國立臺灣大學文學院哲學系 碩士論文
Department of Philosophy College of Liberal Arts National Taiwan University
Master Thesis
關於天擇本性的爭論
The Debate over the Nature of Selection
李浩德 Hao-Te Lee
指導教授：王榮麟 博士 Advisor: Rong-Lin Wang, Ph.D.
中華民國 103 年 7 月 July 2014
近期關於天擇本性的爭論關注在兩個問題上：天擇是否是因果過程，以及天擇 應如何刻畫。本論文檢視在此爭論中最具代表性的三個理論：Matthen 和 Ariew 的形式樣式觀，Bouchard 和 Rosenberg 依附於適存性概念的刻畫，以及 Millstein 不倚靠適存性概念的刻畫。Matthen 和 Ariew 主張一般所認為的作為因果過程的 天擇並不存在，天擇其實是由特定一個數學定理所代表的形式樣式。Bouchard 和 Rosenberg 認為天擇是一類因果過程，其原因為適存度關係而結果是生殖成功度 差異。Millstein 則認為天擇是一類因果過程，其原因是族群的性狀差異性而結果 是生殖成功度差異。他並認為性狀差異性是一個族群層次的性質。這三個說法都 有重大缺陷。其中，前兩個主張因為錯誤太過根本所以必須完全放棄。Millstein 的說法雖然概念上有混淆之處且形上學上也有瑕疵，但經過進一步釐清並重新詮 釋之後可以引導出一個更令人滿意的對天擇的刻畫。此一刻畫是，之於不同可定 性狀的天擇是不同類的涉及整個族群一整個世代的因果過程，每一個這樣的因果 過程類是由複數個不同但相似的涉及生物一生的因果過程類所刻畫，而後面這些 不同的因果過程類則是刻畫成同一可定性狀之下的不同類的確定性狀之對不同範 圍的確定生殖成功度有因果貢獻。
本論文結構如下。第一章介紹這個爭論的背景，區分相關的問題與無關的議 題，並指出一些基本假設以及本文的目標。第二章檢視天擇的形式樣式觀。本文 認為作為形式樣式的天擇並非如 Matthen 和 Ariew 所稱是可多重實現的，而且此 一意義下的天擇根本沒有解釋上的效用。進一步，本文反駁他們對於作為因果過 程的天擇並不存在的論證，同時釐清並重建這個意義的天擇概念。第三章處理 Bouchard 和 Rosenberg 的看法。此處將指出，在適存性確實是生物的一個性質的 假設下，適存性是一個二階功能性質，因此，說適存性對生殖成功度或其差異有 因果貢獻會招致形上學上必然相依性問題以及原因互斥問題。對於幾個試圖解決 前一個問題的方式以及對於原因互斥論證的反對此處亦將一一反駁。第四章處理 Millstein 的看法。本文澄清在他的說法裡，天擇其實並不是一類因果過程，而是 一群不同類的因果過程。此處將詳細對比依附於適存性概念的刻畫以及獨立於適 存性概念的刻畫，最終，適存性作為一個生物性質以及依附其上的對天擇的刻畫
論上是冗贅的將促使我們重新詮釋 Millstein 的看法。最終，我們可以得出一個更 令人滿意的對天擇的刻畫，它不仰賴任何形上學上可疑或存有論上冗贅的設定，
同時不會有任何語意上的扭曲或造成其他不必要的麻煩。第五章為本文結論。
關鍵詞：天擇、適存性、原因效力、多重可實現性、原因互斥論證
Recent debate over the nature of selection centres upon the questions of whether selection is a type of causal process and how selection should be characterised. Three representative accounts of selection in this debate are critically examined in this thesis:
Matthen and Ariew’s formal-pattern account, Bouchard and Rosenberg’s fitness- dependent characterisation of selection and Millstein’s fitness-independent characterisation of selection. Matthen and Ariew contend that there is no selection as a causal process as normally conceived; instead, selection is a formal pattern characterised by a mathematical theorem. Bouchard and Rosenberg assert that selection is a type of causal process identified by the fitter-than relation being the cause of difference in reproductive success. Millstein characterises selection to be such a causal process that trait-variations as population-level properties are causally relevant to differences in reproductive success. All these accounts will be shown to be seriously defective. However, the former two are so fundamentally mistaken that they will be rejected outright. On the other hand, Millstein’s account, while conceptually confused and metaphysically flawed as it stands, can be clarified and re-interpreted so that it can pave the way for a more satisfactory characterisation of selection. In this latter account, selections with respect to different trait-determinables are different types of population- wide, generation-long process each jointly identified by a plurality of distinct yet similar types of organismal-level generation-long process that are identified by different types of trait-determinates under a common trait-determinable being causally contributory to different ranges of determinate degrees of reproductive success.
The present thesis proceeds as follows. Chapter one introduces the background of the current debate, distinguishes what is at issue from what is not, and states some basic assumptions and the objective of the thesis. Chapter two examines the formal-pattern account. It is argued that selection as a formal pattern is, contrary to what Matthen and Ariew claim, not multiply realisable and has no explanatory utility at all. Meanwhile, their arguments against the ordinary talk of selection as a causal process will be refuted and the ordinary idea be clarified and re-established. Chapter three is devoted to Bouchard and Rosenberg’s account. I’ll show that, on the assumption that fitness is an organismal property, it is a second-order functional property, and therefore the assertion that fitness is causally responsible for reproductive success or the difference thereof
general objections to the exclusion argument will be rejected. Chapter four explores Millstein’s fitness-independent characterisation. It will be revealed that selection in her account is actually a family of different types of process rather than a single type of process. This leads to a comparison between the fitness-dependent characterisation and the fitness-independent one and ultimately to the rejection of the posit of the property of fitness as well as the fitness-dependent characterisation. Yet, the posit of the so-called population-level properties also has the exclusion problem; Haug’s attempt to save their causal efficacy will be criticised and rejected. Their causal inefficacy and ontological redundancy prompts a re-interpretation of Millstein’s characterisation and eventually a more satisfactory alternative that does not rest upon any ontologically redundant or metaphysically suspicious posit and does not create any semantic twist or other unnecessary complications. Chapter five is the conclusion.
