In his recent book, Godfrey-Smith (2009, 53-63) develops a principled method for distinguishing natural selection from drift from the perspective of entities forming a population that is perfectly compatible with a fully deterministic set-up. His motivation for doing so is that, in their classical versions, natural selection and drift rely on the notion of expected fitness or expected reproductive output. Yet, the classical interpretation of expected values both in philosophy of biology and theoretical biology is the propensity interpretation of probability which, as we have seen, is highly controversial. Thus some of Godfrey-Smith’s and Bouchard and Rosenberg’s motivations for revising the classical account of fitness seem to be similar. With his framework, Godfrey-Smith intends to replace the classical interpretation by another one that does not rely on the expected values.
While the classical notion of natural selection relies on differences in expected reproductive outputs between entities of different types, the classical notion of drift is more difficult to grasp. Classically, the notion of drift is often associated with the notions of ‘sampling error’, ‘indiscriminate sampling’ or ‘random sampling’ (e.g. Beatty 1984; Brandon 2005; Crow & Kimura 1970; Gillespie 2004; M. Hamilton 2009; Hartl & Clark 1997; Millstein 2002) although others, such as Rosenberg have proposed a different interpretation, namely that drift is a useful fiction (Rosenberg 1994). In population genetics, differences in reproductive output occurring by ‘chance’ (random drift) or ‘accident’ between entities of a population and leading to
evolutionary change are often considered as synonyms for evolution due to drift. Although this statistical description is pragmatically useful in evolutionary theory, it does not capture substantially, that is causally, what the concept of drift amounts to. In fact, although the notions of ‘chance’ and ‘accident’ can be quantitatively apprehended via the notion of expected value, the interpretation of what expected values on reproductive output mean is highly problematic. The propensity interpretation of fitness (Beatty & Finsen 1989; Brandon 1978; Mills & Beatty 1979; Sober 2002) we encountered earlier is an attempt to make sense of those expectations. It tells us that fitness is a tendency or disposition in the same way that ‘resistance to impact’ is one. Entities are ‘expected’ to produce a certain number of offspring in a given environment in the same way that a glass is expected to break under certain conditions.
Yet, propensity interpretations of probability, as we have seen, are controversial for they suffer from the charge of being causally empty in evolutionary theory (e.g. Byerly & Michod 1991) and more generally empty accounts of probability without a clear explanation of what propensities are (Eagle 2004; Hájek 2012). To capture more substantially the notion of drift, we need conceptual tools that do not rely on a particular (and controversial) interpretation of probability, and that are compatible with a fully deterministic world since it is not clear whether quantum events percolating up to biological processes or brute biological propensities have an important role for natural selection (Bouchard & Rosenberg 2004). Godfrey-Smith (2009, 53-63) precisely develops tools that are compatible with those requirements. He holds the view that the distinction between natural selection and drift has something to do, among other things (more on this below) with the notions of intrinsic and extrinsic properties of entities forming populations. Although I believe Godfrey-Smith’s account has some problems, I think it is on the right track.
Godfrey-Smith proposes that when in a population there is variation in intrinsic properties between its members, and it leads them to have different reproductive outputs, the resulting evolutionary change should be attributed to natural selection. Conversely, when this difference in reproductive output is due to differences in extrinsic properties, the evolutionary change resulting should be attributed to drift. He defines an intrinsic property as a property that, contrary to an extrinsic one, does not depend on the existence and arrangement of other objects. A good example of intrinsic property is the chemical composition of an organism. Examples of extrinsic properties include being in a particular location or someone’s cousin.
The rationale behind this view is that with extrinsic properties as opposed to intrinsic ones, the differences in reproductive outputs they are causally responsible for cannot be systematically attributed to their bearers. In some sense intrinsic properties are constitutive of an entity while extrinsic are not. Another way to understand this distinction is to use counterfactual dependences. When evolutionary change is due to drift, had at least some circumstances the entities are found in been otherwise, the extrinsic properties of their bearers would have been different and led these entities to have different reproductive outputs. When evolutionary change is due to natural selection, had the entities themselves been otherwise (that is, their intrinsic properties), then their reproductive outputs would have been different, independently from the circumstances these entities might have been found in. By associating extrinsicness with drift, we recover the classical notions of accident and chance classically associated with drift, where chance or accident can be understood as ‘that does not depend on the entities forming the population’. It follows that two types of entities with different extrinsic properties (that do not ultimately causally depend on intrinsic properties), even in a fully deterministic set-up, should exhibit drift. Under this view, the statistical nature of a type’s reproductive output should not be regarded as
the result of indeterminism, but as a measure of the extent to which the member of a population differ in the extrinsic properties insofar as they lead to differences in reproductive outputs.
This view, if correct, has one important consequence: it follows from it that had each extrinsic property been the same for all the members of a type, each member would have had the same reproductive output. This means that this notion of drift does not predict that under a perfectly homogeneous environment for each entity of a population, smaller populations would exhibit higher levels of drift as predicted by a statistical definition of drift (I will come back to this point in Section 1.6). However, it explains why this is observed in most cases. To see this clearly, let us suppose a population composed of two types of entities (that is, with one difference in intrinsic properties) that can be found independently in several different states of the environment. Each state is supposed to lead to different consequences for the reproductive output of an entity. If a population is composed of a small number of entities and there is nothing intrinsic to the entities that determines in which state of the environment it should be, it is very unlikely that for each entity of a given type in a particular state of the environment, there will be another entity of the other type in the same state of the environment. However, the probability of finding matching entities of each intrinsic type in the same environmental state will increase as population size increases. When the population size is infinite, any given entity of one type in a particular environmental state will have a matching entity of the other type in the same environmental state. At that point, any difference in reproductive output between the two types will be attributable to a difference in intrinsic properties only. This notion of drift is thus compatible with its classical statistical interpretation. It should be clear that real populations are usually composed of more than two types and many more than several ‘micro-states’ of the
environment. Yet, the simple reasoning provided here remains valid for populations with any number of types and environmental states.
