To enhance any system, one first needs to identify its capacity-limiting factor(s). Human cognition is a highly complex and multiply constrained system, consisting of both independent and interdependent capacity-limitations. These “bottlenecks” in cognition are reviewed below as a coherent framework for understanding the plethora of cognitive training paradigms which are currently associated with enhancements of working memory, executive function and fluid intelligence (1,2, 3, 4, 5, 6, 7, 8, 9,
10, c.f. 11, 12, 13).
By far, the most common complaint about limitations in cognition is something along the lines of “I have a bad memory.” As described below, this is probably because what we call “memory” is the emergent result of multiple constraints operating in parallel. To improve memory, we need to understand what these parallel operations are, and how they may constrain the emergence of memory.
At the coarsest level, we can distinguish long term from short-term memory. Evidence for this distinction can be observed pharmacologically (with midazolam, among other drugs), clinically (Patient H.M., for example), neurally (long-term memory seems to require the hippocampus, whereas short-term memory may not) and behaviorally (long-term memory has an unlimited capacity, but short term memory is limited to 7 items [or 4, or 1 – the jury is still out]). Thus, this basic distinction is supported by a variety of evidence as well as common wisdom (but see this).
Although long-term memory is basically unlimited in capacity (according to most theories, anyway), it is temporally limited – after a certain amount of time has elapsed, or a certain number of contextual changes encountered, stored information may become inaccessible. There are some reasons to believe information stored in long-term memory is never truly lost, but rather becomes inaccessibly difficult to retrieve into short-term memory via attention (what Unsworth & Engle have called “cue specification“). Thus, at least for the moment, let’s suppose that long-term memory has no inherent limitations, but rather inherits its limitations from those in short-term memory.
Some theorists have persuasively reconceptualized the human memory architecture such that short-term memory is simply the activated portion of long-term memory. I think they are probably right, but for our purposes the distinction is mostly irrelevant: all long-term memory must still “pass through” short-term memory and therefore is constrained by it. Accordingly, these reconceptualizations of memory architecture tend to focus on the capacity limitations of short- rather than long-term memory.
Within short-term forms of memory, we can further distinguish between “working” and “iconic/echoic” memory. Iconic/echoic memory can be thought of as the “trace” of stimulation that reverberates for a short time in sensory cortex after the stimulus is no longer present in the environment. A subset of this sensory information can be placed into working memory and thereby preserved across time. Thus, while iconic/echoic memory is capacity-unlimited, it is temporally-limited; working memory, in contrast, is capacity-limited but (theoretically) temporally-unlimited.
So, roughly speaking, both forms of short-term memory are bottlenecked. Were it possible to lengthen iconic memory – the time that environmental stimulation reverberates in sensory cortex – the disadvantages of such “temporal blurring” might supercede its advantages. Maybe precisely this kind of temporal expansion in iconic memory is the underlying cause of eidetic or so-called “photographic memory” (note, however, that the existence of true photographic memory is highly controversial: only one person has ever been reported to have passed the most rigorous test for photographic memory, and has disappeared from the scientific literature since that report. This coincidence may be indicative of fraud.)
In contrast, the capacity limitations of working memory may be a “safer” target for enhancement because working memory is under conscious control and is unlikely to maladaptively interfere with more basic cognitive functions (c.f.).
The first limiting component of working memory is the selection of important information currently represented in iconic memory, or in sensory cortex, possibly via a ventral attentional network which monitors for important stimuli. Fascinating work from Lavie and colleagues has shown that there are dissociable capacity effects at this “selection from perception” level; more recent work suggests that these perceptual load effects may be related to an object-based limit of approximately 4 items (or features) in the intraparietal sulcus.
Information that has been selected and updated can then be preserved for further processing through a process known as “binding” (Cowan et al) or “consolidation into working memory” (Vogel et al) in several ways:
a) via sustained attention (which may itself require what Unsworth & Engle call a “focus switch“).
b) via rapid storage within an “episodic” or long-term memory buffer (requiring only transient attention, as suggested by “dual mechanisms of control” theory), or
c) via both mechanisms simultaneously (as suggested by the fact that in the absence of long-term memory support, working memory span is capacity limited to 1 item).