Keywords: natural selection, fitness, causal efficacy, multiple realisability, causal exclusion argument
摘要 ... iii
Abstract... v
Table of Contents ... vii
1. Introduction... 1
1.1. Aspects of the Debate... 1
1.2. The Focus, the Assumptions and the Objective ... 5
2. Problems of the Formal-Pattern Account of Selection and the Possibility of a Process Characterisation ... 11
2.1. From the Force Analogy to the Formal-Pattern Account... 11
2.2. Objections to the Formal-Pattern Account of Selection... 18
2.3. Redemption of the Traditional Notion of Selection ... 29
2.4. Summary and Prospect ... 40
3. The Account of Selection as a Type of Causal Process Characterised by Fitness and the Causal Inefficacy of Fitness... 42
3.1. Fitness and the Principle of Natural Selection (PNS) ... 42
3.2. Fitness as a Second-Order Functional Property ... 50
3.3. The Problem of Metaphysically Necessary Dependency... 56
3.4. The Causal Exclusion Problem ... 69
3.5. Summary and Prospect ... 76
4. Selection with Respect to a Trait as a Type of Process and Its Characterisation ... 78
4.1. The Fitness-Free Characterisation of Selection and the Redundancy of Fitness... 78
4.2. The Causal Inefficacy of Population-Level Properties ... 91
4.3. Towards a Satisfactory Characterisation of Selection... 102
5. Conclusion ... 123
1. Introduction
It is widely acknowledged that natural selection is a cause of evolution. The most important cause for that matter. What is evolution? Population genetics formally defines evolution as the cross-generational change of allelic frequency of a population of organisms. This definition is straightforward and incurs no substantial controversy. By contrast, we don’t have an equally explicit and unproblematic definition of selection in evolutionary biology. So, what is selection? How should it be best characterised? This is where all sorts of disputes arise and is the topic of the present thesis.
1.1. Aspects of the Debate
The idea of natural selection was originally an analogy with that of artificial selection (Waters 2003). In artificial selection, the most prominent factor regarding the evolutionary change of domestic groups of organisms is human’s act of selection (selective breeding). In the wild, however, there is an absence of intentional act of selection. Accordingly, the idea of natural selection, in its nascent stage, is understood in a metaphoric way, as if it were Nature that did the selection. Natural selection is thus selection by Nature or natural environments. This is the idea of natural selection as it was explicitly presented in the early versions of the Origin (Sloan 2010).
Personified natural environments and the metaphorical sense of selection soon raised criticisms. Among them are the lack of clarity of the whole notion and the excessive non-naturalistic, intentional/agentive talks. In response, Darwin expressly adopted Huxley’s now well-known cliché that selection is “the survival of the fittest”. This way of understanding natural selection erases those charges. But it also definitely makes natural selection a sufficiently distinct notion from artificial selection. Natural selection
is no longer metaphorically conceived as an act, performable only by intelligent beings.
Rather, when using the notion of natural selection to explain evolution, the most prominent explanatory factor becomes some attributes on the part of organisms (here fitness, presumably). We have thus a completely naturalised notion of natural selection, as one may say, and therefore can do away with the initial, heuristic yet troublesome analogy.
Nonetheless, new problems arise. Since explanation (of evolution or, more narrowly, of reproductive success) by selection boils down to explanation primarily by some attributes on the part of organisms, it is undoubtedly a question of paramount importance as to what exactly these attributes are. Phrases such as “(being) fitter than”
and “(being) fit (in an environment)”, which Darwin made use of when illustrating selection, have been taken to express, or to be based upon, the property that accounts for selection. This (supposed) property is known as “fitness”. But, then, what is fitness? Is it a causal property or not? Or perhaps terms like “fitness” do not designate any property at all? Provided that selection has to be characterised in terms of fitness, the nature of selection obviously largely hinges upon the nature of fitness, and no wonder why questions like these occupy a locus of the current debate. Moreover, any account of fitness cannot fail to address the notorious tautology problem: What does, say, “being fitter than” mean? If it means “having better survival/reproductive result/success”, then the particular (albeit loose) characterisation of selection that fitter organisms have better survival/reproductive success (than less fit ones) is a patent tautology. This creates a serious problem for those who think that the notion of selection has empirical content and the notion of fitness is essential for characterising selection. Accordingly, the debate over the nature of selection also concerns how the tautology problem is supposed to be solved and, more fundamentally, whether it is indeed avoidable.
Another part of the debate has to do with whether selection is a force. Selection, along with migration and so on, has often been said to be an “evolutionary force”. This was a mere metaphor or a casual way of speaking in the first place. It was not until Sober (1984) set up the analogy with Newtonian force that the force talk took a clearer shape and got philosophically anchored. The force analogy attempts to establish the thesis that population genetics, the theory of evolution, is a theory of force, by showing that it shares the same structural elements with Newtonian mechanics. However, are the two theories substantially similar in formal terms, even if we acknowledge that they belong to different domains and most likely do not completely resemble each other?
Further, are there any reasons beyond formal similarity, or lack thereof, to Newtonian theory of force, in view of which selection and so on cannot literally be forces? And if selection is not a force, then what is it? All these also fall within the scope of an account of selection.
Now, mainstream philosophical opinions, at least before the turn of this century, are that fitness is a causal property and that selection is a force. However, from a decade ago, these views have been vigorously challenged by several philosophers, notably Matthen and Ariew (2002; 2005; 2009) and Walsh, Lewens and Ariew (2002). These authors contend that fitness is reproductive growth rate and hence is a statistical construct rather than a causal property. And they argue that selection is not a force, partly by disputing the force analogy. On these bases they go further to claim that there is no selection as a cause of evolution. Instead, selection is something else: It is the
“formal pattern” between time rates of changes in proportions of different types of components of an ensemble and those proportions over time, e.g. between fitnesses or reproductive growth rates of genotypes or trait-types in a population and their representations over generations. As a formal pattern, selection is “not even a biological
phenomenon as such” (Matthen and Ariew 2009: 222) and is completely domain- unspecific.
This characterisation of selection strikes many philosophers and biologists as unacceptable, for selection is regularly understood exactly as a cause of evolution.
