Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Biology and Philosophy (2005) DOI 10.1007/s10539-005-9021-7 Ó Springer 2005 John Maynard Smith and the natural philosophy of adaptation ALIRIO ROSALES Facultad de Humanidades y Educación, Escuela de Filosofı´a, Universidad Central de Venezuela, Caracas, Venezuela; Author for correspondence (e-mail: [email protected]) Received 20 March 2005; accepted in revised form 18 August 2005 Key words: Adaptation, ESS, Evolutionary game theory, Evolutionary stability, Optimality, Phenotypic evolution Abstract. One of the most remarkable aspects of John Maynard Smith’s work was the fact that he devoted time both to doing science and to reflecting philosophically upon its methods and concepts. In this paper I offer a philosophical analysis of Maynard Smith’s approach to modelling phenotypic evolution in relation to three main themes. The first concerns the type of scientific understanding that ESS and optimality models give us. The second concerns the causal–historical aspect of stability analyses of adaptation. The third concerns the concept of evolutionary stability itself. Taken together, these three themes comprise what I call the natural philosophy of adaptation. Introduction ‘This book is about a method of modelling evolution, rather than about any specific problem to which the method can be applied’. Thus began the first chapter of Evolution and the Theory of Games, and it expresses the side of John Maynard Smith’s contribution to evolutionary biology that I want to explore in this paper. In developing novel ways of modelling phenotypic evolution, Maynard Smith was simultaneously concerned with explaining adaptation and evolutionary change, and with the philosophy of biological modelling. He was thus both a scientific theoriser and a natural philosopher of evolutionary biology. The aim of this paper is to examine the philosophical underpinnings of Maynard Smith’s work on evolutionary modelling, and to explore their ramifications. The structure of the paper is as follows. I examine Maynard Smith’s views on models and explanation in relation to three broad themes. The first is the role of optimization and game theoretical models in explaining phenotypic adaptation. Following Maynard Smith, I argue that these models facilitate understanding by locating the actual in a space of possibilities. The second theme is the causal–historical dimension to adaptationist theorising. I argue that stability analyses of adaptation are historical in something like Peter GodfreySmith’s ‘Modern History’ sense (Godfrey-Smith 1994). The third theme concerns Maynard Smith’s concept of evolutionary stability. I argue that to achieve a dynamical understanding of adaptation, evolutionary stability à la Maynard Smith is not enough; evolutionary attainability is also critical. Modelling adaptation Theorising about adaptation did not proceed by way of building dynamical models of change, as was the case in population genetics. Rather, it required the development of novel interpretive concepts and the importation into biology of mathematical methods from engineering and economics.1 One key concept was that of an optimal adaptive strategy, later generalized in Maynard Smith and Price’s (1973) concept of an evolutionarily stable strategy (ESS). The mathematical methods included optimization theory and the theory of games. For Maynard Smith, these methods provided a novel way of understanding biological structures and behaviours, and the diversity of life itself (Maynard Smith 1978a, b). The ESS concept was ‘useful in interpreting animal contests’, Maynard Smith wrote, where ‘interpreting’ meant accounting for the selective forces responsible for the evolution of the observed behaviours (1979, p. 475). More generally, Maynard Smith argued that to theorise about adaptation, one needs to ‘start out with a set of formal categories in terms of which one attempts to interpret phenomena’ – such as an ESS (1987, p. 119). Such categories allow the theorist ‘to explain phenomena in terms of existing theory’ or to see that a given class of phenomena ‘does not fit existing theory’ (ibid. p. 119). Put differently, these categories are necessary to mediate the application of theory to the world. As the opening quotation above illustrates, optimization and ESS models provided a general framework for investigating the selective pressures shaping phenotypic evolution, rather than hypotheses about particular evolutionary histories.2 One important aspect of optimality modelling, emphasised repeatedly by Maynard Smith, is the need to impose constraints on the set of possible phenotypes on which the selective process operates. He wrote: It is clearly impossible to say what the ‘best’ phenotype is unless one knows the range of possibilities. If there were no constraints on what is possible, the best phenotype would live forever, would be impregnable to predators, would lay eggs at an infinite rate, and so on. It is therefore necessary to specify the set of possible phenotypes, or in some other way describe the limits of what can evolve. (Maynard Smith 1978a) 1 These methods include calculus, optimal control theory, game theory and theories of comparative statics in mathematical economics. It is interesting to note that the development of the ESS concept brought about theoretical novelties within both evolutionary biology and game theory itself (cf. Weibull 1985). Day (2005) provides an interesting review of the main modelling strategies in theoretical evolutionary ecology, including a discussion of evolutionary game theory and its relation to the novel approach of adaptive dynamics. 