This writeup is meant to be read after the one by
Plebius, which it comments on:
The turnover of new ideas in psychology is fast, and showing no signs of slowing down. Stephen Jay Gould mentions occasionally in his essays that he's appalled at the content of biology textbooks, which are/were often blatant plagiarisers copying inaccurate and outdated data from old textbooks, which copied it from even older textbooks - psychology is almost certainly no different. Pretty much everything in my first-year psych textbook (at least everything related to the areas I studied in greater detail later on) is wrong, oversimplistic, or outdated. Working memory is no exception.
Firstly and somewhat confusingly, Baddeley has two theories called "working memory." In his 1986 book Working Memory, he carefully makes the distinction between his general model of working memory, and his specific model of working memory. Baddeley's general model of working memory is essentially the concept that short-term memory is "a system for the temporary holding and manipulation of information during the performance of a range of cognitive tasks such as comprehension, learning, and reasoning" (Baddeley, 1986, p.34). In contrast, Baddeley's specific model of working memory is much like the description that Plebius noded above, involving phonological loops, central executives, and so forth.
Secondly, nowadays, Baddeley's specific model of working memory is simply outdated. The visuospatial sketchpad has superceded by the Loftus & Maclean (1999) model of visual memory (a mathematical model which is too complicated and boring to explain here, but which is also very good at explaining experimental results). The central executive has been attacked for being wishy-washy, a relic of the fact that Baddeley has simply tried to stretch his experimental findings too far.
Even the strongest part of the theory, the phonological loop, is simply unable to explain many recent experimental findings. For instance, it cannot explain how we can remember things in the right order (Burgess & Hitch, 1996). Additionally, it is unable to explain how information from long-term memory appears to be used in the phonological loop (e.g., according to Poirier & Saint-Aubin, 1995, we remember lists of words from a category better than lists of semantically dissimilar words). For these reasons (and others), the phonological loop has been superceded by connectionist models of short-term memory like those of Gupta & McWhinney (1997) and Burgess & Hitch (1999).
These days, it seems likely that, while there are definitely brain systems devoted to holding and manipulating information, it's just not as simple as Baddeley's specific model makes out (the brain? extraordinarily complicated? surely not!). For example, recent experimental evidence (e.g., Martin, Lesch, & Bartha, 1999) suggests that there may not be any difference between the phonological loop and speech perception and production processes, and so theories attempting to explain verbal short-term memory also have to explain speech perception and production.
Baddeley's specific model is a little like Newton's theory of physics; it's influential, clearly the work of a very clever man, and it explained stuff pretty well at the time it was drafted. However, experiments testing more recent theories have simply thrown up findings that the traditional model wouldn't have dreamt of trying to explain.
That's science for you.