On Modules

As the next installment in my series on evolution psychology (see previous posts here and here), I thought that I would write about some thoughts on evolutionary modules.  As should be obvious from previous posts, I have serious concerns about evolutionary psychology.  Nonetheless, I don’t want to repeat the knee-jerk criticisms that attended the rise of what you might call (and Symons (1989) did call) “Darwinian Anthropology.”  Like Anthropology more generally, I have found that the level of discourse in human evolutionary studies tends to be particularly low and this surely hinders progress toward our presumably shared goals of understanding human behavior, the origin and maintenance of human diversity, and how people respond to social, environmental, and economic changes.

In this spirit, I am taking seriously the idea of modularity.  The concept of “massive modularity” seems to be pretty central to just about any definition of modern EP and it is one of the ideas that I see as potentially most problematic.  A major question that naturally arises in the analysis of cognitive modularity is: what is a module?  There are two senses of modularity that you find discussed in the EP literature. For a good review of this, see Barrett and Kurzban (2006). In his highly influential (1983) book, Fodor popularized the concept of a cognitive module.  A Fodorian module is characterized by reflex-like encapsulation of critical functions.  It is thought to be anatomically localized, inaccessible to conscious thought and has shallow outputs.  Our senses and motor systems are examples of possible Fodorian modules, as are the systems that underlie language (Machery 2007).

In contrast to the Fodorian module is the second sense of modularity found in the EP literature, the evolutionary module. Like a Fodorian module, the evolutionary module is domain-specific or informationally encapsulated.  That is where the resemblance ends though.  Rather than being defined by a list of attributes, an evolutionary module is characterized by function.  An evolutionary module is a domain-specific cognitive mechanism that has been shaped by natural selection to perform a specific task.   There is no need here to specify their characteristic operating time, the shallowness of their outputs, or their anatomical localization.

Using engineering-inspired arguments about efficiency and design, the proponents of massive modularity suggest that the brain is really a collection of domain-specific modules.  These modules drive not just the reflex-like actions of our sensory-motor systems but also govern higher cognitive processes like reason, judgment, and decision-making.  The brain is not, as we typically conceive it, a single organ.  Rather it is a collection of special-purpose information processing organs.   Needless to say, such a position has been controversial.  Among the notable critics are Jerry Fodor himself, who wrote a whole book with the sarcastic title (referring to Steve Pinker’s (1997) book, How the Mind Works), The Mind Doesn’t Work That Way: The Scope and Limits of Computational Psychology.  Another notable critic is David Buller, the ostensible subject of my last two posts.

Barrett & Kurzban (2006) suggest that much of the controversy surrounding the EP concept of massive modularity arises from confusion over what is meant by a module in the EP sense.  That is, critics are thinking about Fodorian modules when the advocates of massive modularity have something entirely different in mind. Maybe.  I’m no expert, but the argument seems plausible for at least part of the controversy.   I have my own issues with modularity but I will save that for the paper that I am writing (and for which these posts serve as sketches to hopefully help me get some thoughts straight).

One point that I will make here is a fairly orthodox criticism of modularity.  In enumerating possible evolutionary modules, and noting that such modules require domain-specific input criteria, Barrett & Kurzban (2006: 630) include “systems specialized for making good food choices process only representations relevant to the nutritional value of different potential food items.”  Really? I’m not one to fall back on the weak “culture complicates things” argument, but I do think there are other things — including ones potentially important for fitness — involved in food choice than the nutritional quality of a potential foodstuff. Perhaps an anecdote is in order here.

A long time ago, my wife and I were taken out to a fancy Chinese restaurant in Kota Tua, Jakarta by a colleague who wanted to impress us with his esoteric knowledge of a variety of Asian cuisines.  He took the initiative and ordered for the table a range of items including tripe, jellyfish, pig trotters, and chicken feet. For a variety of complex social reasons, we felt it was in our interest to not seem like naïve rubes from America.  So, we ate everything unflinchingly and with smiles on our faces. These were not things we normally would have volunteered to eat (though we now regularly get jellyfish) but the social payoffs of eating these (at the time) unappealing items outweighed whatever distaste we may have experienced.