2 Thus Maynard Smith wrote of the Hawk-Dove game that its goal is not to represent any particular situation, ‘but to reveal the logical possibilities inherent in all contest situations’ (1982, p. 6). Similarly, Peter Abrams writes that optimality models are ‘aids for understanding adaptive scenarios, rather than precise predictions about the outcome of evolutionary processes in specific systems.’ (2001b, p. 274) and Whenever an optimisation or game theoretic analysis is performed an essential feature of the analysis is a specification of the set of possible phenotypes from among which the optimum is to be found. (Maynard Smith 1982, p. 5) The point made by Maynard Smith in these quotations has important implications for the sort of scientific understanding produced by optimality models. An optimal phenotype is not simply the best there is in a population, but the best alternative given a specified range of possibilities. So optimality models, either explicitly or implicitly, must specify a space of biological possibilities from which the optimum is to be selected. To the extent that such models make qualitatively correct predictions about the world, they explain actually observed outcomes by locating them with reference to a set of alternative outcomes that might have evolved, but did not. Thus there is a crucial modal dimension to the scientific explanations that optimality and ESS models yield. This modal dimension is something about which most philosophical accounts of explanation say little, but it plays a central role in the logic of optimality reasoning as described by Maynard Smith. For him, ‘scientific theories say nothing about what is right, but only about what is possible.’ (Maynard Smith 1989, p. 50). Of course, models play a central role in some accounts of scientific explanation, e.g. Cartwright’s ‘simulacrum’ account of explanation and Hughes’ notion of a ‘structural’ explanation (Cartwright 1983; Hughes 1993). However, neither Cartwright nor Hughes addressed the modal dimension emphasised here, which is crucial to the type of scientific explanation that ESS models yield. More recently, Anya Plutynski (2004) has offered a pragmatist account of explanation in classical population genetics in which the modal aspects of modelling are fully taken into account. She argues for a context-dependent explanatory role of models in terms of ‘proofs of possibility’. Plutynski’s modal view of populationgenetic models can be seen as complementary to the modal account of optimality and ESS models offered here. Such models do not establish proofs of possibility but comprise a framework for the theoretical understanding of adaptation. Interpreted as optimal phenotypic configuration, adaptation does not make sense unless it is embedded in a constrained space of phenotypic possibilities. In his essay, Understanding Science, Maynard Smith argues for what might be called a ‘structural’ conception of scientific understanding. He recalls learning that a time delay in a control system increases the likelihood of oscillations in the system (Maynard Smith 1989). So for any oscillating system, be it the menstrual cycle or the number of hares and lynxes in Canada, it is appropriate to look for a delayed feedback. The underlying assumption here is that structure determines behaviour – it is the relations among the components of the system that matter, not the sorts of objects involved. Understanding the behaviour of the system involves understanding how the behaviour is determined by the relations that hold between the parts. Typically the behaviour is described mathematically, by means of a differential equation, but understanding the equation is not enough: If I claim to understand the behaviour of some system, I mean rather more than that I understand the mathematical description of it. I mean that I can in some way analyse it and play with it in my head, imagining how it would behave in various circumstances. For want of an alternative, this ability can be described as having a ‘physical intuition’ about a system. (Maynard Smith 1989, p. 227) Physical intuition is complementary to the mathematical analysis of the system. Now, analysing a system and playing with it in one’s head amounts to constructing theoretical scenarios – models – in which one investigates how the system behaves under different idealising assumptions.3 One explores how robust the relational structure is under different conditions. So the features essential for scientific understanding are structural features that are represented in models and equations. What becomes of this structural conception of understanding in the context of modelling adaptation? I suggest that structural understanding obtains once the set of possible phenotypes, discussed above, is coupled with a fitness function relating phenotypic performance to