Bachelor Project Brain & Cognition
Can Visual Information be maintained Unconsciously in
Working Memory?
Max Philipsen Student #: 10545212 Universiteit van Amsterdam
2 juni 2017 Timo Stein, Mentor
Word count: 5.478
Abstract
From a classical point of view, working memory (WM) and conscious awareness are tightly related. The current view states that WM operates only on conscious input and that the contents of WM reflect the contents of consciousness.
However, recent studies found evidence suggesting WM operating on
unconsciously presented information. The present study used a delayed cue-‐ target orientation discrimination task in attempt to replicate this effect. Contradicting results show that participants could not maintain a subliminal memory cue in WM and perform above chance in an explicit discrimination task. However, dissociation was found between actual WM content and what we experience as the WM content, namely that an induced visual WM load interfered with visibility (subjective measure) only, not with accuracy (objective measure). These contradictory results questioned the effectiveness of the visual WM load; future studies will have to investigate the effect of different kinds of WM loads (verbal, auditory) and possible other memory mechanism involved in these tasks.
Introduction
How do we memorize a phone number? We can do this by rehearsing it ‘in our mind’. While we do this, the phone number becomes a conscious mental object. But where do we keep this object? Some will say we keep this object in our short-‐term memory and – a more recent point of view – some will refer to our working memory. Recent studies argued that working memory could operate on unconscious information. This would unlikely suggest that we could ‘unconsciously’ memorize a phone number. Before we continue on this topic, some memory theories will be discussed.
By the end of the 1960s, Atkinson & Shiffrin (1968) introduced their influential memory model suggesting three separate, key systems: a sensory register, a short-‐term store (STS) and a long-‐term store (LTS). Environmental information is thought to flow through the sensory register into a limited capacity short-‐term store, also known as working or short-‐term memory. Here, the information can be held, for instance by rehearsal. In this model, rehearsing allows the information to be stored more permanently in the long-‐term store, in other words: to be encoded. The STS or WM also receives and holds input from the LTS; this is called retrieval. This model implies that the contents of STS or WM reflect the contents of consciousness.
Whereas the STS is responsibly for the short-‐term storage of information, an important feature of working memory is the possibility of manipulating stored information. Baddeley and Hitch (1974) proposed a three-‐component WM model, including the ‘central executive’, supported by two subsidiary slave systems, the ‘phonological loop’ and the ‘visuospatial sketchpad’. The
for holding verbal and acoustic information, whereas the sketchpad is involved in the storage and manipulation of visuospatial information. This model
proposed the central executive as a supervisory system as it coordinates the flow of information from and to its slave systems, intended for controlling and
regulating cognitive processes. Later, a fourth component of the working memory model was added: the ‘episodic buffer’, which stores information and thus provides a temporary interface between LTM and both slave systems (Baddeley, 2000). This buffer binds information from multiple sources into coherent episodes. Baddeley suggested that consciousness plays an important role in all WM input and output. Besides, Baddeley assumes that WM content reflects our conscious experience.
Another theory supporting the idea that WM content reflects our
conscious experience is Bernard Baars’s ‘’global workspace’’ theory. This theory suggests that we posses a ‘’mental router’’, and that what we experience as consciousness is the global sharing of information (Dehaene, 2014). The brain has specialized, local processors that each perform a specific type of operation; the ‘’global workspace’’ allows them to flexibly share information. In this workspace a coherent representation of all the information together is formed and maintained. But, most importantly, whenever information enters the workspace, it becomes conscious.
But what does it exactly mean when something becomes conscious? Every second, a massive amount of sensory stimuli reaches our senses, but our
conscious mind manages to gain access to only a small part of them; this is called
information enters consciousness, it becomes a conscious mental object that we can ‘’keep in mind’’. Bringing the information to the forefront of our thinking enables us to verbally report it. What happens to all the information that does not reach consciousness? Even unattended and unconscious information can be processed which is called subliminal (‘’below threshold’’) processing, and
therefore our human behaviour is prone to subliminal stimuli that are not consciously experienced. For example, when we briefly flash a masked, subliminal word followed by a visual target word, this will speed up the processing of the target word if the unconscious presented word is the same.