Saying that selection is something else is tantamount to elimination of selection in this normal sense. As Matthen and Ariew and Walsh et al. found this view upon both a criticism of the force analogy and a non-causal account of fitness, to defend against it possible strategy to take is: (1) either insist that selection is indeed a force or, while conceding that selection is not a force, maintain that it remains nonetheless a cause of evolution, and (2) either hold that fitness is really a causal property or decouple the idea of fitness from the notion of selection. Regarding (1), most defenders of the traditional view nowadays no longer stick to the force analogy. They tend to speak of selection as a causal process, with the contents of both “selection” and a “causal process” varying with one’s choice on (2). If an account of selection as a causal process is built upon a theory of fitness, the question of what fitness is again comes to the foreground.
Alternatively, one may try to give a non-conventional characterisation of selection that does not require the notion of fitness. This line of thought brings up a whole new set of questions to consider: Is fitness essential for characterising selection? If not, how does a fitness-independent account of selection differ from a fitness-dependent one? How should selection be characterised without using the notion of fitness, then, and without engendering other problems? And does such a characterisation successfully ground the view that selection is a causal process? All of these fall under the rubric of the debate over the nature of selection.
1.2. The Focus, the Assumptions and the Objective
The present thesis deals only with the metaphysical question of what selection is.
Other controversies concerning selection are beyond our scope, which include, notably, the plausibility of selectionism, the distinction between selection and drift, and the unit- of-selection problem. The first is fundamentally an empirical issue: Taken either as an empirical thesis or as a methodological proposal, whether selectionism is plausible is determined by whether selection is (mainly) responsible for many enough evolutionary changes, which is certainly an empirical matter. The second is chiefly about drift, which will not receive explicit treatment here since we focus only upon selection. However, implicit in our discussion of selection is a minimum notion of drift (the so-called
“product view”), where drift is nothing but deviation to the most probable outcome (viz.
sampling error) of selection (when understood in terms of sampling). This is largely for the sake of argument; if one embraces any alternative notion of drift, one just substitutes
“non-typical result” or “less probable outcome” for “drift” in our sense and reserves
“drift” for whatever is intended. That would not affect our discussion of selection in any important aspect.
The unit-of-selection debate, too, is orthogonal to the present topic, although this may not be immediately obvious. The former, given a fixed content of the term “unit of selection” (say, the type of “interactor” involved in selection) and the relevant set of empirical facts about selection and evolution, concerns questions such as: Are all cases of selection accountable in terms of gene/allele as the only unit of selection? Gene/allele aside, is there unit of selection in addition to organism? Specifically, are kin group and even species also units of selection? And are issues like these purely a matter of explanation/description, or they also have certain metaphysical or ontological basis? By contrast, the current debate addresses what selection is irrespective of what the unit(s) of
selection is(/are). If gene/allele is a/the unit of selection, then we ask: Is selection a force upon genes/alleles, a process involving them or a formal pattern between the growth rates of different genes/alleles (types) in a gene pool and their representations over generations? In that case, presumably we can talk about the fitness of genes/alleles. The question naturally arises as to what such fitness is: a causal property of genes/alleles, their growth rates or perhaps something else? Also, is the notion of fitness of genes/alleles essential for characterising selection (when gene/allele is the unit of selection)? If organism is a/the unit of selection, then the organismal versions of all these questions mentioned at the end of the last section follow. And equivalent questions in case kin group and even species are also units of selection.
Nevertheless, our discussion will be set within the framework of organism being the unit of selection. Although any alternative can also do the job, this one is somewhat preferable because its adoption should introduce the least amount of unnecessary complications: For most, if not all, leading philosophers in this debate frame their arguments in just the same way. Besides, many relevant notions (especially fitness) are originated in such a framework and apply most naturally to organisms. True, these alone do not justify the thesis that organism is the (sole) unit of selection. At the very least, one may simply take it as an assumption for the sake of argument.
There are other noteworthy assumptions and unchallenged ideas to be mentioned at the very beginning. We’ll accept the current definition of evolution, even though there is no reason to insist upon its being couched in terms of generational time interval (Sober 1984) or even in terms of allelic frequency distribution. For this reason we always assume discrete populations. Biological concepts are taken in their under- elaborated versions whenever that does not affect our discussion. For example, we don’t distinguish between survival fitness and reproductive fitness; by “fitness” we mean
exclusively the latter since only it has to do with cross-generational changes. Moreover, we leave unchallenged the evolutionist tenet of “population thinking”. The basic idea of population thinking is that evolution, and hence selection, is about a population of organisms taken as a whole, as opposed to organisms taken individually. It is manifested by expressions such as “a population underwent selection” and “selection acts upon a population”. Implicated in such an idea is that the notion of selection applies only to a population. This, however, has been challenged by Bouchard (2011) on empirical grounds: He cites, for instance, the case of a single genet among his counterexamples, where the notion of a population is not applicable yet it is reasonable to apply the notions of selection and evolution to them. We leave out such cases and agree with many philosophers and most biologists that selection involves a whole population. How this is to be taken care of in the metaphysical or ontological dimension is another matter, though.
Also left unchallenged is the distinctively biologist’s causal talk, as in the statement
“selection causes evolution”. This is problematic from the view of standard metaphysics, for selection and evolution (instance) are not events in any obvious sense. One might simply reject that selection and evolution are causally related for that reason, but chances are that the seemingly metaphysically illegitimate talk is a loose and abbreviated way of expressing something in the neighbourhood of causation if not exactly causation proper. Clarification of such casual causal talks require, as may have been expected, a charitable interpretation of them and a somewhat more liberal notion of event/process. We’ll attempt this in order to preserve the insights of Darwinism, so long as no major philosophical principle is violated. An understanding of selection may (and actually will) thus result which not only can do justice to the scientific theory and practise but is ultimately accommodable in standard metaphysics. So our position is
always: To make sense of distinctively biological as well as ordinary ways of talking as far as possible, on condition that they can sit well with all relevant metaphysical principles. In cases where the two desiderata are incompatible after all, metaphysical considerations nonetheless take precedence.
We’ll soon accept that selection is a process (more refinements as we proceed).
Although we will not precisify the operative, relaxed conception of a process, it at least allows a process to span a long time, to consist of or to involve multiple objects and properties (instances), to be composed of other events/processes each of which consists of a subset of the objects and properties constituting it (and may last shorter than it), and to have no well-defined intermediate stages or events. The distinction between an event, a process (a series of events) and an interaction (an event or a process) is thus intentionally blurred, and in our discussion the corresponding terms are usually interchangeable (for “sub-selection” occurrences alone; we won’t call selection an event or an interaction). I believe that something like this is tacitly assumed by those philosophers who advocate that selection is a causal process. Now, what is it that makes a process a causal process? In our context, it hinges upon some of the properties that constitute selection being causal properties, specifically properties (the instantiations of) which causally contribute to certain reproductive results. This, too, is a framework implicated in the current debate. Again we’ll follow it, as it is also plausible by itself.