selective success in a given environment. For in effect, the functional relationship between fitness and trait value constitutes a relational structure. So again, it is the relations among the components of the evolving system, i.e. the variant phenotypes and the environment, not their intrinsic features that determine the system’s dynamics. We arrive at the following view: theoretical understanding of adaptation is achieved by locating actual phenotypes in a space of biological possibilities, constructing a fitness function modulo a given environment, and exploring the resulting evolutionary dynamics. Adaptation is understood through the fitnessphenotypic performance-environment relational structure. The understanding achieved this way is phenotypic in that there is no recourse to explicit genetic models of the traits in question. Maynard Smith has pointed out that this way of thinking has an ‘obvious justification in analysing the evolution of traits of whose genetics we know little or nothing’ (Maynard Smith 1987, p. 128). This is a common justification of the approach, but it can potentially obscure the role that optimality and ESS models play in providing the sort of theoretical understanding of adaptation envisaged above. Optimality and ESS models provide understanding of how fitness varies with phenotypic performance in a given environment (which includes other organisms), and how this affects the evolutionary dynamics. Such understanding would be indispensable even if the genetical basis of the phenotypes were 3 The distinction between physical intuition and mathematical analysis mirrors the distinction between the physical and formal–structural notions of model, emphasized by Suppes (1966). See also Morgan and Morrison (1999), and French and Da Costa (2003). known; it is not the sort of understanding that population genetical models can supply. Alan Grafen expresses this point well: The behavioural ecologist ... does not usually know the genetics underlying the character she studies. While she would be interested to know this genetic system, it is not of primary importance. Her aim is to uncover the selective forces that shape the character. The behavioural ecologist has to hope in her ignorance that her method will work almost regardless of which particular genetic system underlies the character..... (Grafen 1991, p. 5) The crucial point, therefore, is that optimality and ESS models are not simply ‘second best’ explanatory strategies that we adopt for lack of knowledge of the underlying genetics; rather, they furnish a distinctive type of understanding, as explained above. It is also true, of course, that optimality and ESS models are constructed under idealising assumptions concerning the underlying genetics, i.e. we assume that the different phenotypes breed true, and that there is sufficient mutation to allow each to invade (Grafen 1991). But the strategy of idealization lies at the heart of all theoretical model building, as Maynard Smith has repeatedly emphasized.4 A somewhat different way to put the point is this. Optimality and ESS models explain adaptation by considering how fitness varies as a function of phenotypic performance in a given environment (which includes other organisms); they are thus concerned with the ecological causes of fitness differences, and their evolutionary consequences. Such ecological-dynamical understanding is of a different sort from that obtained from genetical theories of adaptation, which study the distribution of phenotypic effects produced by favourable mutations, and more generally, from that obtained from standard population genetics models, which are only indirectly concerned with phenotypic adaptation (Orr 1998, 2005a, b; Kirkpatrick 1996).5 4 Idealisation can be thought of as theorising under ‘as if’ assumptions; it is as pervasive in theoretical evolutionary biology as it is in theoretical physics. There is a growing consensus concerning the cognitive role of idealisation in theory and model construction; see the contributions in Shanks (ed.) (1998). The role of idealisation in the development of population genetics has been analyzed by Margaret Morrison (2000, 2002). 5 H. Allen Orr has recently argued that genetical theories of adaptation fall into two classes: phenotype-based and DNA sequence-based (Orr 1998, 2000, 2005a, b). (What distinguishes a genetical theory of adaptation is that selection coefficients figuring in the models are not arbitrary: they emerge as phenotypic effects from favorable mutational substitutions (Orr 2005a)). The simplest phenotypic model is Fisher’s geometric model; Gillespie’s mutational landscape model is Orr’s example of the DNA sequence approach. A basic difference between the two sorts of model is that phenotype-based models explore a continuous space while DNA based models explore a discrete space. Despite this, Orr’s discussion reveals some striking dynamical similarities between the two types of model; in both cases, the phenotypic sizes of fixed favorable mutations fall off as an approximate geometric sequence and the overall distribution of factors fixed during adaptive walks is nearly exponential (at least among factors of larger effect). This has the philosophically interesting