Clearly, this is a bit of a trivial example.  I nonetheless think that it highlights an extremely important aspect of human decision-making.  The optimal decision in a one dimensional problem may change when one increases the dimensionality of the problem, particularly when the elements of your (vector) optimand trade-off.  Sometimes the optimal nutritional choice is not the optimal choice with respect to social or cultural capital.  The person’s foraging decision is presumably one that balances the various dimensions of the problem. In a less trivial example, this is what Hawkes, O’Connell and Bird and Bird are suggesting is going on with some men’s foraging decisions  (summarized in this review by Bird & Smith (2005)).  According to their model, men make energetically suboptimal foraging decisions in order to signal their phenotypic quality to political allies and potential mates.  Food choice is thus a decision that balances the potential costs and benefits of at least three fitness-critical domains (energetics, politics, and reproduction).   The same logic can be applied to that other staple of EP, mate choice.  What people say they want on pen-and-paper surveys is not necessarily what they get when they actually choose a mate.  The problem is that one’s choice of mate spills over into so many other domains than simply future reproduction.  So it’s not simply a matter of the ideal mate being out of one’s league.  Sometimes, people actually prefer a mate who does not conform to their ideal physical type.

At the very least, this point seems to require positing the existence of yet another module that integrates the outputs of various lower-level modules.  Of course, this is beginning to sound more like a generalized reasoning process, the bane of EP.

There is another usage of the term “module” that I think may have some relevance to this whole discussion.   In evo-devo, modularity refers to the degree that a group of phenotypic characters have independent genetic architecture and ontogeny.  I will call this an “evolutionary ontogenetic module” (EOM) and contrast that with an “evolutionary cognitive module” (ECM) of EP.  Sperber (2002), in his defense of massive modularity, actually discusses EOMs in passing.  Pigliucci (2008) details the various, largely divergent definitions of modularity.  I tend to think about EOMs the way that Wagner & Altberg (1996) do, wherein a modular set of traits is one with (1) a higher than average level of integration by pleiotropic effects (i.e., gene interactions) and (2) a higher than average level of independence from other trait sets.  That is, modular architecture occurs where there are few pleiotropic genes that act across characters with different functions but more such effects falling on functionally related traits. 

Modularity in the evo-devo sense is central to the evolution of complexity as well as the evolution of evolvability (the capacity of an organism to respond adaptively to selection).  Do ECMs need to be EOMs? Does this and other related concepts from evo-devo help provide a means for relating the ideas of EP or HBE to their genetic architecture and ontogenetic assembly?  I think so but I think an elaboration on this topic awaits a later post.  


Barrett, H. C., and R. Kurzban. 2006. Modularity in Cognition: Framing the Debate. Psychological Review 113 (3):628-647.

Bird, R. B., and E. A. Smith. 2005. Signaling Theory, Strategic Interaction, and Symbolic Capital. Current Anthropology 46 (2):221-248.

Fodor, J. 1983. The Modularity of Mind. Cambridge: MIT Press.

Machery, E. 2007. Massive Modularity and Brain Evolution. Philosophy of Science74: 825–838.

Pigliucci, M. 2008. Is Evolvability Evolvable? Nature Genetics 9:75-82.

Pinker, S. 1997. How the Mind Works. New York: Norton.

Sperber, D. 2002. In Defense of massive modularity. In Dupoux, E.  Language, Brain and Cognitive Development: Essays in Honor of Jacques Mehler. Cambridge, Mass. MIT Press. 47-57.

Symons, D. 1989. A Critique of Darwinian Anthropology. Ethology and Sociobiology 10 (1-3):131-144.

Wagner, G.P., and L. Altenberg. 1996. Perspective: Complex Adaptations and the Evolution of Evolvability. Evolution 50 (3):967-976.

5 thoughts on “On Modules”

  1. I would go farther and say that the idea of modularity in evolution is complete nonsense. That some physiological pathways appear to be modular is an artifact of how they evolved.

    In engineering, “modular design” is a heuristic method for designing a system from the top down by separating design needs into discrete subsystems that can function independently. It is a design strategy used to minimize labor in designing and debugging systems. The modules can be designed, built and tested as independent subsystems before integrating them into larger systems.

    What organisms function that way? The answer is none. No organism has evolved via a modular path. An organism did not separately evolve liver cells, nerve cells, muscle cells, endocrine cells, debug them separately and then integrate them together.

    One way that organisms evolved was by duplicating already existing functions (so there is redundancy), and then modifying the redundant capability to do something different. That may give the illusion of “modularity”, but it is only an illusion.

    Functions associated with the liver are found in cells associated with the liver not because a “liver module” evolved, but rather because as the archetypal liver evolved, the new “liver functional capabilities” evolved in cells associated with the earlier liver functions. It is simpler for new liver-like function to occur in cells via epigenetic modification of fetal liver precursor cells than for those same functions to evolve in some other tissue compartment.

