Remembering To Remember: Prospective Memory

In the middle of the work day, you realize you'll need to stop at a store on your way home from work. Your ability to actually do so, hours later, relies on what some psychologists call "prospective memory." Although prospective memory is clearly important for human intelligence, very little is known about how it works.

Clearly there are at least two kinds of prospective memory. In the example above, you may tell yourself "stop at the store" again and again until you pull into the store's parking lot - this is known as a vigilance or monitoring strategy. Or the store may simply catch your eye while driving, causing you to remember you need to stop there - known as a "spontaneous retrieval" strategy.

In a 2005 JEP:G paper, Einstein et al. review the evidence supporting these two theories of prospective memory. According to hard-line monitoring accounts of prospective memory (PM), retrieval of intentions can only happen through monitoring for target events - the store on the side of the road. Accordingly, dividing attention during PM tasks impairs performance, presumably since attention is required for the monitoring process.

But this might occur simply because divided attention does not allow the processing of the cue sufficient for spontaneous retrieval, as Einstein et al note. The authors also suggest that dividing attention could simply reduce the efficacy of selecting the intention from memory, rather than interfering with monitoring per se. Indeed, previous work by these authors suggested that subjects merely had the intention "pop into" their mind at the appropriate time, rather than continuously maintaining the intention and monitoring for the occurrence of the cue. This would be an instance of spontaneous retrieval.

In contrast, the "multiprocess view" of PM suggests that both monitoring and spontaneous retrieval strategies are used by human subjects, but that there is a general bias towards using spontaneous retrieval (mostly because it's easier!). This view suggests that a subject is more likely to use a monitoring strategy when focal processing of the cue is unlikely, and when the importance of correctly implementing the planned intention is emphasized.

To determine how much monitoring is actually involved in PM, Einstein et al. performed several experiments in which they manipulated the tendency of individual subjects to use monitoring or spontaneous retrieval strategies.

In a first experiment, subjects had to perform a category membership judgment task concurrent with a PM task. In the PM task, they had to give a different response either to a particular word (for half the subjects) or to a particular syllable (for the other half of the subjects). Since detection of each word was already required by the category judgment task, this condition was predicted to induce spontaneous retrieval strategies for PM (and was thus called the "focal condition"). In contrast, since the explicit detection of syllables was unrelated to the category membership task (and previous research shows that syllables are not usually processed as independent items), this condition was predicted to induce monitoring strategies for PM (and was thus called the "nonfocal condition") Half of each of these groups was told the category-membership task was most important, while the other half was told that the word/syllable target detection task was the most important.

Critically, reaction time and performance on the category membership task was not decreased by the simultaneous performance of the focal, moderate-emphasis PM task. On the other hand, all other combinations of the PM task were associated with increased reaction time on the category membership task. Einstein et al. concluded that monitoring strategies are made more likely either by increasing the importance of the PM task, by decreasing the amount of attention directed typically towards the cue in the PM task, or by both in conjunction.

Einstein et al. also examined the performance of subjects across the blocks. Only in those conditions where monitoring was made more likely did subjects show worse PM as the experiment continued, suggesting again that monitoring demands are an attentionally demanding - but not always requisite - strategy for PM.

In their third experiment, Einstein et al. found similar effects with tasks that are even harder than the category membership task. In this case, performance and reaction times on sentence completion judgments (e.g., "The warrior's armor makes him ________ to any blows that he may undergo in battle. IMPERVIOUS?") were unaffected by a concurrent prospective memory task in which subjects had to respond to one particular target word. This suggests that monitoring is not needed for all PM tasks, since an active monitoring process would probably slow the reaction times on this difficult sentence completion task. However, when the PM task required subjects had to monitor for any of 6 possible words, Einstein et al. did observe significant slowing of reaction time.

In a fourth experiment, Einstein et al. discovered that these tasks can reliably index individual differences in PM strategies. Less than 30% of 104 tested subjects showed significantly larger reaction times for sentence completion judgments made concurrent with the PM task relative to those made in the absence of a concurrent task. Thus, Einstein et al. conclude that at least half of the subjects were not engaging monitoring processes to complete the PM task.

In a fifth experiment, Einstein et al. showed that spontaneous retrieval happens even after the PM task is over. They demonstrated this remarkable effect by abruptly having subjects switch from a dual task situation where PM was being tested (as in experiments 1-4) to a task where they knew the PM task was irrelevant. In this case, subjects were slower to respond to items that were related to the targets in the PM task, even though they were not monitoring for them. Since continued monitoring for the targets was unnecessary in this task, Einstein et al. conclude that a relatively automatic process of spontaneous retrieval may have caused reaction time slowing on target trials.

These experiments show that subjects can flexibly use at least two strategies in the service of prospective memory: they can actively maintain their intention (monitoring strategy) or they can often do just as well by not thinking about the prospective memory task at all, and passively waiting for intention-related items to occur. This work suggests that the monitoring strategy is attentionally demanding, and that it may decrease in efficacy with time. On the other hand, monitoring strategies may be used when the prospective task is very important, or when the number of target items is large (or otherwise underdetermined).

Mechanistically, both spontaneous retrieval and monitoring strategies may rely on similar neural substrates. In fact, they may actually be two ends of a prospective memory continuum. Monitoring strategies probably require sustained activation of the target items, perhaps in prefrontal cortex. Retrieval of the PM action will be more likely when each PM target item receives more activation, and this in turn will be more likely when prefrontal cortex is maintaining those items. A similar perspective is described in this paper.

According to this view, the relatively underdeveloped prefrontal cortices of young children may not be sufficient to allow for strong bindings of target items (or contexts) with novel actions (probably accomplished by the hippocampus). Prospective memory in the youngest children will therefore likely depend on more external factors, such as motivation (indeed, even 2 year olds will show precocious prospective memory if the task is to stop for ice cream), and the amount of previous experience with this particular item-action binding (increased previous experience makes spontaneous retrieval more likely, even in the absence of active maintenance, because of stronger latent representations).

As context processing and active maintenance improve with age, spontaneous retrieval may be possible for relatively novel prospective memory tasks, including novel item-action bindings. Finally, once context processing and active maintenance abilities are robust, children may be capable of actively monitoring for the presence of targets in novel prospective memory tasks.