consequence that there is a robust dynamical pattern governing adaptation at both the phenotypic and the DNA sequence levels. Explaining adaptation: history vs. equilibrium A unifying aspect of the modelling strategies dealt with in the previous section is an emphasis on equilibrium states. It is often argued that equilibrium approaches to adaptation are essentially ahistorical. Thus Douglas Futuyma writes: ‘most of population genetics, all of optimization theory and much of community ecology is framed in terms of equilibria rapidly approached; history plays no role in the theory and is wished away whenever possible in interpreting data’. In a similar vein, Reeve and Sherman (1993) distinguish questions of evolutionary history from questions of phenotype existence, arguing that the latter are ahistorical in nature. In The Evolution of Sex, Maynard Smith addressed this issue of equilibrium vs. history, by contrasting questions about the origin and the maintenance of adaptation: So far I have discussed the maintenance of sexual reproduction and genetic recombination. But what of their origins? Surely evolution theory should be concerned with the origins of adaptations, not merely with their maintenance once they have arisen. There is much in this objection; indeed, one of the things I have learned while writing this book is that my own insight into the field may have been obscured by an obsession, which I share with most population biologists, with equilibrium situations (1979, p. 6) Maynard Smith went on to offer a pragmatic justification for focusing on maintenance rather than origins, arguing that questions about the former are more empirically tractable. He continued: ‘we have to concentrate on maintenance rather than origins because only thus have we any hope of testing our ideas’ (ibid. p. 7). Though Maynard Smith’s pragmatic point about testing may be valid, I think the underlying contrast between ‘equilibrium’ and ‘history’ actually needs to be questioned. It is true that optimality and ESS models cannot explain the origins of complex adaptations, only their maintenance; but the type of understanding they yield is nonetheless historical, in a sense. The key is to distinguish between different historical timescales – what might be called ‘recent’ and ‘remote’ history. In explaining the maintenance of a trait, its preservation in recent evolutionary time is what matters. As Ron Amundson has said: ‘equilibrium biology does have what we might call a shallow concept of history, in that it is considered legitimate to infer that a trait of especially high fitness relative to alternatives must at least have been maintained in the population by selection. But remote history is irrelevant’ (1996, p. 45). This idea will be further unpacked below. In his work on functional explanation, Peter Godfrey-Smith provides a framework ideally suited for dealing with the temporal dimension of adaptation at stake here.6 Indeed, in ‘A Modern History Theory of Functions’, he refers explicitly to ESS theory: A final illustration of the importance of the distinction between originating and maintaining selection is found in some of the literature applying game theory to animal behaviour. An ESS, or evolutionary stable strategy, is a strategy which, once prevalent in the population, cannot be invaded by rival strategies. However, an ESS need not be a strategy that can evolve from scratch in any situation. Often, a critical mass of like-minded individuals is needed before a strategy becomes stable. Thus to explain a behaviour by showing it to be an ESS is not necessarily to explain how that behaviour originally became established. Rather it is to point to the selective pressures responsible for the recent maintenance of the strategy in the population (1994, p. 358). What Godfrey-Smith says about ESS models applies to optimality models as well. Such models aim to determine the possible selective forces responsible for the recent maintenance of traits. These forces typically act over many generations, but many generations can be a point in time within a phylogenetic frame of reference: that is, such a period would hardly count as historical in the latter frame. The core of the ‘Modern History’ approach is that biological functions are dispositions or effects of traits which explain their recent maintenance under natural selection. To make a claim about functions is to make a claim about the recent past, not the ancient one. If traits are conceptualised as adaptive strategies, as strategies maximising some component of fitness, then their maximising effect on fitness, relative to a certain environment, explains their recent maintenance under selection. Therefore, the phenotypic understanding of adaptation outlined in the previous section can be considered historical, despite its focus on maintenance rather than origins. My point is that the temporal dimension of such evolutionary causation is to be found in modern history, in Godfrey-Smith’s sense. Godfrey-Smith (1993) reserves the term ‘functional explanation’ for explanation of the recent maintenance of a trait, and ‘evolutionary explanation’ for the original establishment of the trait. I suggest that what Godfrey-Smith calls functional explanation is really a type of evolutionary explanation with a shortened temporal dimension. For the inter-generational time over which traits are selectively maintained can also be considered evolutionary history. This conceptualisation allows us to retain the idea that adaptive explanations are always about the causal history of a trait (as Brandon 1990 insists). For recent maintenance is causal history, albeit over a shorter timescale than original establishment. 