What are the functions of System 2 modules? A reply to Chiappe and Gardner

In their response to our paper “Modularity in Cognition: Framing the Debate” (Barrett & Kurzban, 2006), Chiappe and Gardner (2012) argue that the concept of modularity we propose differs from the way modularity has conventionally been viewed in evolutionary psychology, and that it is “is so broad that it includes mechanisms traditionally rejected by EP” (p. 669) They also argue that the concept of modularity we propose can’t explain mechanisms “designed to deal with novel challenges” (p. 669).

There was a reason we chose to subtitle our paper “Framing the Debate.” Our intention was to reject existing framings of modularity, and to propose a new one. In our view, debates about modularity have been unproductive because they have been centered on a definition of modularity introduced by Fodor (1983) that in all likelihood describes only a small part of the mind’s operations: namely, modules as narrow, inflexible, automatic, encapsulated, innately specified devices. The problem is that as time has gone on, this view of modules has become synonymous with “mental adaptation.” We believe this is a false equation because there is nothing in the theory of evolution by natural selection that implies that adaptations need to be narrow, inflexible, automatic, encapsulated, or innately specified (in the sense of present at birth) and/or unshapeable by learning or experience. Instead, all functionally organized information-processing in the mind, whether or not it has the list of properties proposed by Fodor, is likely to be the result of brain systems shaped by natural selection. Therefore, we need a definition of “module”—or some other equivalent term that designates evolved mechanisms, or mental adaptations—broad enough to encompass all aspects of the mind’s activity that are the result of mechanisms shaped by selection, whether or not these are narrow, inflexible, automatic, and so on. That is the framing of the debate, or the reframing, that we intended.

Unfortunately, Chiappe and Gardner prefer the older framing that we argued is not particularly useful, equating “specialization” (and therefore adaptation) only with Fodorian modules. Part of their argument is historical, claiming that we got the facts wrong about what evolutionary psychologists have argued in the past. We are happy to concede that there has been substantial variation in the use of the term in the past, but our argument doesn’t turn on the historical details. The whole point of our effort to frame the debate was to move past old framings that, we feel, have had the very damaging consequences of separating the mind into “modular” bits, which are viewed as evolved adaptations, and “non-modular” bits, which are not. If the objection is to the word “module,” then we would be happy to discard that term in favor of some other mutually agreed upon term: “psychological specialization,” “mental adaptation,” “evolved cognitive mechanism,” or any other equivalent term that designates an aspect of mental structure that was shaped by natural selection. The important thing is this: if we can agree that natural selection shaped both the parts of the mind that have traditionally been viewed as “modular” and the parts that traditionally have not been, then we are in agreement on the main argument we were trying to make in our paper.

However, the second part of Chiappe and Gardner’s argument suggests that we are not in complete agreement. They argue that the definition of modularity that we intended, as opposed to the older Fodorian definition that we rejected, “has limited usefulness in explaining the existence of mechanisms designed to deal with novel challenges and with the development of novel solutions to longstanding adaptive problems” (Chiappe & Gardner, 2012, p. 669). Here, we disagree. The basis for our disagreement is hinted at by Chiappe and Gardner’s use of the word designed in the phrase “mechanisms designed to deal with novel challenges.” If they are using this term as we use it, then they are talking about mechanisms that evolved by natural selection: in other words, mental adaptations. Our view of modularity defines modules in precisely this way: if X is a mechanism, and if it has a design (i.e., has been shaped by the process of natural selection acting over evolutionary time), then it is what we are calling a “module.” So, the kinds of mechanisms Chiappe and Gardner intend do fall within the purview of mechanisms that we think should be analyzed using the conceptual apparatus of adaptationist evolutionary biology.

We suspect, however, that Chiappe and Gardner’s issue is more than merely definitional: they believe that natural selection acting on the design of mechanisms over evolutionary time is not sufficient to account for mechanisms that can deal with evolutionary “novelty.” On this view, specialization and the capacity to deal with novelty are seen in zero-sum terms, such that more of one means less of the other. While this view is widespread in psychology, we believe that it is a mistake, as we discuss in more detail below. However, Chiappe and Gardner go even further, suggesting that brain mechanisms come in two fundamental types: the “modular” ones, which are evolved specializations that are automatic, inflexible, domain-narrow, and innate, and the “non-modular” ones, which are not specialized, not automatic, not innate, and definitely flexible. Following the terminology of “Dual Systems” models (Sloman, 1996; Stanovich, 1999), they call these “System 1” and “System 2” mechanisms, respectively.

Chiappe and Gardner’s dual-systems view is made clear in how they contrast the two systems. System 1 mechanisms are described as follows: “Natural selection can also enable us to quickly respond to situations by producing mechanisms that automatically activate pertinent representations and responses” (Chiappe & Gardner, 2012, p. 674). They contrast this with System 2 mechanisms, which “override the automatic and obligatory computations provided by System 1” (p. 674). This distinction is reinforced by their claim that “[w]hile our individual experience and evolution may instill in us various reactions to stimuli, it is important at times to be able to prevent these reactions from influencing our behavior” (p. 674). This implies that preventing these reactions from influencing our behavior is not the product of experience or evolution—an implication with which we disagree.

