Against Selection-For Introduction 1. What is selection for? 2. What traits are selected for? 3.

Could selection for T be relational or comparative? 4. Could selection for be contrastive? 5. The past and future of selection-for 6. Consequences of surrendering selection-for

Introduction Selection for seems both an unproblematic and an indispensible notion in evolutionary biology. In The Nature of Selection (1984) Sober calls it the causal concept par excellence for the theory of natural selection. It is selection for properties that causes differences in survival and reproductive success. To say that there is selection for a given property means that having that property causes success in survival and reproduction. Selection for is to be contrasted with Selection of: selection of pertains to the effects of a causal process, whereas selection for describes its cause (1984, p. 100). There being selection for a particular property means that a causal process is actually in motion (1984, p. 100). Sober attributes the distinction to Larry Wright (1973, pp. 163-164). Whatever its origin the selection for/of distinction has been widely adopted by philosophers and biologists to separate those properties that increase in a population because they enhance fitness from those that increase without doing so, because for example, they are by-products, side effects, or even fitness reducing properties carried along by the fitnessenhancing ones.1 1 In the last decade or so, a few philosophers of biology have argued that selection for is not after all a causal concept (Matthen and Ariew 2002, Walsh et al. 2002, Matthen and Ariew, 2009). These philosophers do not deny the explanatory role of selection for in Darwinian theory. They argue that selection for traits explains evolutionary trajectories, but they deny it is a causal force. See for example Walsh (2002, p. 453. But, like those they criticize, these authors also accept selection for traits as the central explanatory concept in evolutionary biology. Our problems with this notion therefore undermine their alternative interpretation of the theory of natural selection as a purely statistical one.

The present paper argues that selection for a given property is not in fact a cause of the evolution of that property. In the evolution of trait T, it is not selection for trait T that enhances reproductive chances.2 Accordingly the distinction between selection for and section of cannot play the explanatory role in evolutionary processes that philosophers and biologists suppose it does. At best the notion of selection for is a heuristic one that obscures an important feature of the process of natural selection and significant aspects of evolutionary methodology. Some points to bear in mind: first, we have no intention of challenging Darwinian theory. We consider that its explanations of adaptation, diversity and complexity are substantially correct. Our concern is only with a widespread but mistaken assumption about the process that it accurately describes. Second, in our discussion of selection-for we are uninterested in and unconcerned by philosophical conundrums about non-natural kind traits of the grue/bleen sort due to Nelson Goodman. We do not address the alleged problem of whether there is selection for bleenness in robins eggs, along with selection for blueness. Nor are we concerned with Fodors [2011] allegation that the theory that natural selection cannot discriminate between traits that are extensionally equivalent. Our focus is on traits of biological interest. The plan of the paper is the following. In Section 1, we outline how the concept of selection-for has been used in the literature and then in Sections 2, 3 and 4 we argue that there is no unproblematic answer to the question of what traits are selected for in the evolutionary process. In section 5 we explore why biologists and philosophers have assumed there is such a causal process as selection for traits. In section 6 we consider the advantages of thinking about natural selection without the conceptual tool of selection-for. We suggest an alternative causal concept that more accurately labels the process selection-for is wrongly supposed to describe. 1. What is selection-for? If there is selection for X, every object which has T has its reproductive chances augmented by its possessing X. Surely this statement Sober and Lewontin made (Sober and
2 Nor

for that matter is selection for trait T part of a non-causal, purely statistical explanation of

the emergence of trait T.

Lewontin 1982, p. 171) is uncontroversial and representative of the way the concept of selectionfor is employed in the literature. There are several other features that seem uncontroversially to characterize selection for traits. Selection-for takes traits, properties, or again takes having traits or having properties for its arguments. Traits or properties are types or kinds, not the individuals that instantiate them. Each individual instantiates indefinitely many properties or traits (including negative ones). Selection operates on some of these traits but not on others, ones that are also instantiated by other individuals.3 Selection for property-types has effects on individuals, augmenting their reproductive chances. Its very clear that selection for ____ cannot take individuals as its arguments. This would collapse the selection for/selection of distinction. It would deny us the machinery required to distinguish traits selected for, from traits merely carried along by the process of selection for traits. Selection for a trait is no guarantee of increased reproductive success. And individual may have other traits that reduce its reproductive success. Selection for a trait only increases chances of reproductive success. There may be multiple processes of selection for various traits instantiated by the same set of individuals over time: selection for trait X, for trait Y, fort trait Z. And there may be complex patters of interaction between these traits and between the processes of selection for them. Thus, co-evolution of traits is common. Of course whether a trait is selected for depend on the environment in which the trait is instantiated by individuals. So long as the environment remains constant the process of selection for a given trait will continue, presumably until it reaches fixations and becomes ubiquitous in the population, unless of course there are other incompatible traits equally selected for, which will result in multiple adaptational outcomes. Evolution is as Darwin noticed, the process of descent with modification. The selection for traits is supposed to explain how the distribution of these traits changes over the course of a multigenerational evolutionary trajectory. 3 Properties selected for must be capable of multiple instantiation. Thus, they cannot be tropes, the unique property instances or abstract particulars, instantiated only once in one particular object or concrete individual.

