On the whole, Anderson's theoretical framework appears plausible and advances a flexible computational architecture for brains. Although this framework works well in the abstract, there are several points for further refinement and investigation. Our first suggestion is to better constrain the concept of reuse in order to set clear criteria for evidential support. One way to do this is by focusing on previous adaptive functions, original use. Until we have some sense of the functions that specific parts were optimized to perform in the past, it remains unclear how such parts might (or might not) be reused. Reuse promises (among other things) to go beyond original use. But how do former functions of neural components constrain the possibilities for reuse, if at all? Anderson is largely silent on this account, perhaps advantageously at this stage. Casting the theory abstractly leaves plenty of room for it to be generally accurate, and avoids objections to uncertain particulars. However, filling in more details will eventually be required for the theory to gain explanatory and predictive traction.
Anderson's discussion of modularity could benefit from additional examples, narrowing in the specific thesis of reuse. Modularity is a versatile – perhaps too versatile – concept. Is “massive modularity” a thesis about the size (crudely analogous to mass) or scale of the modules, the large number of modules (whatever their size), or the ubiquity of modular architecture in brains? Carruthers' (Reference Carruthers2006) comparison with hi-fi components may have misled Anderson. A better parallel might be the random number generator and the graphics processing card in a laptop, which can vary independently, and interact in many different applications. However, probably any parallel with technological modules is of very limited utility, since no such module exhibits the sorts of plasticity that neural tissue is known to enjoy. Sperber (Reference Sperber1996), for instance, is a proponent of modularity, but he insists that modules are there to be exploited to meet new demands. Anderson might categorize Sperber's (1996; Reference Sperber2001) views as more closely aligned with reuse than massive modularity, but this suggests fuzzy boundaries between modularity and potential alternatives. A software theory of massive modularity – programs evolved to serve particular adaptive functions within brains – without commitments about implementation (unlike anatomical modularity), could survive Anderson's critique largely untouched. The grain and level of analysis where modularity is applied can make all the difference.
An important point for clarification concerns Anderson's occasional conflation of two partially overlapping (classes of) hypotheses. Reuse and multi-use should be better distinguished. Reuse theories form a set of related hypotheses. Multi-use is a larger set, including cases where original function is lost, as well as cases where original function is preserved (preservation is a defining attribute of Anderson's reuse theory). The term “reuse” strongly suggests exaptation, and Anderson is explicit that his reuse differs from typical exaptation by proposing that components continue to serve some previous adaptive function while also becoming available to “time share” new functions (though he doesn't put it in exactly those terms). Anderson takes the multiplicity of functions – a brain area being activated by multiple different tasks – as evidence for reuse. However, if multi-use is an available move in design space, what reason do we have to assume that original function is preserved? Without preserving original function, reuse is an inaccurate account, and adaptation to multi-use is more accurate. The case for multi-use is strong, but all of the evidence cited implicating multi-use, while consistent with the reuse hypothesis, is not evidence for the more specific hypotheses of reuse. This ties in with our first point. Until the original use of components is specified, along with examples, Anderson hasn't yet made the strong case for reuse.
To illustrate our suggestion that Anderson's theory should be fleshed out with details, we conclude with a specific example. As mentioned above, the picture of reuse that Anderson offers appears analogous to a time-sharing model: (1) At any given time, one high-level process uses the “workings” of multiple lower-level areas, and (2) numerous high-level processes are hypothesized to alternately access a common pool of specialized lower-level resources. While this account may be accurate, we wish to highlight an alternative that focuses on a finer mechanical grain, such as individual neurons (or perhaps small collections of neurons, such as minicolumns). It is possible that specialized brain areas contain a large amount of structural/computational redundancy (i.e., many neurons or collections of neurons that can potentially perform the same class of functions). Rather than a single neuron or small neural tract playing roles in many high-level processes, it is possible that distinct subsets of neurons within a specialized area have similar competences, and hence are redundant, but as a result are available to be assigned individually to specific uses (similar to the way that redundancies due to gene duplication provide available competences for reassignment, leaving one copy to perform the original function). Over development or training, subsets of neurons in a specialized brain area could then be recruited for involvement in distinct high-level processes. This model emphasizes multi-potential of neurons, but single-use of individual neurons, as determined in the course of development and learning.
In a coarse enough grain, this neural model would look exactly like multi-use (or reuse). However, on close inspection the mechanism would be importantly different. In an adult brain, a given neuron would be aligned with only a single high-level function, whereas each area of neurons would be aligned with very many different functions. This model of multi-potential and single-use may account for all the data that Anderson cites in support of reuse, and it also avoids time-sharing for specific neurons. Whether or not the model sketched here is accurate, it illustrates the kind of refinement that could make Anderson's abstract theoretical proposal more concrete, and perhaps subtly improved.
