coin West-Eberhard’s evocative phrase) “genes are followers, not leaders, in evolution”. Yet Dickins and Barton dismiss such cases as merely “lineage differences in the amount of proximate modular architecture there is for an overall phenotype” and “differences in how sensitive to environmental cues those proximate mechanisms are.” They conclude: “the proposal that natural selection and developmental bias are different explanations is a misunderstanding”.
Let us consider the differences between these two explanations (more accurately characterized as natural selection with and without developmental bias) for such examples more closely. Table 1 compares some general features of a conventional account with a developmental plasticity/bias explanation. To us the differences between these two classes of explanation are striking. For instance, in the standard account evolutionary change begins with genetic mutation, which generates phenotypic differences that are subject to selection. Conversely in the developmental plasticity/bias account phenotypic variation can also result from differential environmental induction, with genetic change following. In the standard account, genetic mutations (and novel phenotypes) are random with respect to direction, rate and location, typically disadvantageous, and appear in a single individual. Conversely, in the developmental plasticity/bias account, genetic mutations can be non-random with respect to rate and location, whilst novel phenotypes can be directional, functional, and may appear in multiple individuals. In the standard account, mutations vary from one population to the next, and are typically multiple and small in effect, with strong convergent selection in similar environmental conditions required to bring about similar phenotypes. In the developmental plasticity/bias account, the same mutation may appear in isolated populations, often associated with a suit of coordinated phenotypic changes, and is subject to parallel (that is, identical rather than convergent) selection. Further differences are given in the table.
Table 1 Differences between a conventional account and a developmental plasticity or developmental bias account of isolated populations adjusting to novel environmental conditions
Laid out in this manner, we believe most readers would recognize that that there are real and substantive differences between the standard and developmental plasticity/bias explanations. How then can Dickins and Barton deny this? Perhaps Dickins and Barton are unfamiliar with some aspects of the literature. It is hard to imagine that anyone conversant with the massive current interest amongst the evo-devo community in developmental systems as a source of evolutionary innovation (Gilbert et al. 1996; Arthur 2004; Minelli and Fusco 2004; Muller and Newman 2005a, b; Gilbert 2006), would characterize the issue of “the starting point of the explanation” (i.e. the origin of evolutionary innovation) as “somewhat trivial”, as Dickins and Barton do. Yet we think there is something more here: specifically, that the proximate-ultimate causation dichotomy encourages a unidirectional view of biological causation, where development is viewed as the outcome rather than the cause of evolution. In this respect we believe Gardner’s and Dickins and Barton’s stances are representative of a broader community of researchers (see e.g. Oyama et al. 2001; Keller 2010 for discussion).
We (like many other developmentally minded evolutionists, e.g. West-Eberhard 2003) believe that resistance to these ideas derives in part from implicit models of causation that can channel thinking on these topics, leading to the neglect of potentially important explanations. For instance, in their recent review of phenotypic plasticity’s impacts on speciation, where extensive evidence that plasticity is evolutionarily consequential was presented, Pfennig et al. (2010, p. 459) nonetheless conclude that “recent reviews of speciation generally fail to discuss phenotypic plasticity, indicating that workers in this field do not recognize a significant role for plasticity in speciation”. It is now well documented that there is considerable variation in mutation rates (e.g. Hodgkinson and Eyre-Walker 2011), but we suspect that the implications of this for developmental bias are rarely appreciated. In the particular case of pelvic reduction in sticklebacks, most evolutionary biologists would probably automatically reach for a counter-explanation of pre-existing standing genetic variation for rapid parallel evolution (e.g. Barrett and Schluter 2008), and this may well be tenable. However, the important point here is not that an explanation in terms of the elevated mutability of a key developmental control gene must be correct, but rather that it must be considered. Valid alternative explanations should not be ruled out on fallacious logical grounds.