Finally, the process that drives adaptational evolution is famously short sighted. There can be selection for traits that enhance the reproductive chances of immediate offspring, or even second-generation offspring. Thus, there can be selection for longevity in grandmothers, provided that longevity in grandmothers enhances the reproductive chances of grandchildren to whom she can convey resources. But natural selection cannot reach much beyond the third generation. There can be no selection for a trait that enhances the reproductive chances of descendants without enhancing the reproductive chances of some chain of overlapping intervening offspring. There cannot be selection for a trait in one generation owing to the enhancement in reproductive chances it makes only in some later generation and not at any time before that generation. Speaking metaphorically, natural selection has to be able to detect or see the properties that are selected for. Non-metaphorically, if there is selection for trait T during generation g, or for having T (the instantiation of T), then T or having T must be causally efficacious in some respect or other during generation g. 2. What traits are selected for? Given the uncontroversial features enumerated above that should characterize selection for a given property, call it T, we need to consider whether there are any instances of selection for that property, that is whether there are biologically significant traits which can be taken as arguments of selection for T We will examine a number of potential candidates: specific determinate properties, specific determinable properties, relational properties and comparative ones. We argue that none of them can turn the concept of selection-for T into the description of a causal process that is actually in motion. None of these traits will satisfy the uncontroversial features of selection for T enumerated above. Consider a famous garden variety of selection-for, one apparently well established and even well understood: the selection of wing color in the peppered moth, Biston betularia. Over the 19th and 20th centuries, environmental changes are widely supposed to have first selected for dark coloration in the wings of the moth and then switched back to selecting for lighter coloration. This process of selection first for and later against dark coloration was due to their effects as camouflage from predator birds in forests where tree trunks and

lichen turned from lighter to darker colors owing to increased coal dusting and then back to lighter color owing to decreases in coal dusting. For convenience focus only on the first half of this evolutionary process: when the lighter coloration in the wings of the moth changed to darker coloration. The environment remained constant during this process, so proponents of selection-for should be able to point to a process of selection for this trait that explains this change in this trait. Exactly what property of moths was being selected for in this evolutionary process? Consider the following candidates: a determinate reflectance wavelength property, a determinable reflectance wavelength property,4 a relational or comparative reflectance wavelength property-difference. These three alternatives seem to exhaust the possibilities. But none can be the property selected for in this evolutionary process. The simplest candidate property selected for is some particular package of specific reflectance wavelengths in the red, orange, and yellow bands of the visible spectrum- characterizing wing color of particular moths (or a grey-scale tone if the predator birds were color blind, like owls. We ignore this alternative below). But it is easy to show that there is no such reflectance property selected for over the process that shifts moth wing colors from darker to lighter and back again. Selection for a trait is supposed to be a cumulative process that results in adaptational evolution. But in the case of industrial melanism, there was no single realized reflectance value that persisted long enough to be the engine of evolutionary change. To begin with, the reflectance values that actually obtain in a given generation, and effected survival and reproduction, may not in fact actually obtain in the very next or any other generation of the moth lineage. The smaller the evolving population the more likely 4 The determinatate/determinable distinction is due to W.E. Johnson and is well known in metaphysics. To a first approximation, a property P is a determinate of a property Q, and Q is a determinable of P, if having the property P is a more specific way of having property Q. Colors present the simplest examples of the distinction. Here is an example: the property of being red is the determinate of being colored, but it is the determinable of the determinate being pink. The property of being pink is, in turn, the determinable of being hot pink. Its determinate, being manic panic hot hot pink may be a superdeterminate, that is a determinate which is not also a determinable for any further determinates. For every determinable property an individual has, it will have a nested set of further determinates all the way down to a superdeterminate, if there is one.

this is. Wing color reflectance value is a trait in part determined by a number of genes. Recombination and developmental noise may result in visible differences between the reflectance values of the wings of every member of a generation and its parents. These differences may even result in darker moths having lighter offspring in the very next generation. If things like this happens, then evidently there is no single unique reflectance value trait present over the course of evolution to be cumulatively selected for. Having a particular wavelength reflectance package is a very specific trait. It is a good candidate for being a superdeterminate (see footnote 4 above). But it is likely that no two individuals in the same or contiguous generations of a small population would both possess the same very specific wavelength reflectance package. Thus, we have no reason to suppose there is a relatively determinate wavelength package actually instantiated and available to be selected for over more than one generation. What is more, a determinate wavelength reflectance trait that is the darkest in the present generation, and is presumably selected-for, may be the lightest in a couple of generations and would not be selected for. So these reflectance package determinates are not the properties selected for. Might there be higher-level determinables of these determinates that are selected for in these cases? Perhaps the trait selected for is that of having one of a range of wavelength reflectance properties (say, a range spreading from wavelength reflectance package L1 to wavelength reflectance package L2)? Such a determinable property could be selected for even though no determinate wavelength reflectance is ever repeated in any generation of any peppered moth lineage. However, this proposal faces serious epistemic and causal problems. Seeing the epistemic problem for the biologist seeking to identify such a determinable enables us more easily to understand the causal problems for natural selection in acting on such a determinable. Selection for such determinable traits is supposed to drive the course of adaptational evolution among these moth lineages. The biologists epistemic trouble is being able to specify, identify, locate this determinable trait, call it W, through the changes in the particular individuals in the lineage over time. It is not difficult to identify the actual wavelength reflectance packages manifested by moths that reproduce. But identifying the specific range of values that is selected for over the entire process is a trickier matter. It wont do for example to identify W as the most determinate color(s) or wave length