Then, we’ll take accounts of selection which support that selection is a causal process to be differing primarily in respect of the sort of (supposedly) causal properties as well as objects that are constitutive of selection. Accordingly, our main concerns are twofold:
Are those supposedly causal properties genuine causal properties (or really properties at all)? And more fundamentally, which sort of objects and properties best identify selection as a process, causal or otherwise?
Hence, our goal is towards a satisfactory characterisation of selection. We approach it by critically examining three major accounts of selection. In the next chapter we first evaluate the radical view, as propounded mainly by Matthen and Ariew (2002; 2005;
2009), that there is no selection as a cause of evolution but only selection as a generic formal pattern. While their arguments against the force interpretation are persuasive, their own conception of selection is suspicious and their exposition of that is metaphysically problematic. Besides, they don’t quite rule out the possibility of making positive sense of the ordinary idea that selection is a process and is a cause of evolution.
At this stage, as a first approximation, selection is considered a type of process involving organisms of a population within a generation, such that their differential reproductive results are in accordance with their fitness differences.
Chapter three is a criticism of the view that selection is a causal process (type) identified by the causal relation between fitness and reproductive results. This part is directed at Bouchard and Rosenberg (2004), Rosenberg and Bouchard (2005) and the propensity interpretation of fitness as well. I’ll argue that, if “fitness” expresses a property at all, that property is a second-order functional property and therefore cannot be a causal property due to the problem of metaphysically necessary dependency and the exclusion problem. Thus no fitness-dependent characterisation of selection is
compatible with the thought that selection is a causal process-type.
In chapter four we explore the fitness-independent characterisation of selection put forward by Millstein (2006). While this position is more plausible than the preceding two, the original characterisation as it stands is metaphysically confused and calls for clarification and further revisions. I’m going to elaborate upon it by first making the distinction between selection (simpliciter) as a second-order process-type and selections with respect to particular traits as first-order process-types. This contrast also helps
understand the fundamental differences between fitness-dependent and fitness- independent characterisations of selection and why the latter is more appropriate. Then I shall point out that Millstein’s taking physical differences between organisms, or variations in their traits, as the so-called “population-level properties” not only are causally inefficacious but may also be ontologically redundant. This suggests us to take a step back: to hold that the only properties on the part of organisms that are constitutive of selections with respect to particular traits are their respective traits. At the end I’ll give a sketch of a more conservative yet metaphysically more tenable account of selection with respect to particular trait. This is followed by the conclusion in the final chapter.
2. Problems of the Formal-Pattern Account of Selection and the Possibility of a Process Characterisation
The current debate is initiated by Matthen and Ariew (2002) and Walsh, Lewens and Ariew (2002). They refute the force view reinforced by Sober and contend that there is no population-level cause of evolution. Selection is then re-identified to be a formal pattern. It is further said to be multiply realised and realised even in non-biological substrates. Is this correct? And does their criticism of the force analogy preclude the possibility of a process view, which preserves the biological insights as well as makes better sense of the ordinary process-talk? This chapter aims to show that the answers are no and no.
2.1. From the Force Analogy to the Formal-Pattern Account
Sober (1984) distinguishes four structural similarities between the theory of evolution and the Newtonian theory of force: that the former features some laws or generalisations formally comparable to Newton’s principle of inertia (a zero-force law), to Newton’s second law of motion (a consequence law), to (classical) force laws like the law of gravitation (a source law) and to a principle of force combination. For that very reason he asserts that the theory of evolution is a theory of force. Further, the terms in the theory of evolution that play the similar role to the force term in the Newtonian theory are terms representing the “strength” of migration, mutation and selection. So, by analogy, migration, mutation and selection are forces with respect to evolutionary change. This is the gist of the Soberian force analogy. It obviously rests upon two things:
The two theories are indeed similar and their similarities are sufficient for migration,
mutation and selection to be forces. Most philosophers dissatisfied with it put into doubt some claimed similarities or other between the two theories. For example, Brandon (2006) argues that drift is not a force by demonstrating, amongst other things, that the supposed population-genetic zero-force law (i.e. the Hardy-Weinberg principle) is not a law and there is no zero-force law in the theory of evolution. Attacks by the proponents of the formal-pattern account, on the other hand, are directed mostly towards the (alleged) principle of force combination and the (alleged) consequence laws in population genetics. Matthen and Ariew (2002) contend that there is no principle of force combination in the theory of evolution and those population-genetic counterparts of the Newtonian consequence law are not genuine laws. We begin our discussion by examining their counter-arguments.
Matthen and Ariew hold that there is nothing in population genetics like the principle of force combination. Their main reason is that component selective factors do not combine in a way component forces do. Forces always combine linearly (as well as vectorially). The joint effect of two or more forces upon an object (at a given time and position) is the (vectorial) sum of the individual effect of each of those forces (were they present individually). This is an example of what is known as the superposition principle in physics. By contrast, selective factors do not obey the superposition principle. In statistical terms, this is because causal factors usually have “interactions”.
Since forces cannot interact in this sense, selective factors and their resultant, i.e.
selection, are not forces.
One may think that the above conclusion follows from the requirement that forces be superposing independently. This is not correct. It stands still even if we just demand that individual forces be combined in a universal way, independently or otherwise. And it is plausible to think that there is no universal principle for combining selective factors, as
there is no reason to believe that all selective factors interact with each other in the same way. Thus, the attempt by Stephens (2004) to save the force analogy by stipulating that non-Newtonian forces need not combine additively is in vain. This move is not only ad hoc, but misses the real point of the argument. One may further drop the requirement of
a universal principle of force combination, but then how is the notion of a force supposed to be distinguished from that of a cause? Such a generic, non-physical notion of force is too thin to be significant. It appears that selective factors are called forces merely because they can be conceived as if they pushed, pulled and balanced out. No causal factors cannot be so conceived, however. And we just have to admit that the force talk is, at bottom, a matter of metaphoric use of words.