You have
Access
On the whole, Anderson's theoretical framework appears plausible and advances a flexible computational architecture for brains. Although this framework works well in the abstract, there are several points for further refinement and investigation. Our first suggestion is to better constrain the concept of reuse in order to set clear criteria for evidential support. One way to do this is by focusing on previous adaptive functions, original use. Until we have some sense of the functions that specific parts were optimized to perform in the past, it remains unclear how such parts might (or might not) be reused. Reuse promises (among other things) to go beyond original use. But how do former functions of neural components constrain the possibilities for reuse, if at all? Anderson is largely silent on this account, perhaps advantageously at this stage. Casting the theory abstractly leaves plenty of room for it to be generally accurate, and avoids objections to uncertain particulars. However, filling in more details will eventually be required for the theory to gain explanatory and predictive traction.
Anderson's discussion of modularity could benefit from additional examples, narrowing in the specific thesis of reuse. Modularity is a versatile – perhaps too versatile – concept. Is “massive modularity” a thesis about the size (crudely analogous to mass) or scale of the modules, the large number of modules (whatever their size), or the ubiquity of modular architecture in brains? Carruthers' (Reference Carruthers2006) comparison with hi-fi components may have misled Anderson. A better parallel might be the random number generator and the graphics processing card in a laptop, which can vary independently, and interact in many different applications. However, probably any parallel with technological modules is of very limited utility, since no such module exhibits the sorts of plasticity that neural tissue is known to enjoy. Sperber (Reference Sperber1996), for instance, is a proponent of modularity, but he insists that modules are there to be exploited to meet new demands. Anderson might categorize Sperber's (1996; Reference Sperber2001) views as more closely aligned with reuse than massive modularity, but this suggests fuzzy boundaries between modularity and potential alternatives. A software theory of massive modularity – programs evolved to serve particular adaptive functions within brains – without commitments about implementation (unlike anatomical modularity), could survive Anderson's critique largely untouched. The grain and level of analysis where modularity is applied can make all the difference.
An important point for clarification concerns Anderson's occasional conflation of two partially overlapping (classes of) hypotheses. Reuse and multi-use should be better distinguished. Reuse theories form a set of related hypotheses. Multi-use is a larger set, including cases where original function is lost, as well as cases where original function is preserved (preservation is a defining attribute of Anderson's reuse theory). The term “reuse” strongly suggests exaptation, and Anderson is explicit that his reuse differs from typical exaptation by proposing that components continue to serve some previous adaptive function while also becoming available to “time share” new functions (though he doesn't put it in exactly those terms). Anderson takes the multiplicity of functions – a brain area being activated by multiple different tasks – as evidence for reuse. However, if multi-use is an available move in design space, what reason do we have to assume that original function is preserved? Without preserving original function, reuse is an inaccurate account, and adaptation to multi-use is more accurate. The case for multi-use is strong, but all of the evidence cited implicating multi-use, while consistent with the reuse hypothesis, is not evidence for the more specific hypotheses of reuse. This ties in with our first point. Until the original use of components is specified, along with examples, Anderson hasn't yet made the strong case for reuse.
To illustrate our suggestion that Anderson's theory should be fleshed out with details, we conclude with a specific example. As mentioned above, the picture of reuse that Anderson offers appears analogous to a time-sharing model: (1) At any given time, one high-level process uses the “workings” of multiple lower-level areas, and (2) numerous high-level processes are hypothesized to alternately access a common pool of specialized lower-level resources. While this account may be accurate, we wish to highlight an alternative that focuses on a finer mechanical grain, such as individual neurons (or perhaps small collections of neurons, such as minicolumns). It is possible that specialized brain areas contain a large amount of structural/computational redundancy (i.e., many neurons or collections of neurons that can potentially perform the same class of functions). Rather than a single neuron or small neural tract playing roles in many high-level processes, it is possible that distinct subsets of neurons within a specialized area have similar competences, and hence are redundant, but as a result are available to be assigned individually to specific uses (similar to the way that redundancies due to gene duplication provide available competences for reassignment, leaving one copy to perform the original function). Over development or training, subsets of neurons in a specialized brain area could then be recruited for involvement in distinct high-level processes. This model emphasizes multi-potential of neurons, but single-use of individual neurons, as determined in the course of development and learning.
In a coarse enough grain, this neural model would look exactly like multi-use (or reuse). However, on close inspection the mechanism would be importantly different. In an adult brain, a given neuron would be aligned with only a single high-level function, whereas each area of neurons would be aligned with very many different functions. This model of multi-potential and single-use may account for all the data that Anderson cites in support of reuse, and it also avoids time-sharing for specific neurons. Whether or not the model sketched here is accurate, it illustrates the kind of refinement that could make Anderson's abstract theoretical proposal more concrete, and perhaps subtly improved.