determinable, whatever it is, that includes the wavelength value all moths share at fixation when selection is no longer changing the frequency of their wing colors. The biologist can rarely see that far ahead. If the range can only be identified retrospectively, once a reflectance package has been fixed in the population, then no testable predictions about what is selected for can be made before the process of selection has eventuated in fixation of traits. The epistemic problem may be (inductively) soluble, but it is indicative of a more serious causal problem. The process of natural selection can never (metaphorically) see as far ahead as fixation. If the final determinate value or range of values did not even emerge until some point after the process began, it could hardly have been the target of selection from the outset. So the determinable that includes this determinate as an ending or outcome value could hardly be selected for either. But if the determinable trait at fixationthe outcome trait--emerged only late in the process, it certainly was not the one cumulative selection acted upon through most of the natural history from incipient appearance to fixation. Finally, there are metaphysical problems with the suggestion that the trait selected for in this process is a determinable at all, instead of a superdeterminate. David Lewis (1983) has argued powerfully that it is only such maximally determinate properties that have causal powers, and that if causal relations obtain among determinables, they do so only owing to the same causal process obtaining among their determinates. (See also Crane, 2010). If these metaphysical theses are correct, then there cannot be selection for determinables since there is no selection for their determinates. To summarize, neither determinable nor determinate properties seem to be very good candidates for properties that are selected for. So far we have been assuming that the properties that are selected for are monadic reflectance wavelength properties of moth wings. But given that the property that is being selected for clearly depends on the traits of other individuals, perhaps the selected-for property is a relational property that cannot be fully characterized without reference to the instantiation of wavelength reflectance values by more than one individual in the population. We now turn to this possibility. 3. Could selection for T be relational or comparative?

Might the property selected for in the case we have been considering be the property of being darker than the population average (or mean or median, or the mean of the previous generation)? Or perhaps there is selection for the property of being in the darkest N percent of the individuals in the population (where N is some constant, say 33%). The figure certainly need not be arbitrary. Suppose that the limitation of the environmental resources and the rate of reproduction in the population is such that in every generation about one third of the population can survive and reproduce. In these circumstances it may well be that the property selected for is that of being in the 33d percentile for darkness of reflectance value packages. It is not difficult to see that no matter what figure has the best basis in the biology of the case, none of them picks out the property selected for in this process. As with many traits, reflectance wavelength, or whatever is being selected for in the case of the peppered moth, is a product of balancing selection.5 That is, its optimum value is somewhere on a Gaussian curve of distributions of wavelength reflectance values. Too light a color and the moths are more prone to predation by one species of bird, too dark and they are prone to some other threat, say overheating in strong sunlight. Selection for reflectance values moves the moth population toward higher and higher proportions of darker and darker wing colors. But at some point this process begins to select against the darker morphs owing to their overheating in bright sun. Now neither the suns presence nor its strength is a new environmental factor. It is one that has been present throughout the entire evolutionary process. Indeed, it has been part of the selective environment for moth wing colors, playing a role in selecting for reflectance values just because sunlight is required by the (diurnal) predators to detect moths. Once we factor in the role of the sun in also limiting the reflectance values of moth wing color to some maximum value, it becomes apparent that the process cant be selecting for longer wave lengths (darker colors), since after a certain point longer wave lengths are maladaptive, are selected against.

5 The argument to be given can be made for cases of disruptive selection and directional selection as well. It only requires that there be a peak in trait fitness values beyond which variation invariably produces some individuals. We use stabilizing selection largely for heuristic reasons.