Godfrey-Smith’s (2009) view on drift is actually more complex than the one presented earlier. In fact the dependence of reproductive output on intrinsic properties is only one of five important features to characterise Darwinian populations, that is, populations able to exhibit ENS26 (for a more complete assessment of Godfrey-Smith’s notion of ‘Darwinian population’
see Chapter 2). Godfrey-Smith calls the feature measuring the dependence of realised fitness (for our purpose in this chapter this is equivalent to reproductive output) on intrinsic properties ‘S’. Along with S the four other features are fidelity of heredity ‘H’, abundance of variation ‘V’, competitive interaction with respect to reproduction ‘α’ and what he calls continuity ‘C’. For Godfrey-Smith the parameter C is also involved in drift. But before presenting what this feature entails and why Godfrey-Smith associated it with drift, it is useful to digress a bit and briefly present the three other features.
The features H and V underlie quite straightforward concepts, and represent respectively how reliable heredity is in the population and how much variation there is between the different entities of the population. Without variation there cannot be ENS because there is nothing to ‘choose’ from. If there is no or a very weak heredity of traits between parent(s) and offspring, then natural selection will have no durable consequences on the population. At least that is what I will assume for now, but I will come back to the notion of inheritance first in Chapter 2 and then in chapters 3 and 4.
While the concepts underlying H and V are relatively straightforward, the ones underlying α and C are a bit more demanding. A simple way to understand what ‘α’ represents using Godfrey- Smith’s own words is “the extent to which adding reproductive success to one individual reduces another’s” (2009, 52). With this parameter Godfrey-Smith is introducing the Malthusian notion of competition between the members of a population. Finally ‘C’ represents the level of change in reproductive output induced by small changes in an entity’s phenotype (2009, 57). For each of the five parameters in Godfrey-Smith’s framework, the higher the population scores on this parameter, the more the population is a paradigmatic case of a Darwinian population, leading to paradigmatic form of ENS as opposed to marginal case of a Darwinian population leading to a marginal form of ENS. Godfrey-Smith represents three of the five parameters (H, C and S) in what he calls a ‘Darwinian space’ (2009, 63-67) which allow to visualise how far from an ideal paradigm case of a Darwinian population is an actual population. I present a simplified version of it in Figure 1.1. Although the approach to natural selection in terms of a similarity space is very useful, I will examine some of its problems in more details in the next chapter and more particularly in relation to heredity.
Figure 1.1. Godfrey-Smith’s Darwinian space, reproduced and simplified from Godfrey-Smith (2009b). (S: dependence of realised fitness (reproductive output) differences on intrinsic properties; C: continuity; H: fidelity of heredity.)
Going back to drift, Godfrey-Smith holds that a very low C, that is, when small changes in the phenotype lead to large changes in fitness, should be associated with drift especially when S is also low. This can be visualised the Figure 1.1. I disagree with Godfrey-Smith on this point for several reasons. First, he considers that small differences in “everything about an organism [read ‘entity’]” (2009, 61) (that is, both intrinsic and extrinsic properties) can in some cases lead to large differences in reproductive outputs and should thus be associated with drift. This is problematic. In fact, in cases of large differences in reproductive outputs due to small differences in extrinsic properties, those cases should, in my view, not be considered as of a different nature from cases of differences in extrinsic properties. They represent merely a subset of the cases in which differences in extrinsic properties lead to differences in reproductive outputs. Different levels of C, it seems, modulate the level of drift resulting from difference in extrinsic properties,
but they do not change its nature. Note furthermore that Godfrey-Smith does not provide any principled reasoning as to what thresholds should be used to considered ‘small’ differences between entities’ properties and ‘large’ differences in reproductive outputs.
Concerning small differences in intrinsic properties leading to differences in reproductive outputs, I find Godfrey-Smith’s proposition even more problematic. In fact, these differences should be regarded, as per his own distinction with the parameter ‘S’, as differences due to natural selection, not drift. If Godfrey-Smith’s distinction between intrinsic and extrinsic properties is right, it cannot be the case that small differences in intrinsic properties are sometimes attributed to natural selection and sometimes attributed to drift especially since, in this case too, Godfrey- Smith does not propose a principled way to distinguish which ‘small’ intrinsic differences should be attributed to natural selection and which ones should be attributed to drift.
Second, it seems that the conceptual work made by Godfrey-Smith’s parameter C helps best conceptualising the conditions under-which natural selection can lead to complex adaptations, a point I will briefly develop in the next chapter. If slight differences in properties whether intrinsic or extrinsic lead to dramatic changes in reproductive outputs, that is, if the population exhibits a low C, it is very unlikely that anything like a lensed eye could ever evolve. Godfrey-Smith’s (2009) project, as I see it, is a (successful) attempt to delimit the conditions under which natural selection will lead to adaptations as classically understood by evolutionary biologists, that is, through cumulative selection. Yet, the question as to whether natural selection (or drift) is responsible for the pattern observed (complex structures) in populations and the question of whether natural selection (or drift) is an evolutionary process occurring in the population are quite different and should be distinguished. In fact, we will see in the next chapter that the process of natural selection does not necessarily lead to the production of complex
adaptations. I think that populations with low C are instances of such cases. To me, a low C predicts that natural selection cannot lead to the evolutionary process responsible for complex adaptations, not that evolution is drifty, even if empirically ENS without complex adaptations might be extremely hard to distinguish from evolution due to drift.
1.4. Problems with the intrinsic/extrinsic distinction