When sociobiology passed on its mantle to evolutionary psychology, it seemed that the emphasis shifted away from rigorous mathematical theories in population genetics to plausibility arguments for psychological mechanisms that do not incorporate selection as thoroughly as the theoretical biologists did. The clean-cut equations of Maynard Smith, Williams, Hamilton, Trivers and Price et al are taken as part of the argument for why evolutionary psychology is plausible. However, the more specific hypotheses constituting evolutionary psychology are not supported by similarly rigorous evolutionary reasoning. The standard Tooby-Cosmides argument for massive modularity in our mental architecture considers all the advantages modularity has over domain-generality, but fails to discuss any possible additional energy costs modularity might incur, or whether the details of genetics and molecular biology could possibly conspire such that it is more difficult to acquire many specific adaptational mental modules rather than one monolithic general reasoning device.
I have not found any literature discussing the energy disadvantages a modular brain might have compared to a domain-general brain. I think it is plausible that there are such energy disadvantages, simply because with massive modularity, most specialised mechanisms are not operating most of the time, coming into action only in certain environmental conditions. Therefore, in some sense, these mechanisms are lying dormant most of the time. A domain-general brain, on the other hand, would not have such specialised mechanisms lying dormant. In short, at any given point in time, a domain-general brain has less redundancy — it might not be using its full range of computational resources, but it doesn’t have a massive suite of information processing programmes that are just standing around twiddling their thumbs, as such. Since our brain is not a solid state flash drive, just maintaining these programmes in a dormant state consumes energy, so a massively modular brain would have to channel energy towards the upkeep of these programmes even when they are not working, which most of them aren’t most of the time. The domain-general brain would seem to be free of this burden. So goes my plausiblity argument for why massive modularity could be more costly than domain generality. I have tried Googling for discussions of the energy costs of massive modularity, but have come up blank. Robert Richards, too, says he doesn’t know of any arguments on this score. I suspect that sociobiologists, if they, rather than psychologists, were the ones spearheading the evolutionary psychology movement, would not tolerate having an evolutionary theory that does not take into account the costs as well as the benefits of a particular adaptation.
I also could not find any research on whether it might have been ‘easier’ (more probable) for organisms to acquire massive modularity (or to acquire domain general information processing mechanisms). My intuition is that there must be some bias, in the landscape of genotypes, towards a particular class of mental structures. In other words, I would be very surprised if, in the topology of genotypes, the measure of genotypes with massively modular mental architectures was equal to the measure of genotypes with domain general mental architectures. Now, of course, natural selection would, if we assume equal energy costs for both alternatives and accept the evolutionary psychologists’ arguments for the evolutionary advantages of massive modularity, tend to favour the genotypes with massive modularity, so measure alone, without consideration of fitness functions, cannot tell us everything. But I do not think it is entirely implausible that massive modularity could have a negligible measure compared to domain generality, and if this were so, it would be plausible that we may attain a local fitness maximum on the set of genotypes coding for domain generality and remain there, simply because the set of genotypes coding for massive modularity was, metaphorically speaking, located in an obscure, tiny and relatively inaccessible part of the landscape of possible genotypes, such that even millions of years of recombination and mutations have failed to transport humans to that location.
Such considerations I find more important than the kind of things the psychologists go on about. I do think that evolutionary psychology could do with an injection of good old fashioned mathematical population genetics. Fuzzy arguments make my head spin.
Addendum: The Sperber article I linked to above discusses the problem of allocating energy amongst modules. The idea is that we do not need all modules to be active all the time, so it makes sense to have an energy allocation algorithm whereby energy is allocated preferentially to the modules that would have the largest cognitive benefits. This brings up another possible way in which massive modularity could be more energy-consuming — it demands that the brain have a informational relevance monitoring system and energy consumption prediction system to allocate energy between modules, and these systems naturally consume energy themselves.