Whatever value N make take, after a certain point, stabilizing selection will obtain, selecting against the trait of being in the Nth percentile of darkness. The selective process will in fact result in fixation at some very narrow band of darkness, and will condemn variations in the Nth percentile beyond it to extinction. This problem illustrates even more starkly the requirement that selection for trait T be a process that can detect, can see trait Ts instantiation in individuals. If we seek the property that is selected for in the process of adaptational evolution of moth wing color, we need to identify a property that is not also selected against in the same environment. We may be tempted to suppose that in this environment, there is selection for a goldilocks traitnot too light, not too dark, but just right. But surely this outcome trait is not the trait that is selected for by the Darwinian process that results in its fixity or preponderance in the population. The outcome property, which is not yet instantiated, cannot be any part of the causal process [that] is actually in motion, to use Sobers language (1984, p. 100). To suppose otherwise is to deprive Darwinian theory of one of its important feature. Let us see why. Darwins theory deprived real teleology of any role in the biological realm. It offers a purely causal explanation for the appearance of design. It holds that all adaptations are products of purely causal processes of the same kind that operate in chemical and physical domains. It invites us to locate the source of the appearance of design in the combination of blind variation and passive environmental filtration. The Darwinian process is a winnowing one, continually removing variations from the population, even as variations continually arise in the population. Which variations arise is not dependent on the needs of their bearers and the benefits these variations confer on their bearers; which variations are removed is dependent on the bearers needs and the costs these variations impose. Suppose that the trait selected for in the case of our moths, W, is that of having the optimal reflectance wave length package, a set of wavelengths (or wave length ranges) that optimizes camouflage from birds while minimizing heat exhaustion from the sun. How can the Darwinian process select for this trait without foresight, without somehow aiming at a trait that may not in fact be exemplified at any earlier point in the evolutionary process that eventuates in its attainment. The only way this could happen is if future states can play a causal role in the

10

processes that bring them about. This is not just farsightedness, or even clairvoyance. Its teleology par excellence. But surely the Darwinian process is selecting for those transient traits that will eventually, at fixation, give rise to the adaptationally optimal outcome trait, or some at least locally optimal outcome trait(s), or would have given rise to such a trait or traits had the environment not changed. So, are the traits being selected for the ones that survive a process of Darwinian winnowing long enough so that their descendant outcome traits are the ones remaining at fixation (or would have remained at fixation had the environment not changed)? The answer must be No. If the transient trait selected for is that of being the ancestor of W, the descendant outcome trait present at fixation, then biologists will be unable to identify this trait without knowing what outcome traits are eventually fixed. This is the testability problem that emerged in the last section. Independent of the epistemic problem of the biologist, consider what features these traits could have in common in virtue of which they are selected for? They cannot be selected for owing to the fact that they share some descendant trait in common. Natural selection cannot discriminate among traits owing to their effects beyond one or two generations. It cant select for being the trait or traits whose distant descendant trait goes to fixation (or would do so if the environment remained constant for long enough). There will have to be something else, more immediate, actually co-present with their instantiation, that these traits share in common to which their being selected for must be attributed.6 Our search for the trait selected for in the process of adaptation that culminates in coloration of the peppered mouth has not met with success. Perhaps selection for T is a short hand label for a more complicated process that does involve causally efficacious determinates that actually obtain, a process of contrastive selection for T as opposed to some other trait, S. We consider this option in the next section. 5. Could selection for be contrastive? 6 An example well known in paleontology and sometimes cited by philosophers illustrates the point well: the three bones of the mammalian inner ear are the result of a long process of selection for hearing that begins with selection operating on the the articular and quadrate of the reptilian jaw bone-joint. The process, which began with these bones loss of function as jaw bonejoint, can not be one in which the outcome trait, mammalian accustical sensitivity, plays a persistent role.

11

In each generation natural selection filters out some traits while allowing others to be passed on to offspring. In the case of the peppered moth it filters out reflectances below a certain threshold of wavelength, and allows wavelength packages above that thresh-hold to be passed on. Call this the extinction threshold.7 Should we accept that all of the reflectances that pass through the environmental filter are selected for? Certainly they are not equally selected for. Those closer to the threshold wavelength package will have on average fewer offspring: their traits will be less selected for. Indeed, these traits will eventually be selected against, as the threshold wavelength shortens owing to competition among moths for greater degrees of camouflage. In fact, given the continuum of moth-survival and reproduction values associated with each reflectance, its clear that any given reflectance may be selected for in competition with wavelengths closer to the extinction threshold and selected against with respect to wavelengths further from the extinction threshold. On this view of the process, natural selection is fundamentally a contrastive mattertrait A is selected for more than trait B is, or trait A is selected for with respect to trait B, but trait A may be selected against compared to, or with respect to, trait C. On this view of the process of natural selection many reflectances are selected for to different degrees and just as often the same trait is against as well. Perhaps we can build a concept of Selection for a particular trait sans phase out of the contrastive concept of selection for trait T over trait S. Go back to Sober and Lewontins characterization of selection for: if there is selection for X, every object which has X has its reproductive chances augmented by its possessing X (Sober and Lewontin 1982, p. 171). It is easy, indeed natural, to minimally reformulate this formulation as a contrastive one: if there is selection for X compared to Y, then every object which has X instead of Y has its reproductive chances augmented by its possessing X compared to objects otherwise similar that have Y instead of X. Selection for takes two traits as arguments. It is a relation between properties and not a monadic property of properties.