So, we can accept that this argument successfully establishes that evolutionary factors, or at least selection and selective factors, are not forces. At the very least, the notion of force is superfluous in the theory of evolution. Yet it leaves open the question whether selection is a cause of evolution. This question is addressed, partly and partially, by the argument against there being population-genetic consequence laws. Both Sober and Matthen and Ariew take as relevant in this regard those population-genetic equations between allelic frequencies at the next generation and those at the current generation weighted by coefficients representing the strengths of migration, mutation and selection. For Sober, these are the consequence laws in the theory of evolution. For Matthen and Ariew, however, they are mathematical theorems and ipso facto are not (empirical) laws. And selection is said to be characterised by Li’s growth-rate theorem, which is a population-genetic theorem of that ilk. It is, therefore, a formal pattern, and a formal pattern is not among the right sorts of things that can be causes (e.g. events and property-tokens). Hence, selection is not a cause of evolution. Now, how plausible is this second argument?
I agree that, like Li’s growth-rate theorem, the population-genetic equations containing only frequency variables and coefficients representing the strengths of selection are theorems. They key, of course, lines in the semantic content of those coefficients. They are exactly what are routinely called “fitness coefficients/parameters”
or simply “fitnesses” in population genetics. By both Matthen and Ariew’s interpretation and the standard population-genetic definition (e.g. Futuyma 2009: 306), fitness is basically the cross-generational time rate of change (owing to reproduction) of the frequency or proportion of a (sub-)type of organisms in a population. Population- genetic equations and theorems of the aforementioned family follow solely from the meanings of “fitness”, “(number) frequency” and so on, as well as from the syntax of relevant mathematical operations. They are a priori truths. Since they lack empirical contents, they are not empirical laws. So far, so good.
Yet they go on to claim that selection is characterised by such a theorem. This is where things start going weird. Li’s theorem, which is a simplified version of Fisher’s fundamental theorem of natural selection, states that “in a subdivided population, the rate of change in the overall growth rate is proportional to the variance in growth rates”
(Matthen and Ariew 2002: 72). It surely states a formal pattern. It’s no doubt a truth out of verbal necessity and hence not a law. It definitely cannot be a cause. But, does it represent the notion of selection?
It is the predicament associated with the current debate that there is in fact a plethora of notions of selection among philosophers. When different authors disagree upon the nature of selection, it is likely that they are actually talking about the nature of different things. So, a justification of the adequacy or advantage of a proposed conceptual content of “selection” is no less important than the correct metaphysical categorisation of what a given notion of selection is about. An account of selection as radical as the formal-
pattern one, in contrast to the more conventional ones in which selection is a cause of evolution or a causal process, requires extra efforts to justify. Matthen and Ariew are certainly aware of this. They attempt to justify it along three lines. These are outlined below in the order of increasing importance.
The first has its roots in the view that population genetics is the theory of natural selection. Population genetics is concerned with mathematically representing all sorts of (biological) evolutionary phenomena and exploring possible as well as ruling out impossible evolutionary consequences with the aid of involved mathematical/statistical techniques. It is held to be different from, albeit complementary to, field/lab studies of evolution: It is said to give explanations of evolution in terms of fitness whereas the latter provide explanations of evolution in terms of traits and environmental conditions.
In other words, it yields “explanation by fitness” but doesn’t offer “explanation of fitness” (Byerly and Michod 1991a); that is, it doesn’t address why a given type of organisms of a certain population in a given environment has the fitness (i.e.
reproductive growth rate) it does. Since population genetics is the theory of selection, what it says most generally about selection (i.e. Li’s theorem or Fisher’s theorem, together with the equations containing only fitness parameters and frequency variables), and nothing else, are about selection proper.
The second centres upon the universal applicability of the notion of selection. The fact that the general statements about selection in population genetics are mathematical theorems, and the interpretation that “fitness” means, and can only mean, “reproductive growth rate” in population genetics, are two sides of the same coin. And both of these are further connected to the idea that the notion of selection applies to populations of all sorts of organisms. If “fitness” meant anything else, the notion of selection would not be applicable to all types of organisms across the board. Nevertheless, Matthen and Ariew
do recognise an alternative notion of fitness which they call “vernacular fitness”. The latter is precisely the Darwinian notion of fitness which appears in the cliché “the survival of the fittest”. They accept that it expresses an organismal property that is tightly connected to traits, and Walsh, Lewens and Ariew (2002) even conceded that that property is a causal property in regard to an individual organism’s reproductive success. Yet, they all insist, the Darwinian notion of fitness does not enter into the population-genetic theory of evolution. Consequently, it is not part of the notion of selection.
Now, Matthen and Ariew carry further the generality of selection, for the growth-rate theorem applies not only to populations of organisms. It applies to any sub-typed ensemble regardless of what sort of things/events constitutes it or whether, and how tightly, its constituents are spatially or causally bound together. Also, the constituents of an ensemble need not literally reproduce; all that matters is that the number proportions of the different types of its constituents can change over time for whatever reason. Thus, what enters into the growth-rate theorem is neither the Darwinian fitness nor the population-genetic fitness; it is the more general notion of “time rate of change in number proportion”, a.k.a. “growth rate”. This generic notion as well as the growth-rate theorem itself is required to characterise selection proper, because “[s]election also occurs in nonbiological realms: in the economic domain, for example, as well as in
“clonal selection” in the mammalian immune system, in classical conditioning, and, according to some, in the propagation of theories and other cultural artifacts” (Matthen and Ariew 2002: 71). For that matter, selection is said to have “multirealizability” and is
“realized in many substrates”. It is itself “not even a biological phenomenon as such”
(Matthen and Ariew 2009: 222). It is “wholly abstract, then, but its realizations are shaped by concrete relations—these concrete relations are what determine the value of
the abstract parameters of natural selection” (loc. cit.). Matthen and Ariew consider this characterisation of selection better than its competitors for the reason that it has universal applicability and covers whatever is called selection, especially those outside the domain of biological evolution.
The third line of justification is their direct rebuttal of the traditional claim that selection is a cause of evolution. At the heart of this rebuttal is the “constitution thesis”:
“Ensemble-level selection events are constituted without remainder by individual-level selection events; consequently, the causes of ELSEs are the causes of the ILSEs that constitute them. Thus, ELSEs are wholly caused by [the causes of] ILSEs.” (Matthen 2010: 2) In the domain of biological evolution, an ELSE is just an evolutionary change, whereas an ILSE is the birth or death of an individual organism. This thesis is plausible.
