The Evolution Away from Modularity

Mark Safranski of ZenPundit, in his post “The Movement Toward Modularity,” writes:

Pulling apart segments of an organization and reassembling them to fit the conditions of a new and different scenario is a description of modularity, a critical principle for “managing complexity” (This capacity, incidentally, also increases organizational resilience by increasing the internal link density of the entity). Ideally, with modularity you want to have an organization where the parts, while able to function independently if need be, achieve net gains in effeciency and parameters of capabilities by integrating into a synergistic network.

If your organization must make decisions in a chaotic, “noisy” environment then modularity offers a significant advantage. Unsurprisingly, with war being the ultimate in disorderly environments, the U.S. Army has begun to experiment with a ” modular” structure though the costs and the execution are proving controversial. The next evolutionary step in organizational modularity will be when the modules of an organization are able to self-organize in terms of reacting to an event without requiring central direction to ” pull them apart”. In other words,
” smart modularity”.

The body exists in a chaotic, noisy environment, so is the brain modular? In the fourth chapter of Adapting Minds, David Buller (with Valerie Gray Hardcastle) answer the answer is on. They recontextualize famous psychological tests to give a stunning complex against the modularity thesis, arguing that the mental organ is actually holistic (or, rather, domain-neutral).

My notes below don’t’ do justice to their writing. But perhaps they will be enough for handy-reference to write a paper combining genetic factors in politics with PNM Theory…

Topic: Complex Systems
“B cells don’t even have genes for each antibody. Ratherm, they possess mere gene fragments from which they assemble, on the stop, the genes necessary for building antibodies.” (Buller 140)
“domain-general social learning wouldn’t result in the overgeneralization of an acquired domain-specific solution, it would result int eh acquisition of (imitated) solutions specific to each problem domain in which it operated. In short, the domain-general mechanism would generate domain-specific solutions.” (Buller 145-145)
“The crucial point here is the distinction between information and the mechanism in the mind that process information.” (Buller 149) (but in programming, an array is information that acts like a process)

Topic: Brain Anatomy
“given the complexity of the brain, it appears that its higher cognitive structures are vastly underspecified genetically compared to its more peripheral sensory structures.” (Buller 130)
“Unlike the cells produced in the ventricular zone, the cells produced in the sub-ventricular zone, which makes up the cortex, must actively migrate to their final destinations in the brain, wending their way through a thicket of other cells.” (Buller 132)
“Central to [neurological development] is cell competition and death, in which the cells with the strongest patterns of innervation retain their connections and the other cells die.” (Buller 132) (see similar descriptions in Bloom’s Global Brain
“the best explanation for the rapidity of this response [in neurological change in response to severed digits or limbs] is that silencing the inputs to the medial nerve unmasks secondary inputs from other nerves.” (Buller 138)”
“Imaging studies [of the visual region of the brain lighting up during certain smells] indicate that the somatosensory cortex (which processes the information about touch, pressure, and joint position] projects back to the visual cortex… The degree of informational overlap in our brains shows that brain circuits are not domain specific but they are domain dominant” (Buller 139)
“development of the brain regions that control sexual and emotional responses is primarily under the control of additive events, unlike cortical development… certain sexual and emotional responses may indeed be adaptions themselves [thus] Evolutionary Psychologists may be right about some of our more basic emotional adaptions, but nonetheless wrong in its claims that we possess a lot of cognitive adaptions” (Buller 142-143)
“The limbic system.. produces fear in response to.. all coiled shapes (hoses, ropes, a circular pile of brush [, snakes, etc]” (Buller 143)
“Consequently, an argument that shows the mind must possess innate knowledge of which behaviors are adaptive in particular circumstances (since it wouldn’t be able to glean this knowledge from experience) doesn’t entail that the mind must consist of “hundreds or thousands” of distinct domain-specific mechanism.” (Buller 148)