7 Owing to drift and other probabilistic factors, the filtration process is actually quite variable, consigning individuals of the same color to different immediate fates, but on average and over the long term, eliminating some traits completely. We neglect this complication in the present discussion.

12

As we argue below, contrastive selection is something we should accept as part of the Darwinian process. Alas it is not a relation on which we can build the monadic causal concept of selection for a particular property. Notice to begin with, that two determinate wavelength packages which are actually instantiated, and which stand in a relationship of contrastive selection-for, may only be instantiated once in an evolutionary trajectory. There may not be sets of instantiated traits that stand in the transitive relationship of contrastive selection-for over the evolutionary trajectory that the selection for a trait is supposed to explain. Of course a variety of counterfactual relations of contrastive selection-for will obtain for wavelength values that are not instantiated but could be over this trajectory. However, it is hard to see how counterfactual contrastive selection for uninstantiated traits explains the actual trajectory of instantiated traits in an evolutionary process. Wave length L1 may be contrastively selected for over wavelength L2 in generation 1, but if neither is instantiated in any subsequent generation this fact about the contrastive selection of L1 over L2 has no general explanatory role to play over an evolutionary process. L1 and L2 are highly determinate properties, perhaps even superdeterminate ones. They may not play enduring roles in an evolutionary process. Perhaps one or another of their determinables may do so however. Consider for example contrastive selection for one range of wavelength values over other ranges. Darwinian gradualism suggests that trait-value ranges will always be instantiated in a transitive succession of values. Unless macromutations or other sources of punctuated equilibrium obtain in the evolutionary process, this is an assumption we can have some confidence in. But even with this guarantee of transitivity trait value ranges will face the Goldilocks problem sketched above. Of course there will be some range of wavelength values that is contrastively selected over all other wavelength valuesthe attainable environmentally optimal one, what we can called the outcome trait. But this range will not figure early in the process of selection that brings it to fixation, nor can biologists determine its value prospectively. Indeed it can have any role in a teleology-free Darwinian process. These considerations do not be any means suggest that contrastive selection does not obtain in the evolutionary process. But they do cast doubt on our ability to underwrite a noncontrastive concept of selection for particular properties by building it up out of contrastive selection for determinates or their determinables. The trouble is that non-contrastive selection for particular properties is supposed to have an important explanatory role in Darwinian theory. Its the

13

increase in the proportion of individuals with trait T that non-contrastive selection for a trait T is supposed to explain. It is difficult to se how contrastive selection that does not involve T at all can be used to define or construct a concept of selection for T simpliciter. To see the challenge, start with the simplest candidate for a definition of selection for T via contrastive selection for T over other traits. Suppose there is selection for a trait value T simpliciter in generation g if and only if for any trait value realized in g that is different from T, call it T*, there is contrastive selection for X over X*. Selection for T in generation g is a matter of Ts being a sort of Condorcet winner over each of its competitors. If we defined noncontrastive selection-for simpliciter this, way, we could move from the indefinitely many true contrastive selection-for statements to a conclusion about non-contrastive selection for traits in generation g: the trait selected for sans phrase in g is that trait which is comparatively selected for with respect to all other traits. Alas, the Condorcet winner trait may not be realized in the next or any future generation, to continue to be the target of selection for. So, we need something more abstract. Can we build a definition of selection for T simpliciter out of (iterated or otherwise concatenated) contrastive selection for some set of Ts determinates? The project would be daunting, and would certainly provide targets for the construction of counterexamples (an echo of the succession of definitions for the probabilistic propensity-conception of fitness). But even before such a program begins it is important to note that the concept thus defined would not preserve the role that selection for a particular property was introduced to play. Suppose we could construct a bi-conditional with There is selection for outcome trait T on the left side and some complex set of conditions built up out of contrastive selection either for some of Ts determinates or other traits that share determinables with T. Such a definition would transfer all of the causal power of selection for a particular trait to a pattern or patterns of contrastive selection for some quite different traits! The result would simply be the appropriation of the inscription selection for trait T to describe something importantly unlike the one which it was introduced to describe. Having the trait T would not longer play a cumulative or much of any role in the process that results in it. Thus the redefinition of selection for T is reminiscent of the continued use of the word gravity in the general theory of relativity. It continues to be used, owing to its convenience and familiarity, even as the theory explains away its distinctive causal features.

