Abstract

This paper seeks to clarify and modestly defend the use of adaptive explanationsexplaining the existence of a trait by reference to putative historical selection pressureswithin the evolutionary behavioral sciences. I identify four major areas of concern: (1) the kind of target that behavior is, (2) the explanatory strategy used by adaptive explanations, (3) the forward-looking or model-based approach to adaptive explanations, and (4) the adaptive explanation of human behaviors. After working through each of these areas, I conclude that, adaptive explanation, even via the forward-looking approach, is a viable strategy to explain the behavior of human and non-human animals, but is subject to legitimate difficulties and limitations concerning hypothesized selection pressures, cognitive complexity, and non-genetic systems of inheritance. comments. I would like to especially thank committee member Jeffrey Schwartz and committee member and research adviser Sandra Mitchell who in their many roles as professor, advisor, mentor, and exemplar had a transformative effect on not only this paper, but also my entire undergraduate experience. Dr. Mitchell's ability to be endlessly bombarded with doomed arguments by an over-enthusiastic undergrad and yet remain helpful and encouraging all the way to the completed thesis says more about her than any list of adjectives I could fit in this preface. Before beginning the thesis proper, I would like to make a few clarifications. As a reader helpfully pointed out to me, "adaptive explanation" is casually thrown around in the philosophy of biology, but it is actually quite a laden word. To head off any confusion, when I say adaptive explanation I am referring to the explanation of the current existence of a biological trait by appeal to the historical effect of natural selection on the genotypes of a vi population. This will be discussed in much more detail, but it may be helpful to start with such a platform. However, this entails that those who challenge either the viability or explanatory power of natural selection may object to the meaningfulness of this paper, even if they acknowledge the argumentation. This paper is not for them. Rather, if this were not a mere undergraduate thesis fated to obscurity, I would hope to engage with two distinct groups. For scientists, this paper could serve as a somewhat casual overview, with the interesting feature that it exposes the supporting philosophical skeleton of practices they are likely quite familiar with in the flesh. For philosophers, who are a bit more agnostic on the topic, I have tried to toe the line between preserving their reservations, and defending what I find to be an interesting and fruitful area of biology. vii INTRODUCTION Adaptive explanation refers to the explanation of a trait by appeal to the putative forces of selection that made it, e.g. the antlers of elk evolved in response to the selection pressure of male-male competition for mates. In the 1930s, with the advent of ethology -the evolutionary study of behavior -the use of adaptive explanations was broadened from morphological to behavioral traits. 1 Despite their current broad use across the evolutionary behavioral sciences, they remain controversial, particularly when applied to human behaviors. Here, I aim to provide clarification and measured defense. This paper is structured as series of objections to the adaptive explanation of behavior, concerning: 1 represents and the way the behavioral sciences have responded. (3) and (4) are particularly contentious, and therefore will form the bulk of my analysis. BEHAVIOR AS A SCIENTIFIC OBJECT I begin by briefly sketching one kind of appropriate scientific object. I then discuss behavior with respect to this account, identifying both its unproblematic and problematic features. I conclude by defending the methodological strategies embraced by the evolutionary behavioral sciences as responsive to these challenges, and commensurately holding behavior to be an appropriate scientific object. APPROPRIATE OBJECTS OF SCIENCE Obviously a full theory of what it takes to be an appropriate object for scientific study is well beyond the scope of this (and perhaps any) paper. Therefore, rather than trying to delimit a conceptual space that is occupied by appropriate objects of science and evaluating whether or not behavior fits within that space, I will instead begin with an uncontroversial object of science, and see if behavior shares the features that seem to make that object so scientifically unproblematic. In short, my approach is expansive rather than constrictive. Let us begin by looking to two systems: (1) a ball being rolled down a hill, and (2) a gazelle being approached by a lion. Before proceeding further, it is worth clarifying why the second system constitutes a behaving system. The most obvious difference is that the 2 second system is living or biological -ethologists do not see non-living systems as behaving. However, this is insufficient, after all, excepting the shape, a gazelle being rolled down a hill could be analyzed in the same way as the ball, but one would not consider that the behavior of a gazelle. To be behavior, the gazelle's actions must also be internally mediated, e.g. by neurotransmission, chemical signaling, or muscular activation, as opposed to merely the replication of interactions that occur in non-biological systems (such as gravitational acceleration) in systems that are incidentally biological. Now lets us compare the systems specifically with respect to being appropriate objects of science. Both systems are physical and observable (i.e. they