6 See also Godfrey-Smith (1993). For an application of the ‘modern history’ framework to a different but related philosophical problem, see Kaplan and Pigliucci (2001). If this is right, it implies that adaptive explanations based on equilibrium states, of the sort furnished by ESS and optimality theory, do qualify as causal. However, Elliott Sober (1983) has contrasted what he calls ‘equilibrium explanation’ with ‘causal explanation’, thus apparently implying that the former is non-casual. Sober’s prime example of an equilibrium explanation is Fisher’s explanation of why many species have a sex ratio close to 1:1. (Fisher’s argument is a precursor of the ESS concept, as Maynard Smith (1982) notes.) Fisher (1930) observed that if a given population consisted mainly of females, it would pay to produce more sons than daughters, as more grandchildren would then be produced. If males were predominant, it would pay to produce more daughters than sons. But if there were equal numbers of males and females in the population, there would be no selective advantage to producing a preponderance of either daughters or sons. Therefore, the 1:1 ratio is evolutionarily stable. Sober argues that a causal explanation would show how the population got to its present state from some specified starting point, i.e. it would trace a dynamical path from the initial conditions to the present state. But Fisher’s account does not do this: In fact, Fisher’s account shows why the actual initial conditions and the actual selective forces don’t matter; whatever the actual initial sex ratio had been, the selection pressures that would have resulted would have moved the population to its equilibrium state. Where causal explanation shows how the event to be explained was in fact produced, equilibrium explanation shows how the event would have occurred regardless of which of a variety of causal scenarios actually transpired. (Sober 1983, p. 202) So according to Sober, a causal explanation focuses exclusively on the actual trajectory of the population; while an equilibrium explanation shows that the actual trajectory is in a sense irrelevant, for the actual starting point makes no difference to the evolutionary outcome. It is true that Fisher’s argument does not say anything regarding actual initial conditions or the population’s actual trajectory. But Fisher’s equilibrium argument is still causal, albeit in a ‘modal’ way. The argument shows how the population would evolve to an equilibrium state from any set of initial conditions. There may be equilibrium explanations that lack such a modaldynamic structure and are formulated only in terms of equilibrium and not dynamical convergence, but Fisher’s argument is clearly dynamic in the proposed modal sense. The argument specifies the selective forces that would cause any population, starting from any initial conditions, to evolve to a 1:1 sex ratio. So Fisher’s sex ratio argument is causal, but it concerns possible causal histories rather than just the actual one. In like manner, equilibrium predictions of optimality and ESS models can be made compatible with a manifold of different actual trajectories, but the explanations that such models yield are nonetheless causal. Sober’s distinction between ‘causal’ and ‘equilibrium’ explanations is in reality a distinction between two different species of causal explanation. As we will see below, the fact that some equilibrium explanations are also causal is connected to a point that emerged in discussions that came after Maynard Smith’s original ESS formulation, namely that evolutionary stability involves both equilibrium and dynamical attainability. To summarize, the type of explanation of adaptation yielded by optimality and ESS models concerns the maintenance, rather than the origins, of adaptive traits, in the first instance. However, such explanations are both historical and causal, despite what has sometimes been thought. Godfrey-Smith’s ‘modern history’ account of functions provides the appropriate timescale for the explanation of adaptation by phenotypic models, and opens a way to broaden what we understand as the causal history of a system. How this relates to the notion of evolutionary stability is the subject of the next section. Evolutionary stability and history In this section I explore further the historical side of ESS arguments by analysing refinements of the notion of evolutionary stability. As with all theoretical models, optimality and ESS models do not relate to empirical reality in any straightforward way, for they require idealising assumptions. Only under those assumptions do they provide the sort of theoretical understanding envisaged above. Such understanding, however, neither implies nor requires that populations in fact be at the optimal configurations