Heyes is right to bring cultural evolution to the forefront of cognitive development. As the cognitive and developmental sciences have been slowly drifting away from extreme nativist claims, the glaring question is now, “Where do constraints on social learning come from?” As it happens, cultural and language evolution experts have been working on aspects of this question for some time now. Heyes (Reference Heyes2018) gives the reader a number of useful landmarks from those fields, but the real story of how cognitive development and cultural evolution interact is somewhat more complex than she implies. We unpack the book's central evolutionary arguments, noting some important perspectives that are omitted and highlighting the need for involvement by experts in complex systems. We will also identify problems with her central grist-and-mills metaphor. Though our tone is one of critique, we want to make it clear that we think Heyes’ book is an important contribution to the literatures on both cognitive development and cultural evolution, and we fully endorse her thesis that a deeper synergy between these often disconnected research areas is paramount.
There are two principal components of Heyes’ argument. The first is that cognitive mechanisms are not innate but are shaped during development by social forces – we might call this “gadgetization.” The second component is that the structure of these social forces is the product of cultural evolutionary processes. We address each of these in turn.
The gadgetization that Heyes writes about is an important and underappreciated idea, with thick roots in the cognitive science of the 1990s. Heyes provides compelling demonstrations of why imitation, mindreading, and language should each be seen not as innately-given cognitive instincts but as learned cognitive gadgets. In general, calling these neural mechanisms “cognitive gadgets” will be a catchy and helpful remedy for cognitive developmentalists who have been tempted to accept the notion of “cognitive instincts.” But others may want a bit more detail regarding how these gadgets are actually constructed. For example, many non-nativist cognitive developmentalists will recognize cognitive gadgets as resulting from something like the learning-based modularization process described by Karmiloff-Smith (Reference Karmiloff-Smith1994). Such processes are likely to have substantial innate architectural constraints based on gross neuroanatomy, but fewer representational constraints, because representations are based on more plastic fine-grained neuronal connectivity (Elman et al. Reference Elman, Bates, Johnson, Karmiloff-Smith, Parisi and Plunkett1996). There is already a large literature on how individuals’ environments, including the social and cultural environments, shape fundamental aspects of their cognitive machinery (e.g., Henrich et al. Reference Henrich, Heine and Norenzayan2010; Karmiloff-Smith Reference Karmiloff-Smith1994; Nisbett et al. Reference Nisbett, Peng, Choi and Norenzayan2001; Nisbett & Miyamoto Reference Nisbett and Miyamoto2005), and so we hope the additional evidence Heyes provides signals the end of overly simplistic nativist approaches to cognitive development. Beyond this statement, however, Heyes provides somewhat scant details regarding how gadgetization occurs. One of the few detailed examples given is the rather outdated dual-route model of literacy. Readers interested in a model that actually learns the way children do might be better served by Seidenberg's (Reference Seidenberg2017) neural networks.
Savvy readers will also want to know more about how cultural evolution shapes the social learning environment that imposes this gadgetization. The most interesting part of Heyes’ argument, in our view, is that the social environments that facilitate the development of culturally important cognitive mechanisms also evolve. This is an important point that has, to date, received too little attention. However, Heyes’ presentation of how the relevant evolutionary dynamics occur is somewhat thin, and glosses over both the staggering complexity of those dynamics and the considerable progress that has already been made in describing that complexity.
Heyes characterizes members of the California school as being vague and inconsistent in their descriptions of evolutionary dynamics. Let us first note that “the California school” is an unfortunate term for many reasons, not least because it enables the author to ascribe to a collective a set of viewpoints without naming any individual sources, which would require attribution and subsequently allow scrutiny of those claims. Many of those associated with the referenced “school” have gone to great lengths to individuate their own research programs and clarify exactly the sorts of transmission, inheritance, and population-level dynamics they believe to have occurred in human cultural evolution, including using computational models, archaeology, cross-cultural fieldwork, and laboratory experiments.