Our view is not that the System 1/System 2 distinction is necessarily useless, at least when defined in terms of “automaticity”: for example, some kinds of processes do appear to respond to subjects’ self-reported “volition” more than others (Wegner, 2002). However, we believe that it is a mistake to think of the “automatic” systems as being the result of evolution by natural selection, and the other systems as being the result of something else. If our brains are designed such that some processes are “automatic” and others not, and if there exist mechanisms whose function is to “override” or in other ways inhibit or modulate the activity of the automatic systems, then those “executive” systems are also the products of natural selection. They evolved because of their beneficial fitness effects, and possess design features that produce those effects. This entails that they have functions, and are likely to exhibit the form–function fit relationship that is characteristic of biological adaptations. This, and not domain-narrowness, is what we mean by specialization.

If this is right, then an evolutionary “modular” view is likely to illuminate both the functions and functional design features of System 2 processes. There are many possibilities for what the functions of System 2 mechanisms might be, some of which have been discussed in the literature on dual systems. For example, as Chiappe and Gardner mention, executive modules might monitor the inputs and outputs of other modules and look for productive mappings between them, marshaling cooperation and synergistic interaction among systems. Such “structure mapping” mechanisms are thought to play a role in phenomena like metaphor (Chiappe, 2000; Gentner & Holyoak, 1997). Another function of System 2 modules might be, as Chiappe and Gardner suggest, inhibition of the activity of other systems (Leslie & Polizzi, 1998). In play and pretense, for example, children must decouple their play responses from the responses that “automatic” systems might generate if play were not occurring, such as aggression (Leslie, 1987). And there are likely to be many more kinds of System 2 mechanisms as well, with diverse functions and design features.

Most modular systems in biology, such as the modular genetic systems involved in growth of limbs and organs, involve modulation of some (modular) systems by other (modular) systems. So do modular human-engineered systems like computers and automobiles. In order to function properly, modular systems need interrupt switches and various other higher-level “deciders” that allocate resources like attention and control across different systems in a context-sensitive fashion. While such mechanisms would count as System 2 mechanisms in a dual-process taxonomy, they nevertheless are likely to have evolved functions and design features that were shaped because of their fitness effects in past environments over evolutionary time. Our argument is that thinking about System 2 mechanisms as adaptations with functions and design features is likely to shed much more light on them than simply describing them as “flexible” and “open.”

This brings us to the topic of evolutionary “novelty,” and whether there can be adaptations designed to deal with it. Here, we think, there is a definitional issue: at a certain level, the terms “design” and “novelty” are incompatible with each other, because adaptation is impossible without some environmental signal, even if statistical and fuzzy, to adapt to. If “novel” means “bears no resemblance to anything in the past,” then design to deal with novelty is a priori impossible. Chiappe and Gardner (2012) appear conflicted on this issue. They want to have their novelty cake and eat it too, defining “novelty” as bearing no resemblance to the past, and yet positing “mechanisms designed to deal with novel challenges” (p. 669). To be clear, we don’t think adaptations designed for novelty are impossible, but only if we redefine “novelty” so as to not make adaptation to it impossible. For example, one might use “novelty” to refer to novel tokens of types that recurred over evolutionary time, and that therefore had some recurring features (Barrett, 2006). Foods are an example: our digestive systems can handle “novel” foods, like chocolate, because they are tokens of a broader type that has features our digestive systems can handle. We might also use “novelty” to refer to some aspect of evolutionary environments that has presented a trackable, yet changing, signal over time. Mechanisms for cultural transmission, such as prestige-biased transmission mechanisms, are adaptations to “novelty” in this sense: they allow us to acquire novel cultural variants, yet their design features (which serve to bias learning in favor of high-prestige models) evolve only when there is a stable correlation between social prestige and possession of fitness-increasing information (Henrich & McElreath, 2003). Regardless of the terminology one wants to use to describe such cases, the overarching point is this: If there are adaptations to novelty, then they evolved over evolutionary time (i.e., in the past), and if they do systematically adaptive things in response to novelty, then it must be because of evolved design features that allow them to do so. Therefore, thinking about these mechanisms using evolutionary, adaptationist logic is likely to yield insights about their design.

In conclusion, we reiterate that much of the disagreement between us and Chiappe and Gardner might come down to semantics, and choice of terms. As we’ve said, we’re less interested in who said what about modularity and more interested in arriving at a view of modularity that is likely both to be biologically correct and to allow us to capture the myriad ways that natural selection shapes brain design. In our view, this means we should be talking about adaptive specializations in the brain. We are happy to use the term “module” to refer to these but are happy with any other equivalent terminology as well, as long as it is used in a clear and consistent way that does not import false dichotomies and associations, such as the false dichotomy of “evolved” versus “learned,” or the false association between natural selection and automaticity. Similarly, if the System 1/System 2 distinction ends up being a biologically real distinction in the brain, then it cannot be equivalent to the distinction between having evolved by natural selection and not having so evolved. Design features of both System 1 and System 2 must have an evolutionary origin, and must be explained in functionalist terms, both ultimately (what selective pressures brought them into being) and proximately (what design features they have that allow them to carry out their functions, including, for example, their ability to respond adaptively to “novelty”). In our view, when discussions of brain architecture are framed this way, many unproductive debates about the role of evolution in shaping the mind are likely to go away. In our opinion, that would be a good thing.