In What Sense Can There Be Evolution by Natural Selection Without Perfect Inheritance?

ABSTRACT

In Darwinian Population and Natural Selection, Peter Godfrey-Smith brought the topic of natural selection back to the forefront of philosophy of biology, highlighting different issues surrounding this concept. One such issue is whether the perfect transmission of characters from parent(s) to offspring is necessary for evolution by natural selection (ENS). Drawing on the classical summaries for ENS, Godfrey-Smith's answer is that it is not, and opposes his view to the replicator framework. In this paper, I show that Godfrey-Smith's approach to ENS is only one of two legitimate perspective on ENS. One focuses on natural selection in the context of other evolutionary processes, while the other assumes their absence. After having presented these two perspectives, which I call the ‘contextual’ and the ‘pure’ perspective respectively, I draw on a framework which conceptualises the difference between natural selection, drift, and mutation in a causal rather than statistical fashion developed elsewhere. From there, I show that following the pure rather than the contextual perspective, perfect inheritance of characters is a necessary condition for ENS. This is because, I argue, imperfect inheritance is inevitably associated with an evolutionary process conceptually distinct from natural selection, namely mutation. I conclude by proposing that the classical summaries for ENS correspond more to the contextual perspective and the replicator framework more to the pure perspective.

Acknowledgements

I am thankful to Paul Griffiths, Peter Godfrey-Smith, two anonymous reviewers of International Studies in the Philosophy of Science, and James McAllister for feedback on previous versions of this manuscript.

Notes

1 By ‘perfect’ I mean here ‘exact copying most of the time’. Note that recently Earnshaw-Whyte (2012) has claimed, using a simple verbal model, that ENS does not even require stronger degrees of similarities between parents and offspring—that is it does not require heritability between parental and offspring characters. However, as shown in Bourrat (2015b), this claim relies on an imprecision surrounding the concept of heritability. When understood in its most abstract form, even in the model presented by Earnshaw-Whyte, I show using a version of the Price equation (see Okasha 2006, chap. 1), that there is some heritability between parent and offspring. The upshot of my analysis is that an individual persisting without change in character over time should account for part of the heritability observed in a population exhibiting variation between two times.

2 Note importantly that whether such populations exist in nature should not be regarded as more problematic for the soundness of the pure perspective on ENS than imagining a body being subjected to a given Newtonian force without frictions.

3 The evolutionary force known as ‘migration’ will not be treated here since I suppose a population isolated from any other population.

4 In a series of papers, Grant Ramsey (2006, 2013, 2016; Pence and Ramsey 2013) develops of a convergent framework. One difference though is that his approach relies on the propensity interpretation of fitness while mine does not.

5 Godfrey-Smith's view on drift is actually more complex. However, the extrinsic/intrinsic distinction is the only part of his account I need for the purpose of this article. In Bourrat (2017), I criticize his view that drift also results from what he calls ‘low continuity’, that is, when small differences in individual properties (whether intrinsic or extrinsic) have large effects on reproductive outputs.

6 For an introduction to the notion of ‘screening off’ see Brandon (1990, 83–84).

7 Note that drift is defined here very broadly. It entails what is known as ‘correlated responses’, i.e. evolutionary changes which are not causally due to the trait under scrutiny but to another known cause.

8 This assumes here that the failure does not depend on changes in extrinsic properties within the range of the extrinsic properties encountered by the species, otherwise the pattern of change is itself an intrinsic-variable property.

9 Heritability is a population level measure of inheritance from parent to offspring. There are roughly two notions of heritability used in the literature, namely broad-sense heritability (H²) and narrow-sense heritability (h²). In evolutionary theory only narrow-sense heritability is of interest and it can be defined in two different ways. Under one, h² is the ratio of additive genetic variance to phenotypic variance (Falconer and Mackay 1996). A less biologically centered definition would be ‘the ratio of intrinsic-invariable phenotypic variance in a given environment and at a particular grain of description to phenotypic variance’. Under the other, more abstract, but formally linked to the first one, h² represents the slope of regression of average offspring character on parental or mid parental character (Falconer and Mackay 1996; Rice 2004; Okasha 2006). This latter definition has been favored by many authors. I follow suit, although see notes 13 and 14. The concept of heritability is associated with a number of philosophical issues (for more details see Sesardic 2005; Godfrey-Smith 2007; Downes 2009, 2017; Bourrat and Lu 2017; Bourrat, Lu, and Jablonka 2017; Lynch and Bourrat 2017).

10 This is a more detailed version of Godfrey-Smith’s (2009, 24) model.

11 The fact I use only two individuals in my figures instead of a population composed of a higher number of individuals is for illustrative purpose only and has no consequences on the conclusions drawn later on.

12 In Case 1 there is no difference in viability between the individuals: they all die simultaneously after a unique reproductive event. That makes of Case 1 a ‘fertility selection’ example, i.e. in which selection comes primarily from difference in fertility between some individuals of the population. Yet, all the reasoning involved could have been made using a ‘viability type’ example, i.e. in which selection comes primarily from differences in viability.

13 Note here, that some have argued that the regression definition of heritability (see note 7) should not be regarded as a definition but rather as a method of estimation of heritability (e.g. Lynch and Walsh 1998, 171). In fact, arguably, because all the variation in character is due to variation in the environment of organisms in the Case 2, it should not be associated with heritability. This is a similar case to the ones of gene-environment covariance discussed in Lynch and Bourrat (2017).

14 Following on from notes 7 and 13, this suggests that the variance approach to heritability might, in some respects, be superior to the regression approach to discriminate the effects of natural selection from those of drift or correlated responses. This is mainly because the variance approach takes the independent population variable to be an intrinsic-invariable property of individuals (most of the time genes) while the regression approach takes any property (extrinsic, intrinsic-invariable or intrinsic-variable, i.e. any phenotype) as an independent variable. As such the regression approach to heritability is blind to the causes that produce resemblance.

15 ‘New’ should be understood here from the point of view the parental individuals. Unfaithful transmissions produce characters that are new and different from that of their parents. These new differences might also be new with respect to the whole population, but this is not the sense in which I am using the word ‘new’ here.

16 The idea that an intrinsic-invariable property can change by mutation might appear as contradictory. However, by ‘invariable’ one should understand here ‘invariable given a range of environmental conditions in which events changing the property are so rare that they are considered as exogenous to the environment’. This means that the relevance of making explicit the grain of description—which will fix what is meant by ‘rare’—is not limited to natural selection and drift, as argued in Section 2 and Bourrat (2019), but also to mutation.

17 It goes without saying that each approach can used for both perspectives on ENS. I am only suggesting here that they respectively fit better with one perspective rather than the other.

Additional information

Funding

This research was supported by a Macquarie University Research Fellowship and a Large Grant from the John Templeton Foundation (Grant ID 60811).