In contrast, the evolutionary dynamics Heyes proposes are themselves vague. Let us consider her discussion of cultural group selection, which draws little from more established discussions (e.g., Boyd & Richerson Reference Boyd and Richerson2002; Henrich Reference Henrich2004a; Richerson et al. Reference Richerson, Baldini, Bell, Demps, Frost, Hillis, Mathew, Newton, Naar, Newson, Ross, Smaldino, Waring and Zefferman2016; Smaldino Reference Smaldino2014; Wilson Reference Wilson2002; Zefferman & Mathew Reference Zefferman and Mathew2015). The key point about multilevel selection or group selection is that selection on a social trait can strongly depend on the social environment, and so the contributions of each level of selection must be taken into account to analyze the overall population dynamics (Okasha Reference Okasha2009). Altruists can be exploited by free riders within their group (individual-level selection against altruism) but still increase the overall fitness of their group relative to other groups (group-level selection for altruism). Additionally, prosocial norms often require coordination among social actors to provide marginal benefits, and therefore face difficulties in propagating in new communities (Bicchieri Reference Bicchieri2006; Boyd & Richerson Reference Boyd and Richerson2002; Richerson et al. Reference Richerson, Baldini, Bell, Demps, Frost, Hillis, Mathew, Newton, Naar, Newson, Ross, Smaldino, Waring and Zefferman2016). Understanding how such social traits spread is a key research area throughout the social sciences, encompassing not only the selection-based work in cultural evolution, but also the epidemiology of beliefs pervasive in other areas of the social sciences (e.g., Centola Reference Centola2018; Sperber Reference Sperber1996). Heyes’ evolutionary model (pp. 199–201) pays lip service to group selection, but closer inspection shows that her model is completely unaffected by group structure. In her model, trait M′ always has higher fitness than trait M, so that it will spread regardless of group structure. There is no selection at the group level that is any different from selection at the individual level. This is a mistake, of course. The traits Heyes is talking about are social and will likely have different fitness gradients at different levels of selection; it's just that there is no mention of what the different gradients might look like. Heyes does briefly mention the potential importance of social complexity, but hardly enough. The consideration of emergent social structures that are important in cultural evolution is explored at length in Smaldino (Reference Smaldino2014). There is still a lot of important work to be done on understanding these evolutionary dynamics, and we agree with Heyes that the inheritance mechanisms for social structures is an important target for future research.
What makes understanding the cultural evolution of social organization so challenging? We suspect that the missing piece from Heyes’ discussion is the complex feedback between extant traits and selection pressures and principles of organization (cf. Kauffman Reference Kauffman1993; Thompson Reference Thompson1942) that shape those traits. In terms of the cognitive gadgets discussed in Heyes’ book, some of the clearest perspectives on this front can be found in the work of Kirby and Smith (Kirby Reference Kirby2017; Smith & Kirby Reference Smith and Kirby2008). They have proposed that once humans had genetically evolved sufficient capacities for sociality, imitation, and cooperation, cultural evolution was sufficient to shape early communication systems towards easier understanding, production, and flexibility. Most importantly, they have provided dynamic models of how such a process might occur. Tomasello et al. have suggested more generally that social cooperation was a likely driver of the emergence of many uniquely human cognitive features (Moll & Tomasello Reference Moll and Tomasello2007; Tomasello et al. Reference Tomasello, Melis, Tennie, Wyman and Herrmann2012; Tomasello & Gonzalez-Cabrera Reference Tomasello and Gonzalez-Cabrera2017), which is consistent with, but not a direct consequence of, Heyes’ central arguments.
In general, the processes of feedback across multiple organizational and temporal scales needed to explain the evolution of complex systems (Caporael Reference Caporael, Scher and Rauscher2003; Caporael et al. Reference Caporael, Griesemer and Wimsatt2013; Smaldino Reference Smaldino2014; Wimsatt Reference Wimsatt, Schaffner and Cohen1974) are missing from Heyes’ story, though their importance is hinted at in her final chapter. Implicit in this discussion of feedback is that, at some point in our evolutionary past, humans lived in social environments that were not well adapted to facilitate the development of many cognitive gadgets in the forms we now know (relatedly, it is likely that other gadgets, adaptive in past environments, have since been lost). Over time, humans constructed environments which promoted the development of new cognitive gadgets, which in turn facilitated the development of new environments, and on and on. Co-evolutionary dynamics like this have, in fact, been increasingly studied under the banner of cultural niche construction (Kendal Reference Kendal2011; Laland & O'Brien Reference Laland and O'Brien2011). This body of work explicitly targets the evolutionary feedback processes by which humans modify their environments (e.g., by producing new social institutions like writing), which in turn creates new selection pressures (e.g., by encouraging literacy), which in turn creates new opportunities for modifying the social environment (e.g., by producing new divisions of labor in which some individuals are express stewards of written knowledge), and so on.