make no appeals to undetectable or supernatural forces). However, the ball-hill system is also (fairly) noncontingent, accessible to experiment, predictable, and able to be specified with the precision needed for scientific study (individuated). The ball-hill system operates with the law-like regularity of physics. An identical system can be recreated simply by putting the ball back on top of the hill. I can easily change just one thing involved in the system, e.g. the coefficient of friction, and see a direct and isolatable change in outcome. If I cover the hill in glue and observe the ball rolling more slowly, I do not have to worry about the possibility that maybe the ball was simply rolling more slowly because I was standing too close and making it skittish. And in fact, given the current knowledge of physics, I can predict the results of almost any change I make on the system, from changing the coefficient of friction to increasing gravity. Finally, the system is easily specified and precisely measurable, starting and stopping under exact parameters. I can describe my setup, e.g. I put a 5kg bowling ball on top of a 30-degree grass-covered hill 3 meters in height, and scientists can then observe an identical system. While balls on hills may not be the most exciting, they are extremely cooperative. 3 A gazelle being approached by a lion is not so tractable. If the gazelle sees the lion it may run, but it may not see the lion, or it may jump in the air (stot) first -this makes the system both more contingent and less predictable. Nor can I intervene on the system by changing many things. Finally, the gazelle's actions exist as a stream of information. Sure, it may run from the lion, but then it eats some grass, and then it licks itself, etc., until it dies. Which stretch of actions should I describe such that a scientist will be able to identify (approximately) the same stretch in another gazelle? And how should I describe or quantify them? Given this apparent divergence from the ball-hill system, there are two broad strategies. First, one could study behavior in a way that makes it as much like the ball-hill system (i.e. physics) as possible. Second, one could defend a particular approach to behavior that, while relevantly different from that used by physicists, nonetheless is scientific. 2 The first strategy is certainly possible. Scientists could approach the actions of the gazelle as applied physiology: this brain region was activated, these neurotransmitters were released, these muscles twitched, etc. In fact many scientists, e.g. physiologists, neurologists, look to exactly these factors. However, to accept only this strategy is problematic. It does not address the fact that behaviors seem to do something for organisms, a sort of apparent goal-directedness, which is a particularly interesting and salient feature of behavior. Moreover, it is this feature of behavior that the evolutionary behavioral sciences and adaptive explanations utilize, e.g. 4 wasps stinging spiders in order to lay their eggs in the them, rather than appealing to the mechanism of the sting. In light of this, I will undertake a defense of the second strategy. Particularly, I will argue that behavior exhibits the same features that make the ball-hill system a good object of science, but either exhibits them to a different (but still sufficient) degree, or exhibits them in a different manner. BEHAVIOR ON ITS OWN TERMS I will structure this defense of a behavior level science based on the features of an appropriate scientific object described above: non-contingent, accessible to experiment, predictable, and able to be specified with the precision needed for scientific study (able to be clearly individuated). In the second section I will go into functional approaches in more detail; however, it is important to my defense here that behaviors can be thought of as doing certain things or performing certain functions for organisms. This statement seems to commit to a 5 research is directed influence the framework used to approach and explore behaviors Behavior as contingent Like other biological features, the existence of a behavior is contingent upon evolutionary history, development, and experience. Moreover, whether or not a behavior is performed is contingent, dependent upon both internal factors and environmental conditions. Finally, even when performed, the manner of expression is contingent, likewise dependent upon both environmental factors (particular environmental demands) and internal factors. All this adds up to meaning that simply observing a behavior does not mean it occurs predictably or is easily generalizable. To get more traction on this problem, let us imagine that behaviors, instead, operate in a more lawful manner. Upon a lion getting within a certain distance of a gazelle, the gazelle will begin to move away and the lion will pursue. This happens regardless of whether the lion is spotted, the gazelle is hungry, or an experimenter is present -this is simply how gazelles and lions interact in this universe. Scientists can derive the behavior of new a gazelle from this law-like interaction. If something unusual happens, e.g. the lion 6 turns away, scientists can look at this with respect to the lion-gazelle law to identify a cause, e.g. a Wildebeest intervened. Unfortunately, evolutionary behavioral scientists do not have it so good. There is the possibility of a superficially similar situation being very different, e.g. gazelle A running from approaching lion A, but gazelle B not running from approaching lion B. Moreover, there may be no easy way to identify the cause of the