During the past few decades, multilevel selection has been one of the most discussed subjects in the philosophy of biology and evolutionary theory. But, even though the subject has generated an enormous interest, I believe this interest remains fairly ambiguous and multilevel selection remains an essentially marginal or even questionable issue. This may seem like a gross overstatement: after all, we have come a long way from the 1960s and 1970s, when the heavy critiques directed against Wynne-Edwardsâ group selectionist theories on population regulation led multilevel selection into serious disrepute. Indeed, since then, the idea of multilevel selection seems to have become widely accepted, both by biologists and philosophers, even though many disagreements about its details may subsist. So, how could one claim that this notion is still marginal, or even questionable?
To understand this, note that there is a serious ambiguity that has affected recent multilevel selection theory. It stems from the widely accepted idea that there are two types multilevel selection. In one type, what is of interest is the evolution of individual (or lower-level) characters in group-structured populations: this is usually called âmultilevel selection 1â; on the other hand, in âmultilevel selection 2â, what we are interested in is the evolution of group (or higher-level) characters (Damuth and Heisler 1988).
But the distinction between these termsânowadays abbreviated as MLS1 and MLS2 (Okasha 2006)âhas progressively come to be seen as a distinction between two kinds of multilevel selection scenarios: scenarios in which group selection refers to some groups making more individuals than others due to a given group trait (MLS1), and scenarios in which some groups make more groups than others (MLS2). I believe this semantic transformation is a mistake (though arguing why this is so is beyond the scope of this introduction and will have to be done elsewhere 1 ). Instead of seeing MLS1 and MLS2 as two perspectives that we can take on any multi-group scenario (and this is how Damuth and Heisler (1988) understood these notions when they first proposed them, though not without a hint of ambiguity), we have now come to reify these two perspectives and turn them into two kinds of cases. Once this is done, one can allegedly choose between these types of cases, depending on oneâs research interests. For example, if one is interested in the extremely popular issue of the evolution of altruismâthis being an individual-level trait pertaining to the behavior of organismsâone would seem encouraged to concentrate on scenarios involving groups making more individuals (i.e. MLS1 scenarios), rather than on scenarios involving groups making more groups. And, indeed, as the issue of altruism is one of the main issuesâif not the main issueâfuelling our interest in multilevel selection, in recent decades, more and more of the interest of researchers seems to have been directed towards scenarios involving groups making more individuals. To the point that a recent survey on these issues (West et al. 2007) calls ânew group selectionâ the scenarios and models involving groups making more individuals and âold group selectionâ those involving groups making more groups; and their contention is that old group selection has long been refuted by empirical and theoretical research and that it is now only of interest for the historian of science.
Of course, many authors working in multilevel selection theory may contest the fact that âold group selectionâ has been convincingly refuted and should be seen as a thing of the past; indeed, it is striking that, in his response to West et al.âs paper, D.S. Wilson (2008), one of the main proponents of ânewâ group selection (Wilson 1975), has argued against the view that âoldâ group selection has been refuted by empirical and theoretical research. But the point I want to stress here is that, even though many authors may theoretically contest the idea that âoldâ group selection is a thing of the past, by the mere fact that most issues addressed nowadays both by philosophers and biologists are concerned with ânew group selectionâ (groups making more individuals), West et al.âs position is, or, in any case, tends to be, de facto confirmed. To put it otherwise, though many authors may disagree with the idea that old group selection should be abandoned, it does in fact get progressively abandoned simply because multilevel selection research seems to be more and more concerned with ânewâ group selection (with groups making more individuals). So if, as I will argue below, contemporary multilevel selection theory may risk marginalizationâand possibly even a new controversial statusâthis is less due to theoretical considerations, but rather due to the actual practice in the field, i.e. due to the fact that most researchers seem to devote their work to scenarios involving groups making more individuals.
(As a caveat, let me add that there are important exceptions to this insistence on ânewâ group selection, i.e. there are researchers that are still proposing models or doing extremely valuable field or laboratory work devoted to âoldâ group selection: e.g. Jablonski and Hunt (2006), Pruitt and Goodnight (2014) and Michael Wade (see his 2016 book, for a comprehensive overview of his work), to name just a few examples. And, as already mentioned above, I believe it is the experimental work of the latter that could be used in order to dispel some of the ambiguities that are, to my mind, affecting the contemporary status of multilevel selection theory.)
Going back to my argument, I believe that, as long as it concentrates on scenarios and models with groups making more individuals, multilevel selection theory is bound for marginality and even for a controversial status. Indeed, for scenarios with groups making more individuals, it has been repeatedly shown during the past few decades (see West et al. 2007 for a list of references) that kin selection and multilevel selection are mathematically identical and the two approaches are nothing more than different ways of conceptualizing what is essentially the same evolutionary process. Even authors much more sympathetic to multilevel selection than West et al. have highlighted this point. For example, Sober and Wilson (1998, p. 57), having kin selection in mind, note: âThe theories that have been celebrated as alternatives to group selection are nothing of the sort. They are different ways of viewing evolution in multigroup populations.â This compatibility between multilevel selection and kin selection theory explains, I believe, why multilevel selection theory seems to have reached a respectable level of acceptability in the scientific community over the past few decades. However, this âin theoryâ acceptance of the legitimacy of multilevel selection masks a significant risk or, more precisely, a de facto marginalization of multilevel selection, because an explanation appealing to multiple levels of selection will always present significant pragmatic disadvantages with respect to an explanation based on entities at a single level (and whose fitnesses are affected both by direct and indirect effects of their traits, as is the case for kin selection). The latter approach will usually make model construction easier; the predictions the latter yields will be easier to interpret and, also, the latter would be applicable to much more general cases (for a full list of these pragmatic advantages, see, again, West et al. 2007). In short, the recent insistence, by supporters of multilevel selection theory, on scenarios and models involving groups making more individuals may potentially have ensured (but see below) a wide theoretical acceptance for multilevel selection explanations; but this comes at the cost of pushing multilevel selection theory towards a very marginal status from a pragmatic point of view: a theory may be accepted in general theoretical terms, but if it is not actually used, or if it is only very rarely used by scientists in their explanations of actual phenomena, then its future is not looking favorable.
But, leaving its marginality-inducing effects aside, even the much wider theoretical acceptance of multilevel selection theory that we see today is somewhat controversial. When one accepts that kin selection and multilevel selection theory are mathematically identical when we are dealing with groups making more individuals, it is not always clear what exactly is thus accepted. As Goodnight (2013) has pointed out, this equivalence is only true if we adopt a âcontextual analysisâ framework for understanding such scenarios (based on Heisler and Damuth 1987), and not if we adopt an understanding of group selection based directly on Priceâs (1972) hierarchical equation. However, the issue is further complicated when we bear in mind that, as Jeler (2017) has shown, contrary to the Pricean approach, when it is applied to cases with groups making more individuals, the âhigher-level selectionâ notion from the contextual analysis framework is nothing more than a form of lower-level selection, in the sense that groups are not conceived of as being under selection therein, but only individuals are. Putting these elements together, we are lead to the puzzling conclusion that when one accepts that kin selection and âmultilevelâ selection are mathematically identical and explanatorily interchangeable, this statement does not refer to multilevel selection at all. There may be a wide acceptance of the legitimacy of multilevel selection explanations, but whether what are thus accepted are genuine multilevel explanations is far from certain: suddenly, multilevel selection becomes questionable again.
The recent insistence of researchers on ânew group selectionââi.e. on scenarios with groups making more individualsâthus seems to lead multilevel selection theory towards marginalization, at best, ...