All this creates problems for Heyes’ central metaphor of grist and mills. There is a reason that the social skills essential to human cultural learning are reasonably well described as being handled by cognitive “gadgets,” in the sense that the word usually refers to cobbled-together thingamajigs. The reason is that the environments that led to the evolution of those skills were socially constructed ones – a set of social niches constructed by the same species that was itself developing those skills. These gadgets were pieced together over time by a nonlinear unguided process, and, therefore, they are not pristine engineered devices. The grist was not already there to cause the formation of the mill, nor was the mill already there to cause the formation of the grist. As Heyes herself notes on p. 203, “the inheritance mechanisms for mills overlap with the inheritance mechanisms for grist.” Thus, the social environments influencing development (e.g., the grist) co-evolved with the cognitive gadgets (e.g., the mills), bringing each other into being in a fashion not unlike autocatalysis (where two chemical reagents cause each other to come into prominence). To understand something like autocatalysis, one needs some facility with the dynamics of complex systems. Treating cultural evolution and cognitive development as though they are linear feed-forward processes that straightforwardly turn selection pressures into human traits just will not cut it. For example, the social mechanisms of language use and the neural mechanisms of language processing may not be well treated as “a grist” and “a mill,” respectively, precisely because they overlap so much with one another (e.g., Clark Reference Clark2008; Kirby et al. Reference Kirby, Cornish and Smith2008; Spivey & Richardson Reference Spivey, Richardson, Robbins and Aydede2009).
Real mills are traditionally made of wood and stone, or whatever modern materials are currently in fashion. In our unpacking of Heyes’ analogy, the mill is formed by the grist, which it then processes in such a way that changes the construction of subsequent mills. If grist can change the way the mill works, and vice versa, then perhaps grist-and-mill is not the right metaphor for understanding the cultural evolution of thinking (most mills don't reshape themselves as a result of changes in the grist that they are milling). If a metaphor is needed, a more apt one might be rivers and the water that runs through them. A riverbed channels the water that runs through a geographical area, but it can also get reshaped by that water. And the quality and flows of that water can change over time. If one embraces a river metaphor to illuminate this mutual relationship between cultural evolution and cognitive gadgets, it is easier to see how culture and brain can indeed shape one another. It also becomes clearer that culture and brain are not two separate factors that additively combine to generate mind. They are sufficiently interdependent that they might be best treated as one complex system: a distributed cognition composed of information that is transmitted via both neural fibers and social fibers.
Heyes is right to bring cultural evolution to the forefront of cognitive development. As the cognitive and developmental sciences have been slowly drifting away from extreme nativist claims, the glaring question is now, “Where do constraints on social learning come from?” As it happens, cultural and language evolution experts have been working on aspects of this question for some time now. Heyes (Reference Heyes2018) gives the reader a number of useful landmarks from those fields, but the real story of how cognitive development and cultural evolution interact is somewhat more complex than she implies. We unpack the book's central evolutionary arguments, noting some important perspectives that are omitted and highlighting the need for involvement by experts in complex systems. We will also identify problems with her central grist-and-mills metaphor. Though our tone is one of critique, we want to make it clear that we think Heyes’ book is an important contribution to the literatures on both cognitive development and cultural evolution, and we fully endorse her thesis that a deeper synergy between these often disconnected research areas is paramount.
There are two principal components of Heyes’ argument. The first is that cognitive mechanisms are not innate but are shaped during development by social forces – we might call this “gadgetization.” The second component is that the structure of these social forces is the product of cultural evolutionary processes. We address each of these in turn.
The gadgetization that Heyes writes about is an important and underappreciated idea, with thick roots in the cognitive science of the 1990s. Heyes provides compelling demonstrations of why imitation, mindreading, and language should each be seen not as innately-given cognitive instincts but as learned cognitive gadgets. In general, calling these neural mechanisms “cognitive gadgets” will be a catchy and helpful remedy for cognitive developmentalists who have been tempted to accept the notion of “cognitive instincts.” But others may want a bit more detail regarding how these gadgets are actually constructed. For example, many non-nativist cognitive developmentalists will recognize cognitive gadgets as resulting from something like the learning-based modularization process described by Karmiloff-Smith (Reference Karmiloff-Smith1994). Such processes are likely to have substantial innate architectural constraints based on gross neuroanatomy, but fewer representational constraints, because representations are based on more plastic fine-grained neuronal connectivity (Elman et al. Reference Elman, Bates, Johnson, Karmiloff-Smith, Parisi and Plunkett1996). There is already a large literature on how individuals’ environments, including the social and cultural environments, shape fundamental aspects of their cognitive machinery (e.g., Henrich et al. Reference Henrich, Heine and Norenzayan2010; Karmiloff-Smith Reference Karmiloff-Smith1994; Nisbett et al. Reference Nisbett, Peng, Choi and Norenzayan2001; Nisbett & Miyamoto Reference Nisbett and Miyamoto2005), and so we hope the additional evidence Heyes provides signals the end of overly simplistic nativist approaches to cognitive development. Beyond this statement, however, Heyes provides somewhat scant details regarding how gadgetization occurs. One of the few detailed examples given is the rather outdated dual-route model of literacy. Readers interested in a model that actually learns the way children do might be better served by Seidenberg's (Reference Seidenberg2017) neural networks.