    In many cases it is pretty obvious that physiology is not modular because functions are distributed across the entire organism. How is the heart controlled? A major factor in the regulation of the heart is the pressure in the aorta. If that pressure is low, the heart beats more to increase it. What sets the pressure in the aorta? The resistance of the vasculature down stream of the aorta. Where is the resistance of the vasculature controlled? Local to each endothelial cell.

    One can make any system appear to be modular by artificially imposing boundaries and ignoring signals that pass through those boundaries. In the case of the heart, the decision to ignore the very complex pressure signal that propagates back through the aorta ignores the distributed control system which is as much a part of the “blood pumping system” as is the muscle that provides the motive power. One could include the endothelium, but then what about the signals of O2, NO, CO2, glucose and other things that pass through the endothelium into the vascular smooth muscle where the vessel tensile stress is generated?

    I am not exactly sure what modularity is in the sense of brain activity or how it would evolve (other than the quite obvious mating call mimicking behaviors discussed earlier ;).

    To me, insisting that physiological systems are modular is akin to saying they are the irreducibly complex. A functional system comprised of multiple modules is irreducibly complex. You can’t remove one module and have the whole system continue to function. If you could remove a module and retain functionality, then how essential is that module? How modular is the functionality embodied in that module if it can be removed and sufficient functionality remains for the organism to remain viable?

    “Approximately two thirds of all knockouts of individual mouse genes give rise to viable fertile mice.” (Laurence D. Hurst and Nick G.C. Smith. Do essential genes evolve slowly? Current Biology 1999, 9:747–750.)

    This shows that the systems utilizing 2/3 of individual mouse genes are sufficiently redundant that those individual genes are not essential. Another way of looking at it is that the systems are comprised of the products of genes (proteins) and those systems are sufficiently redundant that they function without the smallest modular unit (a protein).

    I see the perception of modularity as an artifact of a top down perspective, rather than from a bottom up perspective. Things evolved from the bottom up. The imposition of a top down hierarchical view is an artifact of some individual’s perceptions, not of how things actually are.

    A top down, vs. a bottom up perspective is related to the local vs. global perspective differences observed along the autism spectrum.


    Which relates to my research on neurodevelopment and how and why that difference in perspective evolved.

  2. An argument against the idea of the brain being modular is the occurrence of synesthesia. If there is cross-talk between sensory "modules", then they can't be very robust and discrete. If the sensory modules are not robust and discrete, then what use is the module hypothesis?

    Healing of brain injury by re-routing of signals through other parts of the brain also argues against a modular structure/function. If a module can change its function into something else, then it isn't very module but is rather plastic.

    I have been trying to understand the various definitions of modularity as it pertains to EP, and it seems to me that the definition of "module" is being modified so it fits what is observed, rather than "modular" being a useful way to characterize and understand what is there.

    The more I think about it, the more I think that the whole concept of modularity is simply an artifact of how the various properties of cells and tissue compartments are arbitrarily conceptualized and categorized. Modularity relates to the conceptualization, not to the cells or to their functions under different circumstances (which remains mostly unknown).

    For example a particular nerve cell has various inputs and outputs, some of which relate to the neural activity of other nerve cells. It also has inputs and outputs that relate to supply of O2, glucose and other nutrients. That makes that nerve cell part of the control system for the heart, lungs, erythropoiesis, gluconeogenesis, insulin secretion, and everything else the blood carries. When that nerve cell was growing, it produced signals that made a niche for itself and so it regulated the surrounding cells including the skull. The mass of the skull (including that nerve cell) regulates the strength of the other bones and the skeletal muscles to support them.

    In modular design in engineering (which is the only way I can think of modular design), the different functions are instantiated in modules and the modules only interact through defined interfaces. A module that consumes electrical power is separate and operates independently of the module that supplies electrical power. There may be a control module that acts as an interface, but that control module is independent of the power source and the power sink. That is simply not how physiology works. The systems in physiology are not independent. They are coupled. One may treat them as independent modules for simplistic analysis, but that is a useless approach to understanding how they evolved because they did not evolve down a modular path.