difference -maybe gazelle B did not see the lion, or maybe it had a small stroke. However, simply the fact that the evolutionary behavioral sciences deal with these contingent objects is not grounds for dismissal -every science must do so to some extent or the other (see Mitchell, 2009). Two specific questions come to the forefront. First, are behaviors necessarily so contingent that they can never be approached scientifically? Second, if not, can one evaluate the extent to which a behavior is contingent? The answer to this first question seems to be a definite no. This would entail that every behavior is a sort of miracle contingent upon situations so elaborate that they are never relevantly repeated. Certainly some behaviors may fit this. A starling, a bird noted for its abilities of mimicry, may somehow end up copying a once in a lifetime performance by Yo-Yo Ma. This would seem to tell us little about the normal behavior of starlings. 4 However, most animals definitely perform what prima facie appears to be the same behavior a sufficient number of times -look to stotting in gazelles, collecting acorns in squirrels, wasps stinging spiders, etc. In terms of evaluating the contingency of behaviors (i.e. the obscurity of the combination of causes upon which behaviors are dependent), evolutionary behavioral scientists have at least three strategies available to them: fixed-action patterns, 4 To clarify, this would tell us little about the behaviors of starlings; however, it would tell us a great deal about the capacity of Starlings. Similarly, even a single instance of tool use or manufacture could be very powerful, even though the information it provides is not about the behavior of a species per se. 7 comparative methodology, and modeling. A fixed action pattern --easily recognizable and often repeated complexes of actions -picks out a specific behavior from the stream of information that behavior represents Behaviors are susceptible to this approach, and looking at behaviors across environments and species (or even across many member of the same species) can lead to meaningful inferences about the contingency of behaviors. Finding similar behaviors across a wide variety of taxa and environments e.g. mobbing, or mating dances, provides evidence they are contingent upon a less obscure set of causes. 5 In contrast, behaviors found only in a small number of members of one species are likely to be extremely contingent. A third option discussed by Trestman (2011) is hypothetical models of behavior to which natural behaving systems can be matched. For example, given an understanding of optimal foraging behavior, one can model how long shorebirds will search for food in a specific patch and when they will move to a new one -which can then be matched to the birds actual behavior demonstrating a functional character to the bird's behavior 5 See Trestman (2011) for a much more thorough analysis of the comparative or "Lorenzian" approach, albeit the context for his analysis is somewhat different than the "dealing with contingency" problem I am tackling here. Also see Accessible to experiment Behaving systems are simply not manipulatable to the degree of balls on hills. The biological nature of the components makes everything messier, and it is very hard to intervene on specific components. However, again, this is far from a death sentence. First, behaving systems can still be experimented upon. Second, there are other ways to get the same kind of information that would be provided by experiments (or other relevant kinds of information). In addition to simply observing, evolutionary behavioral scientists do engage in experiments. Performing experiments involves toeing a very difficult line, as those in the ethological tradition would like to control as many causal factors as possible, but not control so many factors that behaviors performed are not representative of how the organism performs in its natural environment (see Lorenz, 1982, pp. 47-52, 64). This difficulty can be somewhat overcome through possessing a broad understanding of the organism and the behaviors it performs (ibid. pp.52-53). On this background of a general knowledge of the organism, if experimental intervention leads to aberration, the experimenter is far more likely to identify it. Alternatively, experiments can also be performed in the wild. To assess whether the mane of lions impacted mating, West put one of two dummy (plush) lions that differed only in mane length by a female lion (2005). He can evaluate whether this experiment is useful by looking to both whether behavior induced by one dummy differs with respect to behavior induced by the other, and whether behavior induced by the dummies differs qualitatively from behavior that occurs in response to real male lions. 9 It is also well established that scientists use non-experimental approaches. Scientists in the ethological tradition have a rich observational literature that highlights many correlations between ecology (or situation) and behavior. And again, the comparative and model-based approaches do work here. The comparative method sets up a sort of natural experiment, in which one can attempt to identify relevant causal factors by looking at an array of different situations. In Shear's research on web-building in spiders, he looks at a connection between particular features of the environment and the structure of the web being built (1994). The finding that there is a correlation between web design and ecology independent of the species of spider making the web evidences that webdesign is a reaction to the environment. Predictable The complexity and