The format of this preserved information appears to be a series of slots each of which has a discrete resolution, which may be related to the posterior parietal cortices. Some fascinating work suggests that the resolution and slot capacity may represent characteristics of the inferior and superior parietal cortex, respectively.
(Orthodoxy in cognitive psychology holds that two “slave systems” [the “inner scribe” and the “phonological loop”] can be used to rehearse this information, although there are a number of problems with this idea and these two systems have not been well supported by more recent work in cognitive neuroscience.)
The components of working memory listed in bold in the preceding paragraphs are all potential targets for cognitive training and enhancement; they are:
- 1) selection (filtering of perceptual information by posterior parietal cortex)
- 2) updating (basal ganglia-mediated entry of sensory information to working memory)
- 3) focus switching (switching attention among items in working memory)
- 4) sustained or transient attention (aka “proactive or reactive control”)
- 5) binding (parietally or hippocampally mediated; see this discussion)
- 5a) storage in a episodic or long-term memory buffer (hippocampally-mediated)
- 5b) storage in short-term memory buffer (parietally-mediated)
- 6) # of working memory “slots” (Zhang & Luck’s term)
- 7) resolution of working memory “slots” (Zhang & Luck’s term)
Now, which of these bottlenecks can be alleviated?
As reviewed over the past few weeks, updating can be enhanced with n-back training, but this may not widely transfer to other tasks; “focused attention” may be made more efficient with mindfulness techniques (as assessed via the “orienting” and “conflict monitoring” components of the ANT task and via the attentional blink) but might not be parallelized or expanded via practice (as determined via constant “focus switch costs” as a function of practice on the n-back). Somewhat counterintuitively, focus switch costs are unrelated to working memory capacity and fluid intelligence; thus the immutability of focus switch costs are not a lost opportunity for executive control and intelligence enhancement.
Interestingly, although the more reactive or retrospective mechanism of working memory is the newest to be proposed, it is probably the mechanism behind the majority of the oldest memory training techniques, which encouraged strategies for “chunking”. (Such strategies thus target the encoding of information; might future work reveal undiscovered techniques for enhancing retrieval?)
As far as I know, there are no reports of improvements in “selection efficiency” as operationalized in the Luck & Vogel paradigm, nor whether “selection efficiency” changes as a function of more generalized working memory training. It is also unclear whether selection efficiency is related to “updating” (and the basal ganglia, as suggested by McNab & Klingberg‘s work) or to the “alerting” / VAN system, but both have been shown to be plastic with training. In all likelihood, “updating” and “alerting” work in conjunction, with “altering” being accomplished by a ventral attentional network which then triggers a basal ganglia “updating” process (as supported by some analyses of the p3 response).
In addition, no work has focused on enhancing the resolution of working memory’s discretely precise “slots”, nor on whether the number of those slots appear to increase with practice. This is probably due to the fact that Zhang & Luck’s mixture-model method is just too new to have been used in these paradigms.
Different areas of the brain are thought to underlie the two forms of information preservation in working memory discussed above (continuous “focused attention” in the service of active maintenance, and “rapid storage” in the service of a more reactive/retrospective form of working memory). While the same basic cortico-striatal architecture probably subserves both mechanisms, parietal cortex may be particularly important for the form of working memory that involves active maintenance and continuous, focused attention, whereas the hippocampus may be more involved in the reactive or retrospective form of working memory. In fact, computational models suggest the two regions may interact or coordinate their activity. (The reactive/retrospective form of working memory might also [or alternatively] involve a form of cortical “fast-weights,” similar to those underlying iconic memory. By this view, reactive/retrospective working memory could involve the “reconstruction” of a memory from decaying memory traces exclusive of hippocampal pattern completion.)
Finally, attention training (whether of sustained or transient attention) has been the focus of some of the most successful and long-standing cognitive training paradigms: those carried out by Posner and colleagues. One can speculate that more recently-discovered cognitive training techniques also work directly on attention itself, perhaps increasing its efficiency by increasing its strength or coherence (see, for example, this previous post on the tightening of neuronal temporal tuning as a result of practice, and this work suggesting that uniformity in reaction time distributions subsumes and surpasses the predictive variance in IQ captured by elementary cognitive tasks.)
In summary, human memory is a heterogenous entity with multiple constraints operating both in parallel and in series. This does not, however, preclude a coherent framework for understanding the multiple behavioral training methods which are being discovered to enhance cognition.