predicted. As is well known, biological complexities such as frequency-dependence and environmental variability conspire against fitness maximisation and stability. As Hofbauer and Sigmund note, ‘the idea that a population will somehow evolve until it happily reaches the safe heaven of evolutionary stability is not always valid’ (1998, p. xxiv). Theoretical developments within evolutionary ecology have shown how evolutionary stability involves dynamical relations between predicted equilibria and their evolutionary attainability (Eshel and Motroo 1981; Eshel 1983; Taylor 1989; Christiansen 1991). These developments mean that Maynard Smith’s original ESS concept needs modification. For a state that is an ESS may not be reachable by a population from its actual starting point. As Troy Day says, ‘evolutionarily stable strategies can be evolutionarily unattainable’ (2005, p. 292). To address this problem, some authors have argued that the notion of evolutionary stability should be defined more carefully, to include evolutionary attainability (Christiansen 1991). The importance of evolutionary attainability can be explained as follows. Maynard Smith’s original ESS concept describes a situation where the fitness of an individual is affected by what others are doing in a population. The basic ESS argument establishes what happens once a population has become evolutionarily stable: the (average) population composition would not change if threatened by potential invaders. The strategy adopted at the monomorphic stable state is superior to any strategy used by an invading mutant or immigrating group. The result is that the population ‘will remain effectively monomorphic, with all individuals using the ESS’ (Hines 1987, p. 200). However, the biological meaning of an ESS is not complete unless one can determine that a population will remain at an evolutionary stable state in the face of perturbations; the original definition of an ESS does not guarantee this. An early concern with a dynamical understanding of evolutionary equilibria was expressed in Taylor and Jonker (1978): An equilibrium state, if undisturbed, should persist, since all existing strategies are equally fit. However, in real life, an equilibrium state will surely be disturbed, so that we are only interested in it if the system, once disturbed, returns to the equilibrium state. If this is the case, the equilibrium is said to be stable (Taylor and Jonker 1978, p. 146, emphasis in original.) Recast in these terms, a genuine evolutionary equilibrium requires that two local stability conditions be satisfied (Taylor 1989; Christiansen 1991). One coincides with the ESS condition, the other requires that a population situated near an evolutionary equilibrium (ESS) converges to such an equilibrium (Christiansen 1991). An ESS precludes possible invading mutant strategies from modifying the current stable population composition. Convergence stability ensures that the population has a tendency to move towards the stable state: variants that bring the population closer to the ESS will be selectively favoured. The type of stability that is crucial for evolutionary history includes both the ESS condition and the convergence (i.e. attainability) condition. I suggest that the shift from ESS alone to ESS-plus-attainability yields a causal– historical conception of evolutionary stability, where ‘causal–historical’ is understood in the modal and ‘modern history’ senses elaborated above. The erroneous idea that optimality and game-theoretic explanations are ahistorical and non-causal probably arises from defining evolutionary stability in terms of the ESS condition alone. Once dynamic convergence to ESS is included in the definition of stability, the causal–historical dimension of these explanations becomes immediately apparent. This point can be appreciated from a different angle, by a brief excursion into history. Early pioneers of optimality analysis, such as Richard Levins, took an explicitly dynamical approach; this was clear from Levins ‘optimization principle’, according to which ‘populations will differ in nature in the same direction of their optima’ (Levins 1970, p. 399). The same is true of more recent life-history theorists who have used perturbation techniques to investigate how optima move when conditions change (Hernández and León 1995). Though ESS modelling began as an extension of optimality analysis to frequencydependent contexts, the dynamic dimension was initially obscured. The first precise definition of an ESS was given by Maynard Smith and Price (1973), in their study of the evolution of ritualised animal fights. The ESS concept as defined by Maynard Smith and Price turned out to satisfy the Nash equilibrium condition of game theory; as a result, the ESS concept was seen as limited to equilibrium situations in which dynamics and history played no role. So the use of the classical ESS concept made history irrelevant to phenotypic analyses of adaptation, and evolutionary stability was regarded as an ahistorical notion. But, as we have seen, the ESS condition turned out not to be the whole story for understanding evolutionary stability. Indeed it could not have, for evolutionary adaptation is a dynamic, historical