Yet it entails that there is no population-level cause of evolution that is to be identified with selection. Colloquially, one may say that the causes of births and deaths of individual organisms, e.g. predations and matings, are causes of evolution. But this is not metaphysically adequate. The correct way is to say that they (severally) cause those births and deaths, and the latter (collectively) constitute an evolutionary change. That is, there is no cause of evolution: It is an epiphenomenon. There are certainly causal processes that are explanatorily relevant to evolution, but they are to be identified only with those processes involving individual organisms.
But what about statements such as “variation in wing darkness caused the evolution of the moth population”? While Matthen and Ariew (2009) admit that it is a statement about a causal relation in the sense of cause as probability raiser, they refuse to acknowledge that it means that selection caused the evolution of the moth population, for “variation in wing darkness” doesn’t mean the same thing as “selection” does.
According to them, saying that selection is a cause of evolution is reifying selection. It
amounts to positing a tertium quid called “selection” over and above, say, variation in wing darkness and evolutionary change, such that it (arises from the former variation and) acts upon the moth population to produce the latter change. Evidently this tertium quid is explanatorily redundant. Therefore, they claim, from the ontological
consideration there is no selection as a cause of evolution.
Thus are their justifications of the formal-pattern account of selection. Even though there are many insights in them (which will be picked up in due courses), they have serious problems that they fail to justify it after all. The first line, in itself, manifests an unduly preoccupation with mathematical representations and statistical reasonings to the exclusion of empirical studies. Moreover, there are population-genetic equations that are by no means mathematical theorems but nevertheless are thought to model selection.
There are no convincing reasons why they are not about selection. So, the first line of justification is very flimsy by itself. It has to be backed up by the other two in order to maintain that those fitness-free equations are not about selection proper and that they cannot be taken to mean that selection causes evolution. This suggests that what really do the supposed justificatory work are the concern about generality and the argument against the existence of selection as a cause of evolution. These latter two, on the contrary, are more substantial and demand more elaborated treatments to show that they are mistaken. To these we now turn.
2.2. Objections to the Formal-Pattern Account of Selection
This section attempts to refute the formal-pattern account of selection by arguing against the second and positive line of its justification outlined above. We specifically focus upon the metaphysical character of selection as a formal-pattern and the
pattern multiply “realisable”, as declared by Matthen and Ariew? As we shall see immediately, saying so betrays a confused metaphysics, which eventually can be traced back to what is wrong with the formal-pattern account.
To answer the question, a clarification of the notion of realisation is required. As a philosophical jargon, the term “realisation” makes its debut along with Putnam’s introduction of machine functionalism into philosophy of mind, where it is used to convey the idea that “a physical computing machine “makes real” or “brings into concrete reality” an abstractly characterized Turing machine, a mathematical entity”
(Kim 2010: 103). In its current use, however, it is more closely associated with role functionalism. The latter dictates that mental properties are functional/causal “roles”, i.e.
they are functionally/causally identified or defined. In the physicalist framework, a functionalist mental property, being a functional/causal role, is said to be played, occupied or “realised” by some physical property or other. And the realised (functional) property is also said to be a second-order (functional) property, i.e. the property of having some (first-order) property or other that satisfies a given functional role. Thus, realisation as it is now understood is a relation between properties, and the standard talk of realisation is tied to the second-order view about functional properties (op. cit.).
Given this standard notion of realisation, selection as a formal pattern is not multiply
“realisable”. For what is realisable is a property, not a pattern. Moreover, realisers have to be properties too. But Matthen and Ariew seem to take the category of objects or just things to be the realiser in this context, as they speak of selection as being realised in many “substrates”. So, saying that a formal pattern is (multiply) realisable in or by (different types of) objects or things is making category mistakes and is misusing the notion of realisation, at least as it is currently understood.
Then, can we say that selection as a formal pattern is realisable in different things in
the same sense as “an abstractly characterized Turing machine, a mathematical entity”
is realisable by different “physical computing machine”? Not really. First of all, in the latter case realisers are concrete things, i.e. physical objects. Contrarily, the supposed realisers in the current case include “things” like one’s account balance, ideas, theories and cultural practises. These “things” are certainly not physical objects. In order for the original philosophical notion of realisation to be correctly employed, at least the possible realisers must be confined to the physical realm. Suppose we do so. Then there are further problems concerning the realised. In the case of machine functionalism, the realised is an entity. This can be understood in two ways: First, it is an abstract entity, i.e. a universal, and then realisation is nothing but instantiation in a universal ontology.
Second, it is just a property, say, being a mental thing, for which that universal is posited, and therefore realisation is merely the ontologically noncommittal type-token relation. We can simply take up the second understanding because the choice of ontology is a topic independent of the talk of realisation. Now, does realisation taken as the type-token relation apply to our case, since the realised is a formal pattern of change rather than a property? Provided that some modifications are made, yes, it does. The talk of type and token/instance isn’t restricted to objects; it applies to changes, i.e.
occurrences, as well. If a type is a pattern of change, its instances are particular changes exhibiting that pattern. Consequently, the instances or “realisers” in our case are not the particular physical objects involved in or partially constituting the particular changes described by the said mathematical theorem; instead, they should be the particular occurrences themselves. So, saying that a formal pattern of change is multiply realisable is shorthand for saying that it is a type of change of which many particular changes are instances and its different instances can involve objects of different physical types, if realisation is just the type-token relation.
But this isn’t the whole picture. Even if realisation in the context of machine functionalism is taken basically as the type-token relation, it is a special one: It applies to all and only those cases where the type is a functional one, or otherwise the talk of
“multiply physical” realisation loses all its significance. This is presumably a reason why realisation is now understood as a relation between a functional property and a physical property. The key question to ask in our case is thus: Is a formal pattern of change a functional type? One might think that it is, for its different instances do involve physical objects of different types, and, like an abstractly or mathematically characterised Turing machine, it is “wholly abstract” and is characterised by a particular mathematical theorem. The truth is that it is not, however. A functional specification certainly abstracts away from all sorts of physical details: It undoubtedly contains terms that do not designate physical properties or state-types, and functional patterns of changes/transitions between property- or state-instances surely can be couched in terms of syntactic or mathematical operations. But not all formal patterns of changes so characterised in are functional patterns. It depends upon whether the characterising operations are over terms that designate functional properties or state-types. If not, the formal patterns are not functional types; they have instances or instantiations (in the ontologically noncommittal sense) but not realisations. As shown below, the terms in the mathematical theorems that Matthen and Ariew think represent selection proper do not designate functional properties and thus the formal patterns characterised by those theorems are not functional patterns.