So far, all discussed WM theories share the same thought: WM content is equivalent to conscious experience. This also means that all components of WM are reportable. Not every theory shares this idea, for instance Cowan’s working memory model (1995), which was later extended to the three-‐embedded-‐
components model by Oberauer (2002), assumes three components are included in working memory: the activated part of long-‐term memory (LTM), the region of
direct access, and a single-‐item focus of attention. LTM serves as a source for
accessible information, while the region of direct access contains representations that more or less corresponds to the focus of attention and has a limited capacity of about four items or chunks. The focus of attention will select one item or chunk and uses this for the next cognitive operation. This model implies that
information can be maintained in WM and remain unconscious until it’s
attended. When information is attend to, it becomes accessibly and reportable. Concluding, from a classical point of view, WM and conscious awareness are tightly related. The current view states that WM operates only on conscious
input and that the contents of WM reflect the contents of consciousness, which means that awareness is essential for the operation of WM. However, recent studies found evidence suggesting working memory operating on unconsciously presented information and thus contradicting the current psychological theories. To be more specific, this new radical view assumes that even if information in WM is attended, it can remain unconscious.
Under some circumstances WM and consciousness can be fully
dissociated. Bona and colleagues (2013) showed participants a memory cue (a grating) of which they were asked to hold the orientation in memory. In
approximately half of the trials, a distracter grating was shown before the cue target orientation discrimination task. The distractor’s orientation was either identical to that of the memory cue, or its orientation differed. Because the distractors were masked and presented briefly, not all distractors were perceived consciously. After this delay period, participants were presented a memory test probe and were asked whether it was tilted to the left or right relative to the memory cue. This was the objective measurement; visual short-‐ term memory (VSTM) accuracy. Participants had to rate the vividness of the original memory cue; this was the subjective measurement. In the end, participants reported their awareness regarding the distractors. Bona and colleagues found that even though participants were unaware of the distractors, the distractors reduced the vividness of the memory cue (subjective awareness), but they did not impair discrimination accuracy (objective measure). The first-‐ mentioned findings suggest that the introspection of WM contents is susceptible by factors that do not interfere with WM accuracy and this suggests a
dissociation between actual WM content and what we experience as the WM content.
In summary, our conscious memory experience does not always reflect the actual memory content. Does this suggest that WM can operate on
nonconscious information? Soto and colleagues (2011) found evidence supporting this idea. Similar to Bona et al., they used a delayed cue-‐target orientation discrimination task. Observers were presented with a masked memory cue and were encouraged to attend and hold the cue in memory. After the delay period participants had to perform a memory test followed by an awareness rating (1-‐4 scale) regarding the memory cue. These awareness ratings were used to assess conscious experience; a rating of ‘1’ meaning that participants were unaware of the memory cue. The results of their first experiment showed that discrimination performance on unaware trials was above chance. This was interpreted as showing that observers could maintain a subliminal memory cue in WM and perform above chance in an explicit
discrimination task, in which a supraliminal orientation probe is compared (clockwise or counter-‐clockwise) with an unconsciously perceived orientation cue.
Thus, the dissociation between the performance in the memory test (objective measure) and subjective awareness of the memory cue (subjective measure) implies that information can be nonconsciously encoded in WM, to be later used in the memory test. Therefore, this contradicts the earlier statement regarding that WM content needs to be conscious.