14

Instead of a definition, perhaps it would suffice to vindicate selection for a particular property if it could be shown to bear a weaker relation to processes involving contrastive selection. Lets consider whether selection for a particular property can supervene on a process involving one or more of its determinate traits contrastively selected for. The epistemic barriers to identifying such a property selected for and its supervenience base may not be great. For almost any adaptive trait or trait value we can retrospectively establish the increase in its frequency in the population over time, and we can make a good experimental or observational guess about which of its determinates the process of contrastive selection operated on to produce this increase in frequency. Can we prospectively identify a determinable as the trait selected for via the contrastive selection for some of its determinates? In some cases this will be possible. But, assuming that the biologists best guess at what trait is being selected for simpliciter is correct, can we accord selection for this trait the requisite causal role in increases representation in the population, while also according a causal role to contrastive selection for its determinates over time? This proposal raises some metaphysical issues already familiar from elsewhere in the literature on supervenience. Most obviously, we will be faced by the dilemma of explanatory competition versus overdetermination that Kim [1989) has famously raised in another context. Kim raised the problem for the attribution of causal roles to a functional property and its physical realization. However controversial the argument is in that case, it must be much less controversial when the two processes that either overdetermine or compete are ones involving just determinables and their determinates. If selection for a determinable trait simpliciter and contrastive selection for one of its determinates are distinct causal processes, then every evolutionary process is overdetermined. If they are not, then either one of them is not a causal process or the explanations that invoke them must compete with one another. There is in addition the biologically more serious problem of explaining how the outcome trait T figuring in the supervening process of selection for T can play a causal role in a process during which it is never instantiated until sometime near or at the end of the process. Of this difficulty more below. 5. The past and future of selection-for

15

If there is really no such thing as selection for traits, the question arises: Why do philosophers of biology and biologists employ the concept of selection-for to characterize the Darwinian process? The answer to this question does not seem to be hard to find. What Haldane (1942) called the modern synthesis of Darwins theory and population genetics, established the conception of evolution as changes in trait frequencies. The debate has raged for several generations as to which trait-frequencies of what sort of individuals are to be explained. No matter what the resolution of this debate, the parties to it shared the view that the explanandum phenomena in evolutionary biology are best characterized in terms of particular traits whose proportions in a population can be measured. If the effect to be explained is a change in the frequency of trait T in a population, it seems natural to expect that the explanans should advert to the causes of trait Ts frequency change. Since Darwinian theory invokes the process of natural selection as the cause of evolutionary adaptation, the default assumption of an inquiry about what explains changes in the frequency of trait T will be some facts about trait T. When T appears to confer an advantage or a benefit, they will be facts about Ts effects in the past; when T does not confer an advantage but nevertheless increases its frequency in a population, the search will shift to Ts association with some other trait that is selected for. Lewontin (1978) famously noted that to produce finely tuned adaptations the process of natural selection must be able to target traits individually from one another, so that adaptational enhancement of one could proceed without producing increased maladaptation in others. It is natural to treat the adaptation of individual traits as the result of selection for those very traits. Even when the traits ultimately produced, the outcome traits, are the result of a sequence of changes among a lineage of distinct precursor traits, it is still natural to assume that all the instances in this evolutionary line of decent share some specific determinable that causation is acting upon, some trait that is selected for through the welter of evolutionary changes that bring about the trait fixed at the end of the sequence. We have seen that evolutionary sequences which appear to be cases of selection for T are in fact cases in which T is a determinable and contrastive selection among its determinates is the actual causal process shifting the frequency of T. Failing to be precise about what traitthe determinable or one of its determinatesis actually seen by the process of natural selection fosters the illusion that there is selection simpliciter for the determinable. Making the

16

determinable/determinate distinction is not one likely to occur to nonphilosophers. And without this distinction, one will have difficulty marking the difference. For a few purposes, especially for the exposition of the theory of natural selection, the assumption that there are traits selected for over the course of an evolutionary process may be harmless. As an idealization, the assumption that there is selection for traits may be a pedagogically convenient simplification. Imagine being required to introduce the machinery of determinates and determinables in order to show that the Darwinian process does not proceed by the selection for particular traits over generations. In some simple textbook cases, the assumption that the evolutionary process does select for a trait may exclude complications safely if temporarily ignored. Consider the case of a traittype, what biologists often call a character, T that comes in only two biologically possible values, T and T. In the evolution by natural selection of values of this character, there is unproblematic route from contrastive selection-for to non-contrastive selection-for. Contrastive selection for T as opposed to T realizes selection for T simpliciter and, as a result, in these scenarios, selection-for can explain the change in trait frequencies of T and T. So long as there are no additional values of the character T actually instantiated in a Darwinian process, the shift from contrastive selection for T to selection for T sans phrase is unproblematical. So long as the number of values of character T is fixed and finite, and so long as there is a transitive ranking among them of contrastive selection-for, there will be a process of selection simpliciter for one of them. Using this simplification to illustrate the process of natural selection strongly encourages the notion that there really is such a thing as selection for traits. However, once we add features to this idealization common to every real evolutionary trajectory, it becomes clear that selection for traits does not actually occur in the evolutionary process. Giving up selection for particular properties as a description of the causes of evolutionary outcomes seems a radical proposal, even if the argument is accepted that no such process really obtains in the biological domain. Should we perhaps explicitly adopt some variant of the redefinition of selection for considered above, so that biologists and others can continue to use it without its misleading suggestion that it is selection for T that has a causal role? We define selection for T as the process of blind variation and environmental filtration of a sequence of some determinates or other of T that results in the increased frequency of T in the population, where the particular sequence and perhaps even which determinates of T will vary