The terms of the growth-rate theorem are mathematical functions of number growth rates. Also, the terms used in the basic selection models in population genetics are number growth rates (of genotypes) and number proportions (of allele types). Number proportions are obviously quotients of number counts, and number growth rates are
essentially differences of number counts (since the only meaningful time interval is one time-unit). Thus, underlying the mathematical theorems that characterise selection construed as a formal pattern of change is the sole concept of number count. It is not in any intuitive sense a physical concept. Yet it is not a functional concept either: It is not a concept defined as a role in a functional specification, and it is not such a concept that all the objects belonging to its extension have a certain common function or common set of causal powers. Objects’ belonging to the extension of “(having) number count of one” are so not in virtue of their falling under a common functional description; they are so just in virtue of their being each individual objects under whatever principle of individuation. And a collection of 1,001 objects’ belonging to the extension of “number count of 1,001” is so just in virtue of its being a collection of 1,001 individual objects.
In fact, the concept of number count applies not only to every object and every collection of objects. It applies to occurrences as well (which Matthen and Ariew tacitly accept, as they agree that selection also occurs in conditioning, i.e. in sets of behaviour- tokens), since events and processes are themselves also countable. All the more, it applies to non-physical “things” such as ideas and theories. This strongly suggests that the concept of number count doesn’t even pick out a type/kind; that is, it doesn’t designate any property at all. If so, it should remain a mere concept/predicate.
Accordingly, there is no multiple realisability of a functional property named “number count”. If anything, there is only the multiple, or literally universal, applicability of the concept of number count.
In so far as the concept of number count does not designate a property, all concepts mathematically derived from it alone do not designate properties either. These include those of number (count) proportion and number (count) difference/change, i.e. growth rate, in our case. As a result, any mathematical equation wholly composed of such terms,
which in fact states a conceptual relation between themselves, does not express a pattern of change between property-instances when it is used to describe particular occurrences, phenomena or histories. It has multiple or even universal applicability just as are all its constituent, number-count-defined concepts. But again, there is no multiple realisability of a pattern of change in the sense of a functionally specified occurrence-type having as its instances particular occurrences that are instances (also) of different physically characterised occurrence-types. For what are describable by such a theorem do not even collectively constitute a type of occurrence, let alone a functionally specified one.
Therefore, selection as construed in the formal-pattern account is not multiply
“realisable”.
This antithesis is not going to be defeated in the following ways. First, one may try to limit the scope of application of such a theorem so that what are describable by it do collectively constitute an occurrence-type. The problem is: how? Trivial cases, i.e.
single-member collections, collections of a plurality of exactly similar things and, generally, no changes/differences in number proportions over time, can be omitted for sure. Differences in number proportions of two unrelated collections may be excluded too. But any constraint beyond these is ad hoc. There is nothing inherent in the concepts of collection, subclass (within a collection), number proportion and change/difference in number proportion or in the growth-rate theorem itself and like equations that confines any such theorem to any specific type of things, collections, partitions or occurrences.
Moreover, even if, by stipulation, it is restricted to populations of organisms and trait- partitions and the concept of growth rate restricted to reproductive growth rates of trait- subclasses, what are describable by it still do not collectively constitute an occurrence- type. For traits are not a single property, and by extension occurrences characterised partially in terms of number proportions and reproductive growth rates of trait-
subclasses of a population of organisms do not form a single type. It follows that even limiting ad hoc the range of application of the growth-rate theorem to the domain of biological evolution cannot make selection as construed in the formal-pattern account something that can have realisations.
Alternatively, one may be willing to recognise “arithmetic types/kinds” and hence arithmetic or formal properties. In that case, number counts and number proportions are formal properties of object-collections (suppose we’re talking only about objects and also ignore trivial cases). Then, occurrences describable by an equation such as the growth-rate theorem may be said to constitute a formally characterised occurrence-type, since any of the terms composing such an equation either designates a formal property or is based upon some such properties. But do these permit one to say that selection construed as a formal pattern of change, now a formally characterised occurrence-type, is multiply realisable by different physically characterised occurrence-types? Absolutely not, because formal properties and occurrence-types, while being themselves types, are not functional types, and ipso facto are not realisable. And, given that the concepts of number count, number proportion and change in number proportion are universally applicable, any of the supposed formal properties and occurrence-types they pick out is instantiable concurrently with any single physical property or occurrence-type (and even with any set of physical properties or occurrence-types when the object-collections are arbitrary or inhomogeneous ones). This plainly renders pointless all the talk of realisation (and supervenience as well) in the case of such formal types.
Lastly, one may think of growth rate as a property of an object-collection that
“causes” the change in its number count/proportion in a certain way (in unit time), or a disposition/propensity of it to change in number count/proportion in a certain way. Such a property, disposition or propensity can indeed be considered functionally specified
and thus can be said to have physical realisers, which are certain physical properties of the object-collection or of the objects of the collection. Then selection understood as an occurrence-type partially characterised by such a property may be said to have physical realisations. This line of thought is very akin to Bouchard and Rosenberg’s view about selection. However, if one thinks about growth rate in this way, one is not thinking about growth rate as such and is really not talking about Matthen and Ariew’s account of selection. Growth rate is about the change of number count/proportion; it is not about any cause of, or causal property in relation to, that change. Selection and the properties characterising it may be said to have multiple realisability in a different account of selection. But that is totally another matter.
So, Matthen and Ariew confuse the universal applicability of a mathematical theorem with the multiple realisability of a functionally specified type. The correct understanding of the metaphysical character of selection as they construe it, however, strongly suggests that the sort of universal generality with which they identify selection has little to no explanatory utility. It’s difficult to imagine an occasion in which one would find that, given the information about the (constant) number growth rates of the various subclasses of an object-collection, (the decrease in) the time rate of change in
the overall number growth rate would require an explanation at all, or at least would be explained by (the decrease in) the variance among those subclass number growth rates.