Savvy readers will also want to know more about how cultural evolution shapes the social learning environment that imposes this gadgetization. The most interesting part of Heyes’ argument, in our view, is that the social environments that facilitate the development of culturally important cognitive mechanisms also evolve. This is an important point that has, to date, received too little attention. However, Heyes’ presentation of how the relevant evolutionary dynamics occur is somewhat thin, and glosses over both the staggering complexity of those dynamics and the considerable progress that has already been made in describing that complexity.
Heyes characterizes members of the California school as being vague and inconsistent in their descriptions of evolutionary dynamics. Let us first note that “the California school” is an unfortunate term for many reasons, not least because it enables the author to ascribe to a collective a set of viewpoints without naming any individual sources, which would require attribution and subsequently allow scrutiny of those claims. Many of those associated with the referenced “school” have gone to great lengths to individuate their own research programs and clarify exactly the sorts of transmission, inheritance, and population-level dynamics they believe to have occurred in human cultural evolution, including using computational models, archaeology, cross-cultural fieldwork, and laboratory experiments.
In contrast, the evolutionary dynamics Heyes proposes are themselves vague. Let us consider her discussion of cultural group selection, which draws little from more established discussions (e.g., Boyd & Richerson Reference Boyd and Richerson2002; Henrich Reference Henrich2004a; Richerson et al. Reference Richerson, Baldini, Bell, Demps, Frost, Hillis, Mathew, Newton, Naar, Newson, Ross, Smaldino, Waring and Zefferman2016; Smaldino Reference Smaldino2014; Wilson Reference Wilson2002; Zefferman & Mathew Reference Zefferman and Mathew2015). The key point about multilevel selection or group selection is that selection on a social trait can strongly depend on the social environment, and so the contributions of each level of selection must be taken into account to analyze the overall population dynamics (Okasha Reference Okasha2009). Altruists can be exploited by free riders within their group (individual-level selection against altruism) but still increase the overall fitness of their group relative to other groups (group-level selection for altruism). Additionally, prosocial norms often require coordination among social actors to provide marginal benefits, and therefore face difficulties in propagating in new communities (Bicchieri Reference Bicchieri2006; Boyd & Richerson Reference Boyd and Richerson2002; Richerson et al. Reference Richerson, Baldini, Bell, Demps, Frost, Hillis, Mathew, Newton, Naar, Newson, Ross, Smaldino, Waring and Zefferman2016). Understanding how such social traits spread is a key research area throughout the social sciences, encompassing not only the selection-based work in cultural evolution, but also the epidemiology of beliefs pervasive in other areas of the social sciences (e.g., Centola Reference Centola2018; Sperber Reference Sperber1996). Heyes’ evolutionary model (pp. 199–201) pays lip service to group selection, but closer inspection shows that her model is completely unaffected by group structure. In her model, trait M′ always has higher fitness than trait M, so that it will spread regardless of group structure. There is no selection at the group level that is any different from selection at the individual level. This is a mistake, of course. The traits Heyes is talking about are social and will likely have different fitness gradients at different levels of selection; it's just that there is no mention of what the different gradients might look like. Heyes does briefly mention the potential importance of social complexity, but hardly enough. The consideration of emergent social structures that are important in cultural evolution is explored at length in Smaldino (Reference Smaldino2014). There is still a lot of important work to be done on understanding these evolutionary dynamics, and we agree with Heyes that the inheritance mechanisms for social structures is an important target for future research.
