Smaldino argues that the division of labor may have influenced the evolution of social activity. We agree. However, he has not established that group-level traits constitute a distinct unit of selection. Moreover, Smaldino's argument that group-level traits are emergent, and therefore irreducible to the individual level, is unsound.
With respect to the first argument, we should follow Hull (Reference Hull, Jensen and Harré1981) and distinguish between two senses of “unit of natural selection.” Interactors are biological individuals such as organisms, which stand in ecological relations to the world. Replicators are biological units such as genes, which are copied from one generation to the next and preserved through evolutionary time. In claiming that group-level traits are units of selection, Smaldino may be claiming that they are replicators or that they are interactors. We will argue that they are neither.
It is unclear how group-level traits could act as interactors because they do not mediate interactions with the world for replicators that construct them. Smaldino himself seems to recognize this and never claims that group-level traits stand in ecological relations to the world. He does, however, emphasize the persistence and transmission of group-level traits, suggesting that he sees them as replicators. But persistence and transmission alone are not sufficient qualifications to be a Darwinian replicator. Darwinian replicators' fecundity must depend on their own nature, coming from something like protein structures or regulatory processes that they encode (Dawkins Reference Dawkins1982). Because downstream products of replicators are not directly replicated themselves, phenotypic traits generally do not qualify as replicators despite their persistence and transmission. For example, human skin color is transmitted and reproduced over time, but is not regarded as a replicator. This is because we can fully account for the evolution of skin color by appeal to selection on populations of humans (interactors) and their genes or developmental systems (replicators). Group-level traits, analogously, are phenotypic traits whose transmission and reproduction are standardly accounted for in terms of individual-level replicators and interactors.
Smaldino disagrees, of course, arguing that group-level traits cannot be accounted for at a more basic level. To block the reduction of group-level traits to the individual level, he appeals to Wimsatt's (Reference Wimsatt1997) analysis of emergence and aggregativity. An aggregative system is one in which the whole is a mere aggregate of its parts, whereas non-aggregative systems display emergent properties that cannot be accounted for by a simple summation of the properties of individual parts. For a system to count as aggregative, Wimsatt argues that the behavior of the system must be invariant under four different types of alterations to its parts: (1) rearrangement, (2) addition or deletion, (3) decomposition and recombination, and (4) linear amplification. Smaldino correctly notes that group-level traits are not likely to be invariant under any of these alterations, so they qualify as emergent in Wimsatt's sense.
We agree with Smaldino up to this point. However, Smaldino goes on to argue that because group-level traits are emergent, their evolution cannot be explained by selection at the individual level. This argument relies on a premise that we reject: that emergence implies irreducibility. This may be true for some emergent properties, but will not generally be true for the properties Wimsatt's test counts as emergent. In fact, a primary aim of the paper in which Wimsatt introduces the aggregativity test is to argue that emergence is fully compatible with reductive explanation. The fact that group-level traits pass Wimsatt's test for emergence does not therefore support the claim that group-level traits cannot be explained reductively in terms of selection on individual organisms or the units of culture of which they are composed.
Wimsatt's aggregativity test is also unsuitable for Smaldino's purposes because it identifies almost all systems as emergent. Even most physical properties (e.g., the combined volume of a mixture of different substances) are emergent in Wimsatt's sense, and nearly all social and biological systems will fail some, if not all, of the four conditions of aggregativity (Wimsatt Reference Wimsatt2000). This is problematic because it means that Wimsatt's test cannot distinguish group-level traits from the social systems Smaldino wants to contrast them with. For example, according to Smaldino, cooperation involves simply contributing resources to another individual but collaboration requires organized coordination and is thus an emergent property of a social system. However, many types of cooperation will vary along with the addition, deletion, and recombination of individuals, because these operations alter the relative distribution of resources. Thus, according to Wimsatt, cooperation is an emergent property. In this case, and many others important to Smaldino, Wimsatt's test cannot distinguish a special class of group-level traits.
Although Wimsatt's analysis of emergence does not support the claim that group-level traits are irreducible, it does illuminate another important aspect of explaining the evolution of social structures. By showing how emergence and reductive explanation can be compatible, Wimsatt helps us understand that we are not limited to methodologies that are either reductive and hostile to emergent properties, on the one hand, or non-reductive and friendly to emergence, on the other. Instead, understanding emergence along Wimsatt's lines opens up a middle path, a methodology that is reductive and nonholistic, but that uses this perspective to understand and explain emergent properties. Wimsatt calls this a reductive heuristic (Wimsatt Reference Wimsatt1997).
As we see it, the state of the art for cultural and social evolution modeling already follows this middle path. For example, deploying individual-based models (IBMs) is a fully reductive methodology in the sense that group properties are built out of individual interactions. Sophisticated IBMs include individual-level relational properties such as relative spatial location and potential roles in collaborative effort. These are the very factors Smaldino emphasizes, and we suspect he approves of the use of IBMs in researching cultural evolution (e.g., Smaldino et al. Reference Smaldino, Pickett, Sherman and Schank2012; Reference Smaldino, Schank and McElreath2013b). However, in individual-based modeling we can see how interactions lead to emergent properties without our having to abandon methodological individualism. For this reason, we remain unconvinced of the need to postulate irreducible group-level units of selection to generate scientific explanations of social interaction and evolution.