process. The recognition that stability involves ESS-plus-attainability brings history back, in the form of maintenance in ‘modern history’, into the theoretical understanding of adaptation. Evolutionary attainability introduces possible dynamical trajectories for a given equilibrium. Three related topics deserve philosophical attention. Recent work has explored whether different approaches to the study of adaptive evolution, namely optimization, game-theoretic and quantitative-genetic, lead to similar predictions about evolutionary dynamics and equilibria (Gomulkiewicz 1998). This has led to a clarification of the inter-relationships between the modelling approaches and has made their assumptions more transparent (Abrams 2001a; Day and Taylor 2003). Philosophy can do an important job here in exploring the structure of the conceptual framework we have for understanding phenotypic adaptation. The second topic concerns the role of frequency-dependence in producing stable fitness minima that could be involved in situations of competition and disruptive selection (Abrams et al. 1993; Abrams 2001b). Indeed, Day emphasises how the distinction between an ESS and convergence stability ‘illustrates the potential for a trait to evolve to a point at which natural selection becomes disruptive’ (Day 2005, p. 295). The third topic concerns the theory of long-term selection, which ‘was originally suggested as a means of closing the gap between the genotypic and the phenotypic approaches to natural selection’ (Eshel and Feldman 2001, p. 161); see also Hammerstein (1996). According to the proponents of this approach, ESS stability holds for long-term selection in phenotypic space but not for the short-term selection analysed by population genetics models, that operates in genotypic space (Eshel 1996). It remains to be assessed whether long-term selection would be compatible with the modern history framework presented here. But an analysis of these topics will have to wait for another occasion. Concluding remarks Maynard Smith’s pioneering work in evolutionary ecology produced novel modelling strategies aimed at the explanation of phenotypic adaptation. These modelling strategies involve a modal dimension and yield a ‘structural’ understanding of adaptation; as we have seen, both the modal and the structural aspects were part of Maynard Smith’s general philosophical stance towards modelling in science. In this paper, I have tried to formulate some of the philosophical questions arising from optimality and ESS modelling, and have offered some preliminary answers. Besides the problem of elucidating the type of scientific understanding that optimality and ESS models provide, there was the central question of whether the notion of evolutionary stability is historical and if so, what sense of history it involves. This in turn led to an analysis of how the concept of evolutionary stability has been modified since the original ESS formulation. If my analysis is correct, then the refinement of the notion of evolutionary stability to include both an equilibrium and an attainability condition shows that stability analyses of adaptation are causal– dynamical in nature, and do involve a historical dimension, in the ‘modern history’ sense specified above. The questions posed here are meant to provoke further analyses of the natural philosophy of adaptation. Acknowledgements I wish to thank Jesús Alberto León, Elliott Sober and Samir Okasha for their comments on an early draft of this paper, and an anonymous referee for valuable suggestions. The ideas developed here have benefited from discussions with David Hull, Peter Abrams, Steven Orzack, Paul Griffiths, Greg Cooper, Robert Brandon, Jim Lennox, Diego Rodriguez, Jesús Alberto León, H. Allen Orr, Steven French and Otávio Bueno. References Abrams P.A., Matsuda H. and Harada Y. 1993. Unstable fitness maxima and stable fitness minima in the evolution of continuous traits. Evol. Ecol. 7: 465–487. Abrams P. 2001a. Modelling the adaptive dynamics of traits involved in inter- and intraspecific interactions: an assessment of three methods. Ecol. Letts 4: 166–175. Abrams P. 2001b. Adaptation, optimality models and tests of adaptive scenarios. In: Orzack S.H. and Sober E. (eds), Adaptationism and Optimality, Cambridge University Press, pp. 273–302. Amundson R. 1996, Historical development of the concept of adaptation. In: Rose M. R. and Lauder G. (eds), Adaptation, Academic Press, pp. 11–53. Brandon R. 1990. Adaptation and Environment. Princeton University Press, Princeton, NJ. Cartwright N. 1983. How the Laws of Physics Lie. Oxford University Press, New York. Christiansen F.B. 1991. On conditions for evolutionary stability for a continuously varying character. Am. Nat. 138: 37–50. Darwin C. 1859. On the Origin of Species by Means of Natural Selection, Edited by J.W. Burrow, Penguin Classics, 1968. Day T. and Taylor P. 2003. Evolutionary dynamics and stability in discrete and continuous games. Evol. Ecol. Res. 5: 605–613. Day, T. 2005. Modelling the ecological context of evolutionary change: déjà vu or something new? In: Cuddington K. and Beisner E. (eds), Ecological Paradigms Lost: Routes to Theory Change, Academic Press. Eshel I. 1983. Evolutionary and continuous