The blooming buzzing confusion and filtering of visual working memory
Thursday, February 18, 2016
William James famously said that the world is “one great blooming, buzzing confusion” to an infant whose sensory apparatus is “assailed by eyes, ears, nose, skin, and entrails at once.”
As adults, we are still assailed by all of the above, but somehow we manage to deal with the complexity of the world. We use our attention to recognize threats in the environment, such as venomous cucumbers, and we use an attentional filter to prevent irrelevant information from entering our working memory.
We recently reported on this blog how attention is used to filter out information that is not relevant to the task at hand.
But how efficient is this attentional filter? One of the many demands placed on our attention is the need to switch rapidly between different settings: One moment we might be interested in attending to and remembering only furry things on four legs—for example, when hiking through the Serengeti—and the next moment we might be more interested in bottled cold liquids—once we safely reach the safari camp.
How do people handle situations in which the requirements to filter irrelevant information alternate rapidly and repeatedly?
A recent article in the Psychonomic Society’s journal Cognitive, Affective, and Behavioral Neuroscience addressed this question. Researchers Jost and Mayr presented participants with objects that had to be remembered for a test of visual working memory. On each trial, participants were cued to remember only certain objects while ignoring the others—for example, on one trial participants might be cued to remember only red objects, while ignoring blue objects. On the next trial, blue might be relevant for memorization and red had to be ignored, and so on.
Several research questions were of interest. The first one related to the efficiency of filtering: If our attentional filter were perfect, then the contents of our memory should not differ between a trial on which the display contained two red and two blue objects, with only blue objects being relevant, and a trial on which only two blue objects were presented and hence no filtering was required. The figure below shows the stimuli used by Jost and Mayr:
At first glance the set size labels seem inappropriate: why are there 4 blue rectangles in the “set size 2” stimulus, and 8 in the “set size 4” display? The answer involves the particular neurophysiological measure used by Jost and Mayr, known as the contralateral delay activity
(CDA). The CDA is a sustained negative wave measured over the posterior cortex by electrodes affixed to the skull. The CDA is evoked in response to a memory load, and is calculated as the amplitude difference between contralateral and ipsilateral activity, thereby isolating the lateralized effects of visual WM from nonspecific bilateral activity. To measure the CDA therefore requires that participants are cued to which part of a display that is presented bilaterally has to be remembered—hence when there are 4 rectangles, the set size is actually 2 because only one half of the display (either to the left or the right of the fixation cross) is to be remembered on each trial.
The events on a single trial in the study by Jost and Mayr are shown below. The fact that the arrow is blue tells the participant that only blue objects are relevant, and the fact that the arrow is pointing left indicates that only the left part of the display is to be memorized.
The figure also shows how memory was tested: after a brief delay, a single probe was presented in the relevant hemifield. The orientation of the probe was either identical to the orientation of the object presented at that location or it had changed. Participants had to indicate whether the orientation was the same or different. In the figure above, the orientation of the probed rectangle changed between study and test.
Let’s return to the neurophysiological measure: One intriguing aspect of the CDA is that its amplitude increases with the number of representations being held in visual WM, thereby providing a measure of the contents of visual WM without requiring an overt response on the part of the subject.
Put another way, the CDA provides an independent assay of filtering efficiency: If filtering is perfect, then the “set size 2” CDA should be identical to that obtained with the “set size 2 + 2 distractors” display. Conversely, if filtering failed completely, the CDA with the latter display should be identical to that obtained with “set size 4”.
The results showed that trial-by-trial changes in the relevant color impaired filtering efficiency: Memory performance was worse with the “set size 2 + 2 distractors” display than the “set size 2” display. The behavioral effect was accompanied by a large difference in CDA magnitude, as shown in the Figure below, which plots evoked potential as a function of time from onset of the memory display:
The figure shows that the CDA for the “set size 2 + 2 distractors” condition (red) overlapped nearly completely with the “set size 4” (blue) curve, but differed considerably from the “set size 2” CDA (black). In other words, people’s memory held as many representations in a condition in which half of them should have been ignored than in another condition in which all items had to be remembered.
Trial-by-trial switches in the attentional filter are therefore not terribly effective: if we never know what to pay attention to until a fraction of a second before we need to filter the information entering our memories, that filtering does not work terribly well.
Does this mean we can never selectively pay attention to memorially relevant information?
Another part of the experiment by Jost and Mayr suggests otherwise.
That other part of the experiment answered a further research question, namely how well the attentional filter can operate under ideal conditions. In that part of the experiment, the relevant color remained the same for a long sequence of trials so that participants could comfortably anticipate which objects had to be filtered out on each trial. Under those circumstances, the CDA pattern looked quite different, as shown in the next figure:
This time round the (red) distractor curve no longer overlapped with the larger set size (blue) and was much closer to the “set size 2” line (black). The CDA therefore tells us that the number of encoded representations was much smaller when half the objects had to be filtered out than when all 4 of them had to be remembered.
The behavioral data confirmed this pattern: memory accuracy was identical between the “set size 2” and “set size 2 + 2 distractors” conditions, also pointing to perfect filtering overall.
When the data were analyzed across individual participants some interesting differences emerged: People with greater working memory capacity exhibited greater filtering efficiency (as reveled by the CDA difference between the distractor condition and the set size 4 condition) than people whose working memory capacity was lower. This effect of working memory capacity was absent, however, in the earlier part of the experiment when the identity of distractors changed on every trial.
Jost and Mayr conclude that two distinct factors determine overall filtering efficiency. The first is the goal-related implementation of an attentional filter, which is related to working memory capacity and which is revealed when the to-be-filtered information remains unchanged for long sequences of trials. The second factor is the fact that even efficient filter settings can be penetrated by information that lingers from previous trials but is now irrelevant. Additional working memory capacity offers no protection against this lingering irrelevant information.
Reference for the article discussed in this post:
Jost, K., & Mayr, U. (2015). Switching between filter settings reduces the efficient utilization of visual working memory. Cognitive, Affective, & Behavioral Neuroscience. DOI: 10.3758/s13415-015-0380-5.