For any (discrete series of) change in the number count/proportion of any object- collection, the set of the number growth rates of its subclasses and the time rate of change in the number growth rate of the whole collection are two different arithmetically/statistically partial- or under-descriptions of the same change, and the variance among those subclass number growth rates is yet another, more informationally reduced statistical under-description that is mathematically entailed by
those subclass number growth rates. To be sure, all these complex terms are derivationally related. But they don’t seem to be able to stand in explanatory relation with each other, because they are just different arithmetic/statistical under-descriptions of one and the same change. At any rate, none of them could explain another in the empirical sense of explaining an occurrence by its cause or explaining the instantiation of a property by the concurrent instantiation of a realiser or a micro-base.
Similarly, at the type-level, it’s highly doubtful that one would ever find the formal patterns of change at issue in need of explanation. Each of those patterns is at bottom a conceptual relation between two different ways of arithmetically/statistically under- describing a single change, which in turn boils down to a matter of mathematical definitions of the describing terms themselves. This is the only reason as to why those patterns hold (and hold necessarily). If this counts as an explanation, then so be it. But it is evidently not an empirical explanation. The simple relation between growth rate and number counts/proportions is purely definitional; it requires no explanation and itself explains nothing in the empirical sense of explanation. It is perhaps because it is so blatantly trivial that Matthen and Ariew do not directly identify it with selection. On the other hand, the conceptual relation expressed by the growth-rate theorem might not seem so trivial, yet describing changes in terms of it is of little epistemic significance.
Since such a description is essentially the application of a mathematical theorem; it is non-empirically and necessarily true of the occurrences described. For this reason, it neither elicits nor furnishes explanations and so cannot further our understanding of the described changes. You can devise a descriptive pattern such that it is void of any empirical content and hence is necessarily true, universally applicable, completely domain-generic and what not. But then you also take away every bit of its explanatory utility and epistemic significance.
By saying that the formal pattern that holds between the time rate of change in overall number growth rate and the variance among subclass number growth rates neither explains nor is in need of explanation, that those two arithmetic/statistical under- descriptions don’t stand in explanatory relation, and that they themselves don’t call for explanation given the set of subclass number growth rates, however, I don’t mean that there’s nothing to be explained about the changes so described. Those number growth rates, i.e. the changes in number counts/proportions themselves, obviously require explanations. We’ll want to ask why the subclasses’ number proportions change in the way they do, as well as why the whole collection’s number count changes in the way it does. We’ll also want to know why the subclasses are divided in the way they are, that is, what grounds the way(s) of partitioning the collection based upon which the given under-descriptions are made. All these are empirical questions and have to be answered by supplying additional information beyond informationally reduced under-descriptions of the changes one after another. Yet, in Matthen and Ariew’s account, selection is not a matter of changes in subclass number proportions of an object-collection; it is a relation between two certain arithmetic/statistical features of the same change in the number count/proportion of an object-collection (with another such feature of that change being a set of changes in subclass number proportions). Consequently, explaining the changes in subclass number proportions is not explaining selection, and using subclass number growth rates to predict long-term trend or, perhaps, to explain it in a wider context is not using selection to predict or explain. Selection as construed in the formal-pattern account is such a useless notion that we can totally dispense with it without affecting our understanding of the world. And precisely this is what is fundamentally wrong with the formal-pattern account of selection.
This conclusion has an interesting and important implication. It implies that it is
pointless to extend the notion of selection as used in the studies of biological evolution so that it becomes applicable to all sorts of changes across the board that merely share a common necessarily true arithmetic/statistical (under-)description and hence becomes completely domain-generic. The extended notion is useless in so far as it has no explanatory work to do. And its supposed merit of being domain-generic is superficial and delusional. The commonality among the changes that fall under such a description is neither physical nor functional; it is non-empirical and merely linguistic. Both of these strongly prohibit the positing of any corresponding properties and occurrence- types. Accordingly, extending the original notion of selection from the domain of biological evolution to other domains does not go beyond the realm of analogy. And a purely verbal analogy resting solely upon a formal similarity for that matter.
Note that this view is not assumed in the first place. I didn’t argue against Matthen and Ariew’s account of selection by criticising that they are not really talking about selection or that they are merely talking about an analogy with selection. Before any substantial argument, it is largely a preconception to think that the notion of selection (proper) concerns biological evolution alone. Rather, I rejected their notion of selection on the grounds that it doesn’t pick out an occurrence-type and cannot play any explanatory role. These provide convincing reasons for accepting that it can only remain an analogy-based idea in relation to the original notion of selection. However, none of these implies that the latter notion is more tenable. Recall that the formal-pattern account of selection is partially founded upon a rebuttal of the ordinary talk of selection as a cause of evolution. If that is a successful rebuttal, then perhaps there is no defensible notion of selection and we may have to recognise that the whole talk of selection is misguided. If it is not, only Matthen and Ariew’s account collapses and we should go on to explore other accounts of selection that stick to the conventional idea.
2.3. Redemption of the Traditional Notion of Selection
Traditionally the notion of selection is specific to the domain of biological evolution:
It is about a cause of evolution. In addition, most philosophers as well as biologists nowadays also speak of selection as a process (if not a force). But, every case of evolution (by selection), and hence every case of selection, is wholly constituted by organismal-level occurrences, i.e. processes/events each involving only some (not limited to one), but not all, of the organisms of a population. According to Matthen and Ariew, this constitution thesis speaks against the reification of the term “selection”: It renders redundant the posit of something called “selection” which is over and above the organismal-level occurrences in a case of what is called “evolution by selection” and which caused the given evolutionary change. Further, although “variation in wing darkness caused the evolution of the moth population” is an acceptable causal claim, it doesn’t have the same meaning as “selection caused the evolution of the moth population” and doesn’t talk about certain process. Can the traditional idea of selection survive these challenges? Yes it can. Below we’ll see that the conventional talk can be understood in a way that is consistent with the constitution thesis. In addition, a careful assessment of the above attacks will reveal that they themselves rest upon several mistakes so that they pose virtually no threat to the traditional notion.
We start by establishing the grounds for reifying “selection” in the ordinary sense of a process such that variations in certain features amongst a population of organisms lead to the variation in their reproductive success. This latter idea is nothing but the Huxley- Darwin cliché and Lewontin’s well-known HVF formula (that the condition for selection is heritable variation in fitness) precisified and neutralised (by leaving unspecified what exactly those features are). It is also what underlies the two accounts