stability. J. Theor. Biol. 103: 99–111. Eshel I. 1996. On the changing concept of evolutionary population stability as a reflection of a changing point of view in the quantitative theory of evolution. J. Math. Biol. 34: 485–510. Eshel I. and Motro U. 1981. Kin selection and strong evolutionary stability of mutual help. Theor. Popul. Biol. 19: 420–433. Eshel I. and Feldman M. 2001. Optimality and evolutionary stability. In: Orzack S.H. and Sober E. (eds), Adaptationism and Optimality, Cambridge University Press, New York, pp. 161–190. Fisher R.A. 1930. The Genetical Theory of Natural Selection. Clarendon Press, Oxford. French S. and Da Costa N.C.A. 2003. Science and Partial Truth: A Unitary Account to Models and Scientific Reasoning. Oxford University Press, New York. Godfrey Smith P. (1993). Functions: consensus without unity. Pacif. Philos. Quart. 74: 196–208. Reprinted in Hull D.L. and Ruse M. (eds), 1998, The Philosophy of Biology, Oxford University Press, New York, pp. 280–292. Godfrey-Smith P. 1994. A modern history theory of functions. Nous 28: 344–362. Hammerstein P. 1996. Darwinian adaptation, population genetics, and the streetcar theory of evolution. J. Math. Biol. 34: 511–532. Grafen A. 1991. Modelling in behavioural ecology. In: Krebs J.R. and Davies N.B. (eds), Behavioural Ecology, 3Blackwell, Oxford, pp. 5–31. Gomulkiewicz R. 1998. Game theory, optimization, and quantitative genetics. In: Dugatkin L.A. and Reeve H.K. (eds), Game Theory and Animal Behavior, Oxford University Press, NewYork, pp. 283–303. Hernández M.J. and León J.A. 1995. Evolutionary perturbations of optimal life histories. Evol. Ecol. 9: 478–494. Hines W.G.S. 1987. Evolutionary stable strategies: a review of basic theory. Theor. Popul. Biol. 31: 195–272. Hofbauer J. and Sigmund K. 1998. Evolutionary Games and Population Dynamics, Cambridge University Press. Hughes R.I.G. 1993. Theoretical explanation. In: French P.A. and Uehling Jr. T. E. (eds), Midwest Studies in Philosophy: Philosophy of Science, Vol. 18, University of Notre Dame Press, pp. 132– 153. Kaplan J.M. and Pigliucci M. 2001. Genes ‘for’ phenotypes: a modern history view. Biol. Philos. 16: 189–213. Kirkpatrick, M. 1996. Genes and adaptation: a pocket guide to the theory. In: Rose M.R. and Lauder G.V. (eds), Adaptation, Academic Press, pp. 125–146. Levins R. 1970. Fitness and optimization. In: Kojima (ed.), Mathematical Topics in Population Genetics, Springer-Verlag, Berlin, pp. 389–400. Maynard Smith J. 1978a. Optimization Theory in Evolution, Annu. Rev. Ecol. Syst. 9: 31–56. Reprinted in Sober E. 1984. (ed.), Conceptual Issues in Evolutionary Biology, MIT Press, Cambridge, MA, pp. 298–315. Maynard Smith J. 1978b. The Evolution of Sex. Cambridge University Press, Cambridge. Maynard Smith J. 1979. Game theory and the evolution of behaviour. Proc. Roy. Soc. Lon. B 205: 475–488. Maynard Smith J. 1982. Evolution and the Theory of Games. Cambridge University Press, Cambridge. Maynard Smith J. 1987. How to model evolution. In: J. Dupré (ed.), The Latest on the Best: Essays On Evolution and Optimality, MIT Press, pp. 119–131. Maynard Smith J. 1989. Did Darwin get it Right? Essays on Games, Sex and Evolution. Chapman and Hall, New York. Maynard Smith J. and Price G.R. 1973. The logic of animal conflicts. Nature 246: 15–18. Morgan M. and Morrison M. (eds), 1999. Models as Mediators. Cambridge University Press, New York. Morrison M. 2000. Unifying Scientific Theories: Physical Concepts and Mathematical Structures. Cambridge University Press, New York. Morrison M. 2002. Modelling populations: Pearson and Fisher on Mendelism and biometry. Brit. J. Philos. Sci. 53: 39–60. Orr H.A. 1998. The population genetics of adaptation: the distribution of factors fixed during adaptive evolution. Evolution 52: 935–949. Orr H.A. 2000. Adaptation and the cost of complexity. Evolution 54: 13–20. Orr H.A. 2005a. Theories of adaptation: what they do and don’t say. Genetica 123: 3–13. Orr H.A. 2005b. The genetic theory of adaptation: a brief history. Nat. Rev. Genet. 6: 119–127. Parker G.A. and Hammerstein P. 1985. Game theory and animal behaviour. In: Greenwood P. J., Harvey, Paul H. and Slatkin M. (eds), Evolution: Essays in Honour of John Maynard Smith, Cambridge University Press, pp. 73–94. Plutynski, A. 2004. Explanation in classical population genetics. Philos. Sci. Proceedings Part II: 1201–1215. Reeve H.K. and Sherman P.W. 1993. Adaptation and the goals of evolutionary research. Quart. Rev. Biol. 68: 1–32. Shanks N. (ed.), 1998. Idealization IX: Idealization in Contemporary Physics, Poznan Studies in the Philosophy of the Sciences and the Humanities, vol. 63. Rodopi. Sober E. 1983. Equilibrium explanation. Philos. Stud. 43: 201–210. Suppes P. 1966. A comparison of the meaning and uses of models in mathematics and the empirical sciences. In: H. Freudenthal (ed.), The Concept and the Role of Models in Mathematics and Natural and Social Sciences, D. Reidel, pp. 163–177. Taylor P.D. 1989. Evolutionary stability in one-parameter models under weak selection. Theor. Popul. Biol. 36: 125–143. Taylor P.D. and Jonker L.B. 1978. Evolutionary stable strategies and game dynamics. Math. Biosci. 40: 145–156. Weibull J. 1985. Evolutionary Game Theory. MIT Press, Cambridge, MA.