Introduction

In the Introduction of this thesis, I have mentioned several evolutionary processes and asserted that they can cause evolution. Mutation, I have claimed, causes transformative evolution while natural selection and drift, once they are isolated, cause distributive evolution. This description is in line with population genetics, the mathematical branch of evolutionary theory, that classically distinguishes four causes or forces of evolution:13 _{natural selection, mutation,}

migration and drift (e.g. Crow & Kimura 1970; M. Hamilton 2009; Hartl & Clark 1997).14 _Yet,

what constitutes an evolutionary cause or force is not immediately clear. It has been and still is subject to an important debate in philosophy of biology. Most of this literature focuses on natural selection and drift while leaving out mutation and migration. In this chapter I follow suit. I will treat mutation in more detail in Chapter 2. Migration will not be treated in the thesis because I will consider that evolutionary change is a phenomenon that arises in an isolated population without external interventions of other populations. Thus, migration is not relevant at the level of generality I will be working. Furthermore migration is usually not mentioned in the formal descriptions of evolutionary change. For the interested readers, Kerr and Godfrey‐Smith (2009) propose to extend the Price equation, a formal description of evolutionary change mentioned in the Introduction of this thesis (see Chapter 4 for more details), to cases of evolutionary change

13 _{Distributive evolution and transformative evolution being undistinguished here.}

14 _{This division into four forces or causes is to some extent arbitrary. Some authors would, for example, add sex or}

that incorporate arbitrary causal connectivity between ancestors and descendants of a population. This formalism accounts for migrations processes and fill the void left by the classical equations.

It is nearly uncontroversial to both evolutionary biologists and philosophers that natural selection is a phenomenon associated with heritable differences in fitness between the members of a population (Godfrey-Smith 2009b; Sober 1984). Although all the protagonists of the debate on the causal nature of natural selection agree on this point, they regard the concept of cause and its links with fitness and natural selection in different ways. The philosophical landscape on this issue can roughly be divided up into three camps. Under what I will refer as the statisticalist view15

it is simply misleading to consider natural selection and drift as forces or causes. The statisticalists claim that what we call natural selection is a mathematical aggregate of unique events happening to individuals forming populations (e.g. Matthen & Ariew 2002; Walsh, et al. 2002). Each type of organism in a population is assigned a trait-fitness value – in its most abstract form an expected growth rate – which is the consequence of different events occurring at the individual organism level such as deaths, births or mating. Yet, those events cannot be equated to natural selection because trait fitness is a non-causal (statistical) property of types at the population level and consequently differences in trait-fitness in a population entail ENS mathematically (statistically) rather than causally (Matthen & Ariew 2009).

Bouchard and Rosenberg (Bouchard & Rosenberg 2004; see also Rosenberg & Bouchard 2005) offer a very different view, in direct opposition to that of the statisticalists. Not only do they argue that natural selection is a causal process, they also claim that fitness should not be understood as a population-level statistical property. They conceive of fitness as a relational

15 _{For different arguments from the statisticalist view see: Matthen & Ariew (2002, 2009), Walsh (2000, 2007, 2010),}

property between individual16 _{entities forming populations. This property must be distinguished}

from a growth rate or reproductive output, since Bouchard and Rosenberg view a difference in growth rate between two types of entities not as a difference in fitness, but as a consequence of a difference in fitness (see Rosenberg 1985, 158-160 for the view that fitness is measured by its effects). In other words, growth rates are proxies for fitness (a proxy of fitness is also called

realised fitness).17 _{As Bouchard and Rosenberg (2004, 710) put it: “selection [is] a contingent causal}

process in which individual fitnesses are the causes and subsequent population differences are the effects”. I call Bouchard and Rosenberg’s view the individual-level-cause view (hereafter the ‘ILC view’).

A number of authors have proposed a third view, namely that natural selection and other evolutionary processes are causes of evolutionary change, not at the individual level, as argued by Bouchard and Rosenberg, but at the population level (e.g. Millstein 2006; Reisman & Forber 2005; Stephens 2004).18 _{The main claim of the proponents of the population-level-cause view}

(hereafter the ‘PLC view’) is that since natural selection and drift can systematically be manipulated at the population level, they represent genuine causes at that level under some legitimate account of causation. An important point to note is that both the ILC and PLC proponents agree that the consequences of natural selection at are the population level. Their disagreement lies in whether the causes of natural selection are population or individual causes.

The main aim of this chapter is to introduce a new conceptual tool to this literature, namely causal modelling, and to use it to elaborate and defend the ILC view. I will also use other

16 _{In this chapter, ‘individual’ should be understood as ‘entity below the level of the population’.}

17 _{See the problem of reifying reproductive output as fitness in Chapter 5.}

concepts from the manipulationist account of causation (see Woodward 2003, 2013) to further defend this view (Section 1.4), such as the criterion of invariance (also called stability) in causal explanations developed by Woodward (2000, 2003, 2010). Before starting, I need to make a few remarks. I wrote in the opening sentence of this chapter that natural selection and drift are usually regarded as ‘forces’ or ‘causes’ of evolutionary changes by population geneticists. Yet so far, I only have used the notion of cause, and will do it throughout the remaining of the chapter while I will be less careful in further chapters, using ‘force’ and ‘cause’ indistinguishably. I make this commitment in this chapter mainly because the concept of cause, although far from being unambiguous, is less vague than the concept of force when ‘force’ is taken to be something different from Newtonian force. In fact, there are no developed philosophical accounts of what makes a ‘force’, whereas by sticking to the concept of cause I can draw upon previous work on causation and apply it to evolution.