large number of unknown factors, e.g. physiological condition, life history, etc., operating in behaving systems injects a certain sense of unpredictability, and make them difficult to deal with in a causal-mechanical sense. To predict what a gazelle will do when confronted by a lion purely by monitoring its physiology would be extremely complicated. However, predictability can be regained through the application of an intentional or functional approach (see especially Able to be individuated Unlike other physical systems, behavior represents a continuous stream of information and therefore must be able to be individuated into usable units (see For behaviors, function seems to be the most powerful as it allows for both specificity and generalizability. Let us take a specific behaving system, say a zebra. I can ask which behavior functions to evade predators and arrive at the answer of running. This both picks out a particular action, and differentiates running that functions to evade predators from other instances of running. 6 Now take a skunk. I can ask the same question -which behavior functions to evade predators -and arrive at another distinct trait for the skunk: spraying. "Predator evasion" serves as a general category that can be used to pick out specific behaviors in any behaving system that is threatened by predation. Morphology can be used to individuate behavior in two senses. First, a behavior can correspond to underlying cognitive and neurological architecture. Second, and in a less reductive sense, behaviors can be more or less like morphological features. Some behaviors, while individuatable, are extremely plastic and contingent. Take for instance the use of spears to hunt in particular populations of Chimpanzees (Pruetz & Bertaloni, 2007). 6 The relationship between functional accounts and trait (or token) individuation is far more complicated than I am able to discuss here. I refer the reader to 11 Other behaviors, such as the stinging of scorpions, are an extremely stable and invariable complex of actions. These invariable behaviors, as captured by fixed action patterns, individuate behavior in the same way as morphology. They provide a stable and specific structural description. As far as I am aware, homology does not serve to individuate behaviors, but I leave open the possibility. 7 SUMMARY Behavior, and more specifically behavior as understood by the ethological tradition, does indeed possess features that make it more difficult to approach scientifically. However, there is nothing about behavior that renders it scientifically intractable. I specifically argued that behavior is naturalistic and observable, and, with the right approaches, also manageably contingent, accessible to experiment, predictable, and able to be specified with the precision needed for scientific study (individuated). Although, I am uncertain as to how homology can serve to individuate at all. Homology is nothing more than similarity combined with a theory of relatedness. Even if one had very good reason to believe two organisms were phylogenetically related, they would not identify two organs as homologous without some morphological (including genetic or developmental) correspondence. In short, the individuation work of homology seems to be done morphologically. However, I do not yet have enough confidence in this argument to deny the possibility that homology may be used for individuation. 12 EXPLAINING BEHAVIORS IN THE EVOLUTIONARY BEHAVIORAL SCIENCES I have defended that the targets of adaptive explanation, i.e. behavior, are appropriate targets of scientific inquiry. I will now address whether the techniques used to explain behaviors are appropriate. I do so in three levels, addressing first "functional analysis," next "selected-for explanation," and lastly "adaptive explanation." While functional analysis identifies the role a behavior plays in its containing system, selected-for explanations correspond to the selective (evolutionary) history of trait, and adaptive explanations likewise appeal to selective history, but specifically for genetically mediated natural selection. I conclude with a discussion on how scientists can move from functional analysis to adaptive explanation. To frame this section, I begin with an overview of functional accounts within biology generally, and the evolutionary behavioral sciences specifically. OVERVIEW There are two main understandings of function within the philosophy of science. First is 13 Cummin's causal-role account, which is elucidated in a more biologically relevant way by Another account of function has been recently introduced by Nanay (2010, also see 2013). His account is quite elaborate and is allegedly differentiated from existing accounts by (1) modal force (how "would" a function contribute), (2) a focus on trait tokens rather than trait types, and (3) the use of counterfactuals. He is mistaken about (1) as causal-role can be interpreted as a disposition or propensity to contribute to capacity of a containing system if certain conditions would be met, (Nanay incorrectly identifies dispositional as strictly "future" accounts. Also see Kiritani, 2011 for a defense of modal force in etiological accounts), but this is beside the point. What matters here is, if Nanay were to be correct, would it change my project. The answer seems to be no. Nanay explicitly identifies his account with usefulness -which maps onto the causal-role a trait plays in the capacity of a containing system. Nanay prevents usefulness from collapsing into use through the use of modal force, which is exactly what disposition does for the causal role