Hi Yingxu,

Interesting question. After Miller's estimation researchers indicated that, actually, Miller overestimated the short-term memory capacity. They estimated it in 4 chunks (e.g., Cowan, 2001). On the other hand, research with experts has shown that they can remember much more information over the short term. The current most important theory in expertise -the template theory (Gobet and Simon, 1996) proposes that the short-term memory capacity is still limited to four chunks, but that experts activate chunks stored in long-term memory while they perform a short-term memory task. The point you are raising is interesting: Is it OK to suggest that all the knowledge we use when we perform a short term memory task does not occupy the short term memory capacity? This is an interesting and important discussion. The way it is resolved may or may not change standard models of memory.

Hi Yingxu,

Interesting question. After Miller's estimation researchers indicated that, actually, Miller overestimated the short-term memory capacity. They estimated it in 4 chunks (e.g., Cowan, 2001). On the other hand, research with experts has shown that they can remember much more information over the short term. The current most important theory in expertise -the template theory (Gobet and Simon, 1996) proposes that the short-term memory capacity is still limited to four chunks, but that experts activate chunks stored in long-term memory while they perform a short-term memory task. The point you are raising is interesting: Is it OK to suggest that all the knowledge we use when we perform a short term memory task does not occupy the short term memory capacity? This is an interesting and important discussion. The way it is resolved may or may not change standard models of memory.

I am in agreement that the estimate of 4 to 7 is accurate, depending somewhat on whether one is referring to verbal versus visual processing. Chunking is the way to increase total volume, with 4 chunks being right on target. In the attached article, I suggested that the number of interconnecting columns is the neurophysiolgical explanation for limited capacity.

Article Of bits and logic: Cortical columns in learning and memory. ...

Perhaps we to define our terms. Perhaps there are different types of short-term memory. Perhaps there isn't a definite demarcation between short-term memory and other forms of memory.

What has brought this to light is an automobile accident I had once. After the accident, there was a blank period in my memory between the moment the car began to fly off the pavement and the moment when I found myself hanging upside down by the seat belt. (I suffered a slight concussion, but I don't think this had anything to do with this memory lapse.) My interpretation of this memory loss is that certain stimuli are stored for a short time (~ seconds) in memory to be processed. If they are not processed, they are discarded. This is different from trying to remember digits because digits have an interpretation and, therefore can be associated with prior knowledge. Stimuli may be, for example, a scene your eyes "photograph" without an interpretation, when you think you are about to die. Quite a different experiences.

A careful discrimination among STM (working memory), LTM, sensory buffer memory (SBM), and action buffer memory (ABM) is the key for clarifying Miller’s observation.

Human memory system is a framework as follows:

SBM --> STM --> ABM

.................^

.................|

.................v

...............LTM

Miller’s estimation was fit well to the capacity of SBM rather than STM, because there are more types of contents in STM such as episodes, procedures, images, texts, emotions, and experiences that are not digits. Nobody may limit these STM contents within 7+/-2 digits.

There are so many nontrivial empirical and theoretical exceptions against Miller’s assertion. For examples, more ordinary people do remember their account numbers, passwords, and crucial data much longer than 7 digits. People in some countries are able to remember names up to 50 letters. Kits can easily recall an essay with more than a few hundred words within a few hours. In all these cases, STM being recalled within less than 24 hours proves that STM's capacity is much greater than 7+/-2 bits.

For further discussions please see related questions on RG: “Where are various memories retained and located in the brain? Are cortical lobes a part of the central intelligent processor or merely memories?” and “Can memory retained in LTM be intentionally forgotten? If yes, what’s the neurological mechanism? If not, where does it happen?”

Wang,

The magic number refers to chunks, which can be associated with an infitude of data readly accessible in long-term memory.

I guess it follows the rationale behind the computer architecture: limited super fast memory (cache memory in the processors), the medium term memory is slightly larger and slightly slower (RAM memory) and than there is the virtually unlimited long-term memory (hard disk, cloud storage).

Training makes the expert hold in 1 'slot' (out of the 7 'slots in the magic number theory), what the novice will hold using many slots...

I think the theory is only relevant in stress conditions, to test reaction time and error proness of stressful activities such as airplaine traffic control. When the activity is not so time-stressful, the speed difference and the limited capacity kind of don't matter much anymore, the equation is much more complex as the analysis timeframe enlarges, along with the complexity of the data. Kind of how the forces studied in physics for the micro world doen't matter much on the macro world.

Everything I've seen on the subject agrees with the majority of respondents: Miller, if anything, overestimated the capacity of STM (perhaps because he loved the "magic number 7" title). I've usually seen numbers like 5.5 instead. And people tend to speak of working memory rather than STM nowadays. Of course, there are ways to boost the absolute amount of information. As others have noted, "chunking" information into units that contain a number of "bits" is one way. Linking the to-be-remembered material with existing LTM stores is anther. (This is the reason why those old mnemonic devices such as the Method of Loci actually work.) And there are at least two (perhaps more) separate working memory systems: the articulatory loop and the visuospatial sketchpad. (I would add an object vision system to that list, and perhaps a kinesthetic loop, but the latter is quite speculative.) Each has its own limited capacity (maybe, on average, 4 chunks), but more than one can be activated at a time. Still, when one is drawing on cognitive resources, less is available for the others, so the total capacity is far less than the sum of the capacities of the individual components.

Although Miller was making a serious argument, in fact the "Magic" in the title and his conclusion was meant to be ironic. The article compares limits in short-term memory to limits on the number of dimensions that people can differentiate in perceptual tasks, which he argued had approximately similar magnitude. But he ends: "And finally what about the magical number seven? What about the seven wonders of the world, the seven seas, the seven deadly sins..Perhaps there is something deep and profound behind all these sevens...For the present I propose to withhold judgment. But I suspect that it is only a pernicious, Pythagorean coincidence." The original article is well worth re-reading, and note the allusion to Senator Joseph McCarthy in the opening paragraph.

Yes, Andrea, the article is quite interesting and really wasn't "about" short-term memory as such. I think Miller's main objective was to build a case for the proposition that information can be quantified in a reasonably objective way, implying that the study of cognitive operations could be a legitimate scientific enterprise. This sounds obvious to most psychologists today, but when he wrote, we were still in the heyday of the behavioral movement, and many (most?) experimental psychologists in America assumed that mental events simply could not be studied empirically. Miller's illustrative examples (short-term memory, subitizing, perceptual discriminations) were just that - illustrations.

In view of this answer, would it be possible to broaden STM in dyslexic people?

Joachim, thanks for links to literature. Best, John