Smaldino argues that the division of labor may have influenced the evolution of social activity. We agree. However, he has not established that group-level traits constitute a distinct unit of selection. Moreover, Smaldino's argument that group-level traits are emergent, and therefore irreducible to the individual level, is unsound.
With respect to the first argument, we should follow Hull (Reference Hull, Jensen and Harré1981) and distinguish between two senses of “unit of natural selection.” Interactors are biological individuals such as organisms, which stand in ecological relations to the world. Replicators are biological units such as genes, which are copied from one generation to the next and preserved through evolutionary time. In claiming that group-level traits are units of selection, Smaldino may be claiming that they are replicators or that they are interactors. We will argue that they are neither.
It is unclear how group-level traits could act as interactors because they do not mediate interactions with the world for replicators that construct them. Smaldino himself seems to recognize this and never claims that group-level traits stand in ecological relations to the world. He does, however, emphasize the persistence and transmission of group-level traits, suggesting that he sees them as replicators. But persistence and transmission alone are not sufficient qualifications to be a Darwinian replicator. Darwinian replicators' fecundity must depend on their own nature, coming from something like protein structures or regulatory processes that they encode (Dawkins Reference Dawkins1982). Because downstream products of replicators are not directly replicated themselves, phenotypic traits generally do not qualify as replicators despite their persistence and transmission. For example, human skin color is transmitted and reproduced over time, but is not regarded as a replicator. This is because we can fully account for the evolution of skin color by appeal to selection on populations of humans (interactors) and their genes or developmental systems (replicators). Group-level traits, analogously, are phenotypic traits whose transmission and reproduction are standardly accounted for in terms of individual-level replicators and interactors.
Smaldino disagrees, of course, arguing that group-level traits cannot be accounted for at a more basic level. To block the reduction of group-level traits to the individual level, he appeals to Wimsatt's (Reference Wimsatt1997) analysis of emergence and aggregativity. An aggregative system is one in which the whole is a mere aggregate of its parts, whereas non-aggregative systems display emergent properties that cannot be accounted for by a simple summation of the properties of individual parts. For a system to count as aggregative, Wimsatt argues that the behavior of the system must be invariant under four different types of alterations to its parts: (1) rearrangement, (2) addition or deletion, (3) decomposition and recombination, and (4) linear amplification. Smaldino correctly notes that group-level traits are not likely to be invariant under any of these alterations, so they qualify as emergent in Wimsatt's sense.
We agree with Smaldino up to this point. However, Smaldino goes on to argue that because group-level traits are emergent, their evolution cannot be explained by selection at the individual level. This argument relies on a premise that we reject: that emergence implies irreducibility. This may be true for some emergent properties, but will not generally be true for the properties Wimsatt's test counts as emergent. In fact, a primary aim of the paper in which Wimsatt introduces the aggregativity test is to argue that emergence is fully compatible with reductive explanation. The fact that group-level traits pass Wimsatt's test for emergence does not therefore support the claim that group-level traits cannot be explained reductively in terms of selection on individual organisms or the units of culture of which they are composed.
Wimsatt's aggregativity test is also unsuitable for Smaldino's purposes because it identifies almost all systems as emergent. Even most physical properties (e.g., the combined volume of a mixture of different substances) are emergent in Wimsatt's sense, and nearly all social and biological systems will fail some, if not all, of the four conditions of aggregativity (Wimsatt Reference Wimsatt2000). This is problematic because it means that Wimsatt's test cannot distinguish group-level traits from the social systems Smaldino wants to contrast them with. For example, according to Smaldino, cooperation involves simply contributing resources to another individual but collaboration requires organized coordination and is thus an emergent property of a social system. However, many types of cooperation will vary along with the addition, deletion, and recombination of individuals, because these operations alter the relative distribution of resources. Thus, according to Wimsatt, cooperation is an emergent property. In this case, and many others important to Smaldino, Wimsatt's test cannot distinguish a special class of group-level traits.
Although Wimsatt's analysis of emergence does not support the claim that group-level traits are irreducible, it does illuminate another important aspect of explaining the evolution of social structures. By showing how emergence and reductive explanation can be compatible, Wimsatt helps us understand that we are not limited to methodologies that are either reductive and hostile to emergent properties, on the one hand, or non-reductive and friendly to emergence, on the other. Instead, understanding emergence along Wimsatt's lines opens up a middle path, a methodology that is reductive and nonholistic, but that uses this perspective to understand and explain emergent properties. Wimsatt calls this a reductive heuristic (Wimsatt Reference Wimsatt1997).
As we see it, the state of the art for cultural and social evolution modeling already follows this middle path. For example, deploying individual-based models (IBMs) is a fully reductive methodology in the sense that group properties are built out of individual interactions. Sophisticated IBMs include individual-level relational properties such as relative spatial location and potential roles in collaborative effort. These are the very factors Smaldino emphasizes, and we suspect he approves of the use of IBMs in researching cultural evolution (e.g., Smaldino et al. Reference Smaldino, Pickett, Sherman and Schank2012; Reference Smaldino, Schank and McElreath2013b). However, in individual-based modeling we can see how interactions lead to emergent properties without our having to abandon methodological individualism. For this reason, we remain unconvinced of the need to postulate irreducible group-level units of selection to generate scientific explanations of social interaction and evolution.