account. Finally, Nanay prevents his account from simply picking out every way a trait could be useful by contextualizing it with respect to the explanatory project. For instance, one could look to how a trait is useful for increasing survival and reproduction. Causal-role accounts make this same move by specifying the capacity that functions are contributing to, e.g. contributes to the capacity to survive and reproduce. His machinery for identifying function is a little different, but it identifies the same functions as the Cummins-style causal-role account. Or, to put it another way, Nanay's account functions identically to the causal-role account. Therefore I will simply speak of the more established causal-role account with the addendum that what I say about it also applies with Nanay's account. Given this symmetry with the causal-role, it is interesting that Nanay puts forth his account largely as a challenge to the etiological account, which as we will see, aims for a completely different kind of explanation (see 14 Causal-role accounts of function aim to explain function through the contribution a feature makes to the capacity of a containing system. Eyelashes would have the function of keeping particulate out of the eye, if possessing eyelashes contributed to an organism's capacity to keep particulate out of its eyes. This can be put in more evolutionarily relevant terms by connecting the function of a feature to an organism's inclusive fitness (again see 9 As should be clear, these two accounts of function, while often overlapping, have different explanatory projects (also see 15 work on the convergent evolution of similar web-design for similar ecologies in spiders is illustrative (1994). This requires the use of etiological accounts (selected-for function), and is discussed by Tinbergen as the study of evolution FUNCTIONAL ANALYSIS (AND BIOLOGICAL FUNCTION) Functional analysis is associated with causal-role function, and very roughly refers to the decomposing of a system into component parts that contribute to a capacity of the containing system (see We can understand biological function as picking out certain features of a behaving system. Specifically, if we take an organism and ask which behaviors are biologically functional, it should pick out every behavior that has a propensity to increase inclusive fitness with respect to a particular environment, independent of whether there has been selection for that behavior. Usually this environment would be its current environment for reasons of ecological authenticity, but one could apply biological functionality with respect 16 to any environment. Behaviors picked out by biological function, for which there was not selection for, are understandable as exaptations or the products of broader cognitive mechanisms such as learning. For instance, begging for food in domestic dogs is perhaps biologically functional, but presumably dogs learn to beg rather than having been selected for begging. In fact, every behavior that a dog gets rewarded for with food, e.g. tricks, could perhaps be considered biologically functional depending on context. Given that biological function will pick out all behaviors that increase inclusive fitness whether they are selected for (as long as they are still functional), incidentally useful, or the product of a broader mechanism, it is reasonable to maintain that it picks out more behaviors than the etiological account. How to engage in functional analysis for biological functionality is a non-trivial problem (see especially One starts with biological functionality, of which survival contributes, within survival eating is identified as contributory, within eating catching fish, and within catching fish, looking for fish, diving for fish, and grabbing fish with talons. A scientist can conclude that looking for fish functions for the catching of fish, rather than, say, predator evasion, based on their total understanding of the osprey. 10 And scientists often do apply biological functionality with respect to environments other than current ones to gain traction on species that either have undergone environmental shift (e.g. humans or zebra mussels) or will undergo environmental shift (e.g. the introduction of non-local crops). 17 Hypotheses of biological function are prior to hypotheses of selected-for function due to a circularity concern (see Returning to the difference between etiological accounts and causal-role accounts, not only do they have different aims, but they also have subtly different targets. The etiological account targets physical objects. It seeks to explain why there is a certain structure or action, e.g. why there are eyelashes, and cannot target things that do not exist. In contrast, the causal-role account targets properties or roles. It seeks to explain the role of a behavior with respect to the capacities of its containing system and can engage with counterfactuals, e.g. how being able to breath fire would impact the survival of canaries. 11 Given this, the notion of biological function cannot explain the current physical presence of a behavior within a system, even though it may explain (or arguably describe) the function of a behavior. In a strict sense, accounts of behavior that do nothing more than establish that a behavior, e.g. running from hungry lions, increases inclusive fitness, are explaining a function not a behavior. 