What makes understanding the cultural evolution of social organization so challenging? We suspect that the missing piece from Heyes’ discussion is the complex feedback between extant traits and selection pressures and principles of organization (cf. Kauffman Reference Kauffman1993; Thompson Reference Thompson1942) that shape those traits. In terms of the cognitive gadgets discussed in Heyes’ book, some of the clearest perspectives on this front can be found in the work of Kirby and Smith (Kirby Reference Kirby2017; Smith & Kirby Reference Smith and Kirby2008). They have proposed that once humans had genetically evolved sufficient capacities for sociality, imitation, and cooperation, cultural evolution was sufficient to shape early communication systems towards easier understanding, production, and flexibility. Most importantly, they have provided dynamic models of how such a process might occur. Tomasello et al. have suggested more generally that social cooperation was a likely driver of the emergence of many uniquely human cognitive features (Moll & Tomasello Reference Moll and Tomasello2007; Tomasello et al. Reference Tomasello, Melis, Tennie, Wyman and Herrmann2012; Tomasello & Gonzalez-Cabrera Reference Tomasello and Gonzalez-Cabrera2017), which is consistent with, but not a direct consequence of, Heyes’ central arguments.
In general, the processes of feedback across multiple organizational and temporal scales needed to explain the evolution of complex systems (Caporael Reference Caporael, Scher and Rauscher2003; Caporael et al. Reference Caporael, Griesemer and Wimsatt2013; Smaldino Reference Smaldino2014; Wimsatt Reference Wimsatt, Schaffner and Cohen1974) are missing from Heyes’ story, though their importance is hinted at in her final chapter. Implicit in this discussion of feedback is that, at some point in our evolutionary past, humans lived in social environments that were not well adapted to facilitate the development of many cognitive gadgets in the forms we now know (relatedly, it is likely that other gadgets, adaptive in past environments, have since been lost). Over time, humans constructed environments which promoted the development of new cognitive gadgets, which in turn facilitated the development of new environments, and on and on. Co-evolutionary dynamics like this have, in fact, been increasingly studied under the banner of cultural niche construction (Kendal Reference Kendal2011; Laland & O'Brien Reference Laland and O'Brien2011). This body of work explicitly targets the evolutionary feedback processes by which humans modify their environments (e.g., by producing new social institutions like writing), which in turn creates new selection pressures (e.g., by encouraging literacy), which in turn creates new opportunities for modifying the social environment (e.g., by producing new divisions of labor in which some individuals are express stewards of written knowledge), and so on.
All this creates problems for Heyes’ central metaphor of grist and mills. There is a reason that the social skills essential to human cultural learning are reasonably well described as being handled by cognitive “gadgets,” in the sense that the word usually refers to cobbled-together thingamajigs. The reason is that the environments that led to the evolution of those skills were socially constructed ones – a set of social niches constructed by the same species that was itself developing those skills. These gadgets were pieced together over time by a nonlinear unguided process, and, therefore, they are not pristine engineered devices. The grist was not already there to cause the formation of the mill, nor was the mill already there to cause the formation of the grist. As Heyes herself notes on p. 203, “the inheritance mechanisms for mills overlap with the inheritance mechanisms for grist.” Thus, the social environments influencing development (e.g., the grist) co-evolved with the cognitive gadgets (e.g., the mills), bringing each other into being in a fashion not unlike autocatalysis (where two chemical reagents cause each other to come into prominence). To understand something like autocatalysis, one needs some facility with the dynamics of complex systems. Treating cultural evolution and cognitive development as though they are linear feed-forward processes that straightforwardly turn selection pressures into human traits just will not cut it. For example, the social mechanisms of language use and the neural mechanisms of language processing may not be well treated as “a grist” and “a mill,” respectively, precisely because they overlap so much with one another (e.g., Clark Reference Clark2008; Kirby et al. Reference Kirby, Cornish and Smith2008; Spivey & Richardson Reference Spivey, Richardson, Robbins and Aydede2009).
Real mills are traditionally made of wood and stone, or whatever modern materials are currently in fashion. In our unpacking of Heyes’ analogy, the mill is formed by the grist, which it then processes in such a way that changes the construction of subsequent mills. If grist can change the way the mill works, and vice versa, then perhaps grist-and-mill is not the right metaphor for understanding the cultural evolution of thinking (most mills don't reshape themselves as a result of changes in the grist that they are milling). If a metaphor is needed, a more apt one might be rivers and the water that runs through them. A riverbed channels the water that runs through a geographical area, but it can also get reshaped by that water. And the quality and flows of that water can change over time. If one embraces a river metaphor to illuminate this mutual relationship between cultural evolution and cognitive gadgets, it is easier to see how culture and brain can indeed shape one another. It also becomes clearer that culture and brain are not two separate factors that additively combine to generate mind. They are sufficiently interdependent that they might be best treated as one complex system: a distributed cognition composed of information that is transmitted via both neural fibers and social fibers.