Atoms of thought

Latest revision as of 04:38, 12 February 2009

I haven't dug much into the model/theory of cognition (http://en.wikipedia.org/wiki/ACT-R) called ACT-R. I definitely haven't read anything formally written by its promoters. I know it is strongly associated with a 'reducible atoms of thought' approach to cognition and I will briefly share my apprehensions with such an approach (which I feel more comfortable discussing separate from ACT-R specifically, which I've admitted that I know nothing about). This helps me get my theories in a row.

Clearly, atoms of thought are a useful approximation of whatever is actually happening in the mind. There are inumerable results that are easily understood by applying the idea of such basic units of thought. Further, no existing more distributed approach to cognition is cohesive enough to account for all the observations that atomic approaches like ACT-R do.

However, there is also much work that confounds the simple picture of atoms, and there is good reason to believe that complicated systems well modeled as 'atoms' emerge from the complex relationships of many neurons at many scales. If this is true, there is no reason to think that the 'atom of thought' scale is clearly abstracted from the lower messy scale. Further, in my green opinion, cracking the brain will be the process of understanding how modules and centralized control mechanisms emerge, develop and evolve from the messy interactions below.

Atomic models will give insight into the process of thought to the extent that their assumptions hold. But understanding when the assumptions hold, and why, will require tools for analyzing and understanding scale and emergence. Any understanding of the mind that lacks a clear understanding of these confounds is incomplete, particularly when they may well be the rule and not the exception in the day-to-day life of the mind. And such 'distributed' tools, if developable, will also provide models for distributed approaches to the mind that are cohesive enough to account for all the observations that atomic approaches like ACT-R do.

So, for my money, I'm betting on emergence.