17

from case to case. Selection for T will always connote selection for outcome trait T with no suggestion that the outcome trait has any causal role in its frequency. One reason not to adopt the concept of selection for under this redefinition is the strong suggestion of teleology that it retains, even after the word selection has been Darwinized. Natural selection is not a kind of selection. That is, it is not a process of intentional choice among alternatives governed by a thought, image, representation of some desired outcome. This makes natural selection inapt in some respects as a label for the process Darwin discovered. Biologists understand the label as lacking any commitment to intentionality. But over and above the intentional suggestion that the word selection conveys, there is a further misleading connotation in the expression for T. This suggestion is exacerbated by the redefinition under consideration. Selection for trait T suggests that the trait T has a biological reality independent of and prior to the process that brings about increases in its frequency. By invoking a determinable and its distinct determinates, and explicitly identifying the determinates as the participants in the causal process that brings about increases in the frequency of the determinable, the redefinition still subtly suggests that the determinable trait has some role in the process that brings it about. From this presumption to the conclusion that the determinable traits role is teleological seems an inviting one to those (almost all nonbiologists) given to teleological thinking. There is still another reason to surrender the notion of selection for, instead of reconfiguring its meaning. The role of the trait, T, in the label selection for T as the causal process that brings about the emergence, persistence, spread and fixity of trait T, obscures the fact that it is contrastive selection for some determinates of T that brings about the presence of T. Failure to see this encourages specious claims that the theory of natural selection is untestable or unfalsifiable. None of these charges against the theory need be taken seriously by anyone who understands it and has a good grasp of how theories are tested. But among those who lack relevant knowledge, the use of the expression selection for T as an explanation for the presence of T makes these charges tempting. If we are right selection for traits turns out not to be the causal concept par excellence for the theory of natural selection or for the process Darwin discovered. In fact it turns out not to describe the Darwinian process at all. We doubt Darwin would be surprised by this conclusion. Eschewing selection for T as the cause of the adaptive evolution of T accords well with his own

18

persistent emphasis on adaptational evolution as the result of a very gradual process accumulating slight variations. His emphasis on the struggle for survival between slight variations directs us to the process we have called contrastive selection between actually occurring traits and trait-values. This relation between actually instantiated traits is the engine of the Darwinian process. But even the notion of contrastive selection for traits retains the expression selection and along with it the teleological baggage that we have complained that selection for traits bears. So, we do not propose that the label contrastive selection replace the label selection for as the name of the fundamental causal process in adaptational evolution. A less misleading alternative label for it might be something like contrastive environmental filtration against T compared with T. While we suggest that the theory of natural selection would be well rid of the notion of selection for traits simpliciter, we are under no illusions that our alternative will be adopted. The most we can hope is that as with the role of gravity in general relativity, scientists will make the obvious adjustments while continuing to use the words selection for. 6. Consequences of surrendering selection-for It should be obvious that treating contrastive environmental filtration against T compared with T as the fundamental causal process in Darwinian evolution does not deprive us of any distinctions we need to make, nor any explanations we wish to offer for adaptational evolution. In particular it is clear how the contrastive notion we have identifiedwhether called selection or filtrationcan distinguish between traits that increase owing to their adaptational effects from those merely carried along, in the parlance we deprecate: traits selected without being selected for. Traits with locally more adaptational effects are the ones that win head-tohead contrastive contests. Individuals bearing these traits will have their reproductive chances enhanced. There will be selection of these individuals. We make no objection to the use of the term selection to describe evolutionary effects. These individuals will bear many traits other than the ones that win the contrastive contests and assure their bearers enhanced reproductive e chances. Many of the other traits of these individuals will also increase in the population just because they are carried along with the local contrastive winner traits.

19

There is a significant philosophical pay-off to the recognition that Darwinian evolution does not proceed by a process of selection for particular properties. It helps to explain perplexing features of the theory of natural selection. For a long time now philosophers of science and some biologists have sought to frame formulations of the theory of natural selection that identify its fundamental nomological generalizations and to show exactly how they organize evolutionary inquiry and explanation. This enterprise has not met with success, or at least it has not resulted even in a set of candidate formulations of the theory that secure much acceptance. Other philosophers and some biologists have sought to explain why this is so, while at the same time arguing that the fact we cannot frame versions of the theory that share the structure of theory in other natural sciences is not a matter of methodological or epistemological concern. Perhaps the appreciation that selection for traits is not the causal concept par excellence in natural selection can shed light on both these mattersthe failure to identify distinctive laws of natural selection that govern the process of adaptational evolution, and why they may not be required by evolutionary biology. To see why this may be so, consider a claim by Sober closely related to the selection for/selection of distinction: that there are in evolutionary theory source laws and consequence laws. The distinction is illustrated in Sober, 2000 (p. 22). Organism/ Fitness Differences Environment ------------ Mutation Relationships Migration Drift System of Mating Source Laws

