I know that the concept of working memory may be subject to discussion. Still, there is a storage capacity limit. This is a real mystery to me. Working memory is known to be positively correlated to various tests (IQ for example). Why did natural selection not grant us a better storage capacity?

Before asking this question, I read some hypotheses by Cowan: The Magical Mystery Four: How is Working Memory Capacity Limited, and Why? Two camps are presented: capacity-limit-as-weakness and capacity-limit-as-strength. Basically the first camp explains this weakness from a neuronal point of view, claiming that keeping many items active is greedy and can cause interference. The second camp pretends that the lesser the number of items accessible is the quicker the cognitive processes are (chunking, searching, etc.). Honestly I don't know what to think. More than that, I don't feel its very convincing. If our neural system has evolved toward some incredibly powerful processing modules and can't even keep 10 items active at the same time, there must be a reason. I am more inclined to think that the capacity limit is a kind of strength somewhere, but I don't feel very convinced by the examples provided in the short passage I mentioned.

Do you have any extra arguments for/against any of these theoretical positions?
My preconception is that this limit constrains us to encode the real in the most powerful fashion. In short, we are always forced to learn more complex patterns. Is it relevant or not? Do you have any source which could challenge or back up this view?

  • $\begingroup$ I've read somewhere a notice, that the researches on autistic savants suggests, that our brain development has reached it's natural limits. More working memory could weaken the central coherence etc. I'd be interested to find some reasearch in that area too. $\endgroup$ – Danubian Sailor Apr 14 '14 at 8:50

There's a huge amount of perspectives you could take in answering a question like this, but I would like to approach it from the perspective of dual process theories (see the Evans, 2008 Annual Review paper for an outline). I'm not going to reference this answer much beyond that, because almost all the information is adapted from this review, with some explanation.

There's any number of specific dual process theories, but they all state that cognition is divided between type 1 processes, which are fast, parallel, automatic, implicit, and generally don't vary between people, and type 2 processes, which are slow, serial, controlled, explicit, and crucially, rely on working memory.

The evolutionary side to this is type 1 processes are the specialized, domain specific evolved functions of our brain to deal with specific problems, like predicting the destination of moving objects, recognising faces, or figuring out how similar lions are to tigers. Type 2 processes (i.e. working memory), on the other hand, are domain general, in that they can be flexibly applied to any domain, rather than just the task they evolved to solve, but this flexibility comes at cost of limited capacity, increased mental effort, and so slower processing.

This limitation is referred to in the dual process literature as 'the singularity principle' (Evans, 2006) - system one, by processing in parallel, can deal with an incredible amount of information at the same time (and needs to, because it takes care of the basic functions that keep us alive from day to day), while the slower type 2 processes can only actively process one piece of information at a time (but can keep "7 plus or minus 2" "chunks" of information in mind if they don't have to be manipulated; Miller, 1956).

...By analogy, think of the difference between the old-fashioned telephone in your house, which was designed for only one function, and so performs it perfectly, with minimal resources, and a smartphone, which can theoretically perform any computational or communicative function, but occasionally disconnects your calls for no apparent reason.

Some References

Evans, J. S. B. (2008). Dual-processing accounts of reasoning, judgment, and social cognition. Annu. Rev. Psychol., 59, 255-278.

Evans, J. S. B. (2008). Dual-processing accounts of reasoning, judgment, and social cognition. Annu. Rev. Psychol., 59, 255-278.

Miller, G. A. (1956). The magical number seven, plus or minus two: some limits on our capacity for processing information. Psychological review, 63(2), 81.

  • $\begingroup$ As I am working cognitive modeling of problem solving, this answer is meaningful to me. From your answer, I presume that S1/S2 theory would be in the capacity-limit-as-weakness camp, right ? But it hardly dispels my interrogation concerning storage capacity limitations. The task "remember these items" seems computationally much more simple than real problem solving tasks as no extra processes are engaged. $\endgroup$ – brumar Apr 16 '14 at 8:44
  • $\begingroup$ Sorry I submitted by error and I did not know that comments could not be edited 5 minutes after that. Sorry about that. This passage was lacking: if we could "store more", the S2 process would be more efficient, am I right ? Then why didn't we evolved towards more storage capacity ? $\endgroup$ – brumar Apr 16 '14 at 8:55
  • $\begingroup$ I can't really give a response backed up by the dual process literature, but my own thoughts are that it's not really right to think in terms of strengths and weaknesses here. S2/WM capacity is limited because flexible, domain general, serial computation isn't something which neural networks do easily, compared to S1 processes, and so are metabolically expensive. The frontal lobe, which seems to be the main 'seat' of these processes, would need to even bigger to allow greater capacity, and this would only happen through natural selection if it gave even more of a boost in evolutionary fitness $\endgroup$ – Eoin Apr 16 '14 at 10:08
  • 1
    $\begingroup$ Ran out of characters - Summary: Greater WM would need more energy, and so should only evolve if it increased evolutionary fitness more than the need for extra food decreased it (which it probably wouldn't, because S1 does most of the important work, for less energy). $\endgroup$ – Eoin Apr 16 '14 at 10:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.