Topic: Cognition
“a mind is equipped with some innate domain-specific information that is processed by a domain-general mechanism employing domain-general rules, which are also used in acquiring additional information about the world.” (Buller 149) (this sounds like interactionism, but how to square with domain specified learning? see definition of “learning” below)
“Tooby and Cosmides believe that what is true of language acquisition [simply unlearnable by domain-general rules] is true of most of the complex problem-solving competences that humans possess [and thus] most of the complex tasks that people can perform… are simply unlearnable… but… language learning could be an exception… language involves a very complex system of rules [and in other fields] necessary domain-specific learning rules could take the form of minimal innate biases.” (Buller 157-158)
“in the Watson selection task [most] subjects choose [the wrong cards but in contrast] choose the logically correct cards when presented with the “drinking-age problem”” (Buller 164) (this purports to show a cheating detection mechanism, but Buller’s suspicious. Ultimately Buller’s line of reasoning relies on different meanings of “if… then…” and would be even more problematic may be wrong if people find pleasure in catching cheaters. the author concludes…)
“[Evolutionary Psychologists interpret the] Watson selection tasks [as meaning] subjects appear to reason more effectively about social contracts than about non-social contracts… subjects ignore the logical properties of the condition they’re evlauation and focus exclusively on whether someone is receiving a benefit without playing the correspond cost… subjects… focus on whether someone has taken a benefit from them without paying a cost to them.” (Buller 171)
“The theory behind the cheater-detection mechanism module should lead us to expect a mechanism that is specialized in detecting cheaters in the domain of social exchanges. But the experimental results that purportedly support the existence of a cheater-detection module involve detecting cheaters int eh domain of social contracts.” (Buller 172) (a problem for EP or a distinction without a difference?)
“the reason subjects perform poorly in Watson selection tasks involving arbitrary indicative conditionals is that most subjects don’t fully comprehend the conditionals in those tasks… performance with indicatives can be as high as performance with deontic conditions, provided that the indicatives are natural” (Buller 180,182)
“autism appears instead to prevent individuals from being able to damp down the total array of irrelevant inputs to the brain.” (Buller 192)
“deaf children of hearing parents perform comparably to autistic children on the false-belief test. The reason Peterson and Siegal found, is that hearing parents do not attempt to communicate with their deaf children about abstractions such as mental states.” (Buller 193)
“Perhaps those autistic children who are able to pass the false-photograph test have acquired some principles of counterfactual reasoning, which accounts for their success in this task.” (Buller 194-195). compare to:

“If Asperger’s people are good systemizers and bad empathizers, with extremely-male brains, the thought arises that there are probably people who are good empathizers and poor systemizers, with extreme female brains.” (Ridley 62)