12 Scientists must necessarily wed biological functionality to an 11 Another way to get at this is through the understanding of etiological accounts as backwards-looking (i.e. tied to actual history) and causal-role accounts as forward-looking (i.e. engaged with potential) (see 12 Two caveats. First, simply identifying a biological function is tremendously useful as it lays the groundwork for an etiological explanation or helps to predict future evolutionary trends. Second, many scientists, for better or worse, likely assume an etiological account in the background when they identify a biological function (see 18 etiological account (or some other kind of causal process account, e.g. a cognitive mechanism)) in order to explain the behavior as a physical object. Selection-for explanations provide a way for scientists to address this difficulty. SELECTION-FOR EXPLANATION A selection-for explanation is an etiological account as applied to a behaving system and aims to explain the existence of a behavior. It is the second step in the development of adaptive explanation. I contend that a selection-for explanation is composed of (1) a present or historical biological function, (2) a selection background (see especially 13 Nonetheless, I discuss all four in turn. The key move of an etiological explanation is that it makes the function of a behavior the reason for that behaviors existence; within evolving systems this entails that at some point the behavior was selected-for because of its function, hence my use of the term selection-for explanation. Specifically a behavior would have been selected for because of its biological function, i.e. contribution to inclusive fitness (also see 19 selected-for function (e.g. exaptations 14 , such as the success of the domestic pigeons nesting behaviors in urban environments Selection-for explanations necessarily assume a selection background (see 15 This is why the foraging behavior of squirrels is a reasonable target of selection-for explanations, but spell-check in Microsoft Word is not. At this level, it would be extremely unlikely any behavior of interest would not have the appropriate selection background. Note that Mitchell, drawing from Sober, 1984, further identifies selection of the target for its function as part of the selection background (ibid.). This is correct, and leads to a much richer account. However, scientists use quasi-independence to get at this aspect of a trait's background, and therefore I will stick with the broad understanding of selection background. Selection-for explanations also invoke the operation of specific selection pressures. As discussed, biological function does not exist as some kind of intrinsic property, but only with respect to particular environments. This applies even to fundamental features such as 14 Exaptation is a problematic concept for etiological accounts of function. As 15 For in-depth discussion of the kinds of populations that can support selective processes see Godfrey-Smith, 2009. 20 reproduction: having many offspring may be extremely biologically functional in a nutrient-dense jungle, but counterproductive in a nutrient-poor desert, where limited resources must be stretched across these many offspring leading to low survivorship. These environmental features 16 can be understood as selection pressures for which specific behaviors are functional in response (i.e. more functional than other available options). Therefore, when a scientist makes a selection-for explanation, it involves the identification of both a function and, at least implicitly, the corresponding selection pressure. Quasi-independence (Lewontin, 1978) accounts for the second element Mitchell identifies in the background of etiological explanations, specifically selection of the target for its function (1989, 1995). to think about this is that the feature must be a heritable evolutionary unit. This is usually applied to genetic features, but cultural features work just as well. For instance, if a cultural anthropologist demonstrated that a belief in Shiva increases fitness, they would have to 16 By environmental features I refer not just to the physical environment, but also to inter and intraspecific interactions. 17 Plasticity can be accounted for unproblematically in this view, as the relevant selection pressures can represents a range of environmental conditions (or even something such as uncertainty itself). 18 This language makes use of Sober's distinction between selection for and selection of (1984). To clarify, selection for relates to what the selection pressure is acting on and selection of relates to what is actually being physically selected. An illustrative example is Belyaev's farm fox experiment. In Belyaev's foxes, selection pressure was only applied to defensiveness around humans, but there was selection of a large number of behavioral and physiological changes such as attention seeking and floppy ears (see 21 demonstrate that a belief in Shiva is sufficiently independent from a belief in Hinduism to argue that the fitness gains from a belief in Shiva explain said belief. Otherwise, even if Hinduism plus Shiva is more fit than Hinduism without Shiva, it is Hinduism itself that is doing the evolutionary work. To review, a selection-for explanation explains a trait just when there is (1) a present or historical biological function, (2) a selection background, (3) specific selection pressures, and (4) quasi-independence. ADAPTIVE EXPLANATION An adaptive explanation is simply a more specific form of the selection-for explanation, and therefore a viable way to explain the existence of a behavior. The difference is that rather than allowing a trait to be mediated through any system of inheritance, 19 an adaptive explanation further specifies that a trait is the product of natural selection mediated through genetic material (see Lewontin, 1978).