-------------- Evolution

Consequence Laws

The figure does not include selection for but the text makes it clear that Sober considers fitness differences a matter of selection for traits since he describing the diagram he substitutes selection for fitness differences: Within a population selection, mutation, migration, recombination, pattern of mating, and drift may all contribute to the changes in frequencies that result (2000, p. 21). The diagram reflects the suggestion that selection, along with sources of variation such as mutation, figure in the consequents of source laws and the antecedent of

20

consequence laws. When it comes to providing examples of source and consequence laws in biology, Sober faces a problem: there do not seem to be source and consequence laws of the sort philosophers of science will recognize. Sober himself has identified mathematical models as the relevant laws, in spite of the fact that they are necessary truths and not contingent claims making claims about causal and other factual relations. 8 The example of a source law he provides is Fishers model of the 1:1 theory of the sex ratio investment. It is interesting to note that in this case of a source law the trait supposedly selected forequal parental investment in males and femalesis an effect and not a cause of the process Fisher describes. This will be unsurprising on the view we defend here: The process which eventuates in the fixity of this trait among populations of almost all sexually reproducing species, is not the result of selection for the trait. The 1:1 sex ratio is an outcometrait, the result of continuous contrastive selection among male/female birth ratios that jointly achieve and maintain the outcome trait as a stable equilibrium by the persistent variation among a range of sex-ratio traits actually instantiated by individuals in a reproducing population. The 1:1 sex ratio is an outcome of continuous contrastive filtration among other ratios that achieve and maintain it as a stable equilibrium in spite of persistent variation among a range of sex-ratio traits actually instantiated by individuals in a reproducing population. A real law, that is a contingently true generalization about what all these cases of contrastive filtration of off-spring sex ratios have in common besides the outcome trait of a 1:1 sex ratio, is evidently impossible to formulate. Whence the absence of a contingent source law in this case. If there are to be universal general statements about traits fixed in a population, such as a 1:1 sex ratio, the only candidates will be necessary truths, analytic statements or mathematical ones. More significant is what Sober notes about consequence laws, ones in which the selection for traits should figure in the laws antecedents, where causes can expect to be found: Population geneticists often work out the consequence laws of evolution. Once the magnitudes of the various causes are specified, a population genetics model allows one to 8 Sober writes, Are there general laws in evolutionary biology? Although some philosophers have said no, I want to point out that there are many interesting if/then generalizations afoot in evolutionary theory. Biologists usually dont call them laws: model is the preferred termThese mathematical formalisms say what will happen if a certain set of circumstances is satisfied by a system (2000, p16).Sobers claim, that despite their being naecesary truths, such mathematical models can provide causal explanations has been subject to dispute. See for example, Sober (2011) and Lange and Rosenberg (2011). W do not pursue the matter here.

21

compute the evolutionary consequences. It is not part of such models to say why one genotype is fitter or why there is a difference between forward and backward mutation rates at some locus (2000, p. 21) In effect Sober admits that there are no laws, no contingent nomological generalizations about the effects of selection for traits. There are only the mathematical models of population genetics that take the fitness values of traits, their initial frequencies, and population parameters as inputs, and provide outputs regarding future trait frequencies. But the absence of contingent consequence laws with selection for traits in their antecedent is exactly what we should expect, if there really is no actual process of selection for trait T, if selection for trait T is never a cause of trait frequencies. Bence Nanay Alex Rosenberg References Crane, T., 2008, Causation and Determinable Properties: On the Efficacy of Colour, Shape, and Size, in Being Reduced: new Essays on Reduction, Explanation and Causation, ed. by Jakob Hohwy and Jesper Kallestrup, Oxford, Oxford University Press. Fodor, Piattelli-Palmarini, 2011, What Darwin Got Wrong, London, Picador Gould, S.J., and Lewontin, 1979, The Spandrels of San Marco and the Panglossian Paradigm, Huxley, J. 1942, Evolution: the Modern Synthesis Kim, J., Mechanism, Purpose, and Explanatory Exclusion, Philosophical Perspectives, Vol. 3, Philosophy of Mind and Action Theory (1989), pp. 77-108 Lang, M., and Rosenberg, A., Can there be A priori models of natural selection, Australasian Journal of Philosophy, 89 Lewis, D., (1983). New Work for a Theory of Universals, Australasian Journal of Philosophy 61: 34377, Lewontin, R., 1978 "Adaptation," Scientific American, vol. 239, (1978) 212-228. Matthen and Ariew Sober, E. and Lewontin, R., 1981

22

Sober, E., A priori causal models of natural selection, Australasian Journal of Philosophy, 89 Sober, E. 1984, The Nature of Selection. Cambridge, MA: MIT Press. Wright, Larry, 1973, Functions, The Philosophical Review, Vol. 82, pp. 139-168