  3. Regarding the absurdity of modules, I'm afraid the idea is here to stay for a while. I am not going to defend the evolutionary psychology sense of an evolutionary module, but I think the evo-devo sense is, well, sensible. You mention duplicating existing functions. This is indeed seen as one of the primary mechanisms for the ontogeny of modularity: gene duplication (and here I use the term "gene" very broadly -- it includes duplication of large stretches of DNA all the way to whole-genome duplication). The duplication of a regulatory gene, for instance, eases the selective pressure on the duplicate and allows it to be recruited for new functions. It is through this mechanism that, as I wrote in the post, pleiotropy can be extensive within a developmental module and minimal across modules. Hansen (2006: 145) writes on gene duplication:

    "A recruited gene may normally come with heritable pleiotropic effects on other characters (its original function) and may act to increase the complexity and pleiotropy of the genotype-phenotype map. In contrast, a gene duplication may produce a new gene that is similar to an old gene, and have the same pleiotropic and epistatic effects. Importantly, gene duplication can also generate genes that are more specialized through subfunctionalization, where each duplicate loses different parts of the original functions (e.g., by losing different regulatory regions). Force et al. (1999) showed that subfunctionalization is a likely outcome of a gene duplication event. A subfunctionalized gene may conduct only a subset of the functions of the old gene and may thus be less burdened by pleiotropic and epistatic constraints. Subfuctionalization may lead to the evolution of modularity (Force et al. 2005) and may indeed be a way to evolve new autonomous evolutionary characters through parcellation in the sense of Wagner & Altenberg (1996)."

    From my reading of this literature, it seems likely that there is a lag in the intensity of pleiotropic effects following duplication. That is, duplication of a regulatory gene eases selection on the duplicate but it takes a while for divergent evolution to act on that complex. My main concern with cognitive modules is how selection could possibly mold all these modules independently in the presence of extensive epistasis over the short time periods in question (i.e., the 1.8 My of the Pleistocene).

    Regarding Stanley, I'm afraid that I (1) am familiar with his work and (2) am not a fan. I think the whole "scale-free" network thing is a whole lot of nonsense. His group wrote a paper in 2001 (Liljeros et al. 2001) that made a mess of STD epidemiology. They argue that sexual networks are "scale free" because a censored plot of the survivor function of the sexual degree distribution on double-logarithmic axes is approximately linear. They conclude that all sexual networks (generalized from their Swedish sample) form by preferential attachment. They claim that the implications of their results are that public health interventions need to be "radically different" by using a total straw-man argument (i.e., that any epidemic control efforts might be randomly allocated). Please. Along with UW statistician Mark Handock, I have written a series of papers that are critical of the approach they use (Jones and Handcock 2003a,b; Handcock and Jones 2004). By specifying competing stochastic process models and employing likelihood-based multi-model inference, we show that the most likely process driving the formation of three different sexual networks (Sweden, USA, Uganda) is a stopping rule (i.e., looking for Mr./Ms. Right). Epidemic thresholds exist and standard epidemic control strategies will work. The real problem is one of funding and political will. In addition to these papers, we have a more discursive working paper that is a rebuttal of a paper their group wrote criticizing us.



    Handcock, M. S., and J.H. Jones. 2004. Likelihood-Based Inference for Stochastic Models of Sexual Network Evolution. Theoretical Population Biology 65:413-422.

    Hansen, T. F. 2006. The Evolution of Genetic Architecture. Annual Review of Ecology Evolution and Systematics 37:123-157.

    Jones, J.H., and M. S. Handcock. 2003. An Assessment of Preferential Attachment as a Mechanism for Human Sexual Network Formation. Proceedings of the Royal Society of London Series B-Biological Sciences 270:1123-1128.

    Jones, J.H., and M. S. Handcock. 2003. Sexual Contacts and Epidemic Thresholds. Nature 425:605-606.

    Liljeros, F., C. R. Edling, L. A. N. Amaral, H. E. Stanley, and Y. Aberg. 2001. The Web of Human Sexual Contacts. Nature 411 (6840):907-908.

    Pigliucci, M. 2008. Is Evolvability Evolvable? Nature Genetics 9:75-82.

    Wagner, G.P., and L. Altenberg. 1996. Perspective: Complex Adaptations and the Evolution of Evolvability. Evolution 50 (3):967-976.

  4. I wasn’t aware of that STD transmission work. I quite agree with your analysis. That is an unfortunate misapplication of modeling.

    I don’t have a good understanding of what a module is and how it is defined. I suspect that there is no good (i.e. unique and unambiguous) definition. Without a good definition and the ability to decide when a “module” first appears and how it changes from generation to generation I think it is a concept of very limited utility.

  5. Pingback: Hitting the Blue

Leave a Reply

Your email address will not be published. Required fields are marked *

* Copy This Password *

* Type Or Paste Password Here *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>