Although the notion of cause is less vague than the notion of force, it is not unproblematic. This is reflected in the fact that there are many different competing accounts of causation in philosophy. Each of these accounts has (at least) some problems and suffers from counterexamples. In a recent survey of the literature, Hall and Paul (2013) distinguish four families of such accounts, namely regularities accounts, counterfactual accounts, probability accounts and transference accounts. Here is not the place to discuss the difference between each of these accounts. Yet this ambiguity plays a role in the debate over the causes of evolutionary change. Huneman (2012) notes that the statisticalists rely on a particular transference account of causality similar to the one proposed by Salmon (1994, 1998), according to which there is causation when there is transference of some physical quantity between a cause and its effect. In contrast, the PLC camp relies on counterfactual accounts of causation. These different

commitments allows the statisticalists to make the claim that all the causal work occurring during an event of evolution lies at the level of individual entities rather than population, and to point out that these individual events (births, deaths and reproduction) do not themselves constitute natural selection. Walsh (2000) for example considers that natural selection is a pseudo process rather than a genuine process. The PLC camp, on the other hand, claims that following the different counterfactual accounts of causation (e.g., manipulationist), natural selection and drift are causes of evolutionary change.19

Bouchard and Rosenberg (2004) do not explicitly state what account of causation they follow. They argue that distinction between natural selection and drift cannot be substantiated from a statistical perspective only,20 _{and that natural selection is a causal process relying on}

individual fitness comparisons, not on (non-causal) population level properties such as trait fitness.

Thus, for them, no notion of population level cause is necessary. Comparisons are at the heart of any counterfactual dependence and assessing whether a property or an event is a difference maker could not be made without them. In fact, to make causal claims under a counterfactual account of causation, assuming that time is asymmetric, is essentially to make a specific comparison between two situations that grants a causal relation. It involves comparing the presence/absence of c21 _{at time t}

1 with the presence/absence of e at a later time t2. In its most

general form c and e can be properties of objects, events or processes. When ceteris paribus c and e are present and had c been absent e would also have been absent, it is safe, under a counterfactual

19 _{Although see Millstein (2013) who uses a Salmonian process account of causality.}

20 _{Which is a claim clearly endorsed by Walsh, Lewens, & Ariew (2002).}

21 _{Or under a contrastive account of causation (see for example Schaffer (2005) for the general case or Northcott}

(2010) in the particular context of evolutionary theory) two different situations c1 and c2 rather than presence and

account, to claim that c is a cause of e.22 _{Thus because Bouchard and Rosenberg’s account is both}

causal and comparative, I will assume that their account relies on a counterfactual account of causation similar to the ones developed by the PLC camp but with a focus on individual properties and events rather than population level ones.

If my interpretation of Bouchard and Rosenberg’s account of causation is correct, it seems that the statisticalists initially disagreed both with ILC and PLC camps on what constitutes a cause in evolutionary theory. It is was thus not surprising that they disagreed about whether natural selection and drift are causes of evolutionary change. Although the debate between the statisticalists and the causalists is not yet settled, it seems that in recent days, the statisticalists have acknowledged the virtue of Woodward’s manipulationist account of causation in the debate over the causes of evolution (see for example Matthen & Ariew 2009). Furthermore, the concept of cause in science is very often cashed out in terms of counterfactuals (or surrogates of them such as controls, see more on this in sections 1.6 and 6.4). For those reasons, in this thesis I have chosen the counterfactual over the transference approach to causation as the relevant one for deciding whether natural selection and drift are causes of evolution.

With these important remarks in place, I will now argue that the ILC view, once developed with the help of the tools of the causal modelling literature (e.g. Pearl 2000), is superior to both the statisticalist and PLC views. To do so, the remainder of the chapter will be divided into five sections. In Section 1.2, I start by presenting some of the arguments motivating Bouchard and Rosenberg’s view that the theoretical underpinning of evolutionary theory, and

22 _{Although the ceteris paribus clause is regarded as problematic for some in the literature (Reutlinger, Schurz, &}

Hüttemann 2014), it is important here because it eliminates any potential confounding variable that could be a cause

more particularly the concept of fitness, is deterministic (or sufficiently close to deterministic) and not fundamentally probabilistic as often argued. From there, I present a slightly revised version of Bouchard and Rosenberg’s claim that fitness should be defined in causally upstream terms from reproductive output. In Section 1.3, drawing upon one of Godfrey-Smith’s (2009b) distinction, I propose a systematic method in individual rather than population terms to distinguish natural selection from other evolutionary causes, and more particularly drift. In Section 1.4, I briefly present causal graphs and depict, from a causal modelling perspective, the problem of teasing apart natural selection from drift with reproductive outputs of organisms as the only available information to the observer that is, without a notion of fitness defined in causally upstream terms from reproductive outputs. In Section 1.5, I show how combining causal graphs and the method proposed in Section 1.3 to partition off natural selection and drift in individual terms can fully explain evolutionary change at the population level. I discuss the methodological and epistemic problems inherently associated with this view, but ultimately conclude that those problems should be kept separate from conceptual considerations. In Section 1.6, I show why the version of the ILC view I propose is superior to the PLC view proposed by Reisman and Forber (2005) and Millstein (2006).

1.2. What fitness is: Bouchard and Rosenberg’s causal account of natural selection