Topic: Definitions
massive modularity thesis .. there are “hundreds or thousands” of modules comprising the human brain.” (Buller 127)
modules… are specific, functionally dedicated to solving a specific range of very closely related adaptive problems… they develop in the absence of explicit instruction in the problem domains in which they specialize… embody innate knowledge about the problem-relevant parts of the world… informationally isolated from cognitive processing occurring in other parts of the mind… information isolation enables modules to be comparatively fast t solving problems … strongly resemble what were traditionally called instincts.” (Buller 128)
Additive events are the formation of neurons, their migrations to their final locations in the brain, and their axonal branching to establish connections with other neurons… Cortical development consists of additive events which overproduce neurons and connections…” (Buller 133-134)
Subtractive events are axonal retraction (whereby neurons lose their connections to other neurons) and cell death…. subtractive events selectively eliminate neurons and connections.” (Buller 133-134)
proliferate-and-prune can produce relatively stable brain circuits that specialize primarily in particular information-processing tasks.” (Buller 134)
neural plasticity… refers to the ability of brain regions to perform different functions, so that a given brain region has the capacity take on the function of any other region… a phenotypic plasticity… a domain-general mechanism with respect to behavioral response” (Buller 138,147)
adaption … there was variation with respect to the trait… individuals with the trait… had higher fitness…. and the difference between having and not having the trait was due (at least in part) to a genetic difference…” (Buller 140-141)
learning .. is also a process of acquiring many domain-specific rules, which are used in solving problems in their proprietary domains.” (Buller 155)
“The input problem arises when the properties to which a module is allegedly responsive aren’t things that can be directly detected by the sense but most be inferred from things that can be detected by the senses.” (Buller 159-160)
“an altruistic act is an act that exacts a fitness cost from the actor while providing a fitness benefit to another individual who is unrelated to the actor” (Buller 161)
reciprocal altruism and what Cosmides calls a social exchange [means] in order for A to perform an altruistic act to B, A must accrue a fitness benefit that outweighs the cost incurred by acting to benefit B .. A must accrue a fitness benefit that outweighs [his cost, meaning] A must perform an act that Benefits B at some cost to A.. Second, B must reciprocate by performing some act that benefits A at some cost to B. Third, the fitness benefit to a of B’s act must be greater than the fitness cost of A of A’s act.” (Buller 161-162)
bilateral cheating option… each party to the contract has the option of cheating the other.” (Buller 168)
social exchanges… occur when two individuals perform acts that benefit one another at a cost to each other.” (Buller 171)
indicative conditionals are formed by making the truth of one fact-stating sentence, Q, conditional upon the truth of another fact-stating sentence, P.” (Buller 174)
deontic conditions are used to impose obligations. Rather than making the truth of one statement conditional upon the truth of another, they make an obligation conditional upon the truth of a fact-stating sentence… all deontic conditionals have the logical form of Old Testament commandments .. [and] rewording deontic conditionals [as Commandments facilitates] performance. Whereas approximately 67 percent of subjects selected the P and not-Q cares in response to the usually formulated social-contract rule, a full 90 percent of subjects selected those cards in response to the Fodorized rule.” (Buller 174-176)
cognitively degraded [means objects] don’t possess sufficient properties to facilitate the cognitive processing that results in accurate representations in logical form.” (Buller 180)

PS: If you made it this far, congratulations! The Gene Expression review of this book I previously mentioned adds an interesting nugget: Dan Sperber, who is a critic of the Watson test mentioned above, is also a strong proponent of massive modularity.

3 thoughts on “The Evolution Away from Modularity”

  1. Great expansion and counterpoint. Not sure if modularity is always antagonistic to holistic conceptions because modules can and should act in synergy, at least at certain times but at other times, it probably is.

    Have to link to this….

  2. Mark,

    Good point about the synergy. I was delighted by how this chapter tied into recent blog conversations [1], but what is actually meant by modularity is not clear to me.

    In the above notes I mentioned how Buller and Hardcastle’s data-and-process division wasn’t clean, because in programming an array is data that looks like a process. You can do the same thing with the authors’ insistence that modules have access to direct information. For example, what to make of

    a PERL script that (depending on command line arguments)
    involves a MACRO language that (depending on flags)
    compiles C++ code that (depending on debugging code)
    builds an ASP presentation layer that (depending on browser settings)
    invokes an SQL command that (depending on the database’s configuration file)
    is converted into DATABASE ACCESS LOGIC (depending on current performance load)
    that is outputed as HTML that (depending on a user’s current time)
    involves an SOAP request that (depending on Internet traffic)
    is returned in XML (which in turn depends on another computer’s settings — a whole new black box there) that
    is processed by JAVASCRIPT and shows a user either his waiting emails or the nightly TV schedule

    Here you have a cascade of modules, some clearly “deeper in” that other, all of which takes some input and much of which relies on real-time or near-real-time settings. Would this be not massively modular in Buller’s and Hardscaste’s eyes because some require information from an internal clock, some from settings that were recently set, others from settings that were set further back, some from social interactions?

    If anything, this chapter (and your comment) helps clarify that definitions are important. Oddly the authors have attacked others for definitions so vague as to be meaningless. But in a chapter on “modularity,” they seem to do the same thing.


  3. No offense but Bullers “not enough genes” argument relies on a “one gene = one trait” view that is just plain wrong. See Barrett & Kurzban (2006), page 640.

    Also it’s Wason selection task, not Watson 😉 (named after Peter Wason).

    Other than that, great post.

Leave a Reply

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