But, the interpretation of these findings supporting the idea of
unconscious working memory has some potential problems. First, the response on the subjective awareness rating is prone to response biases (Schmidt, 2015). For instance, a conservative observer tends to respond ‘’stimulus absent’’ or ‘’not seen’’ more often, also known as a not-‐seen judgement (NSJ), compared to a more liberal approach (tendency to respond ‘’stimulus present’’). In Soto’s experiment (2011) this includes the ‘1’-‐rating on the awareness scale of the memory cue. Second, even true non-‐conscious perception does not necessarily mean non-‐ conscious WM. Stein and colleagues (2016) argue that a conscious ‘’guess’’ could be maintained in WM during the delay period, which is based on non-‐conscious perception of the memory cue. Then, this guess could be used in the objective memory test. So, this would suggest ‘’blindsight-‐like non-‐conscious perception’’, not non-‐conscious WM. Third, an additional methodological problem arises regarding the isolation of trials on which participants rated the stimulus awareness with ‘1’ on the 1-‐4 scale (NSJ’s); in order to perform an correct
analysis concerning the detection sensitivity – the sensitivity measure d’ in signal detection theory (SDT) – both frequencies of misses and correct rejections, and hits and false alarms are required to compute the actual d’ score. Soto and colleagues considered only frequencies of misses and correct rejections, which resulted in the computation of a so-‐called pseudo-‐d’. This new index is not invariant to bias, underestimates actual sensitivity and thus is an incorrect measurement of detection sensitivity for the memory cue (Stein et al., 2016).
Beside these misinterpretations other potential problems exist, for instance the use of subjective visibility ratings in general. Nieuwenhuis and Kleijn (2011) argue that different measures of consciousness determine the
outcome of an experiment. Sergent and Dehaene (2004) presented target words during an attentional blink and asked their participants to rate the subjective visibility of these words on a continuous scale. They found a bimodal perception rating suggesting a discontinuous transition between nonconscious and
conscious processing. When Niewenhuis and Kleijn replicated this study using an alternative measure of consciousness, namely post-‐decision wagering, this
bimodal perception rating disappeared. So, using a specific consciousness measure (subjective visibility) does not always generalize to another consciousness measure (wagering).
An experiment similar to Soto et al. (2011) was done by Trübutschek and colleagues (2016) including a spatial masking paradigm, trying to replicate Soto’s experiment (2011) using a different task. Participants were shown a masked target square on 1of 20 location (in a circle) and were asked to localize the target after a delay (2.5 – 4 s) and rate its visibility on a scale from 1 (not seen) to 4 (clearly seen). In the second experiment an additional task was integrated: 1 (low load) or 5 (high load) digits were shown at the beginning of each trial, followed by the same sequences as in experiment 1. At the end of these trials, participants localized the target, recalled the digits and rated the target visibility. This additional task was intended to create a memory load depleting ‘conscious’ WM and not affect non-‐conscious (WM) performance. Their first results support Soto’s finding and they conclude that the above chance
performance on unseen trials reflect maintenance of non-‐conscious information. Second, they concluded that even when ‘conscious’ WM was depleted by a concurrent memory load, unseen information could be maintained (because of no interference on ‘unconscious’ WM).
However, implications exist regarding these interpretations. First, they concluded that the conscious verbal WM load decreased the precision with which non-‐conscious spatial information was maintained. So, the significant difference in target accuracy between high and low memory WM load in unseen trials is contradicting the idea of WM having a conscious and unconscious part: while unseen stimuli are processed by the unconscious part of WM, a conscious dual task WM load (processed by the conscious part) should not interfere with precision of ‘unconscious’ WM. Thus, this suggests that all WM content is
processed by one conscious mechanism, suggesting all WM content is conscious. Second, they added an additional dual task in order to show that this would only be affective on seen trial (2-‐4 ratings) and not unseen trial (1 rating),
respectively conscious WM processing and unconscious WM processing.
However, the conscious verbal WM load did not affect the performance on seen trial at all, indicating that this additional verbal task was simply un-‐affective, which could explain why the above chance performance on unseen trials were not diminished in the first place. Perhaps, the non-‐interfering of the WM load is the result of the load being verbal instead of spatial. Perhaps, to interfere on the localization (spatial) task mentioned above, a spatial WM load is required. In conclusion, due to these methodological implications and previous discussed limitations regarding Soto’s experiment, replication is necessary. The present study is going to replicate Soto’s experiment (2011) using the delayed cue-‐target orientation discrimination task. In addition to this original
experimental design a visual dual task will be added, that is an additional discrimination task. Imitating Trübutschek’s approach, with the exception of Trübutschek’s load being verbal, this additional dual task will form a visual WM
load. This means that the following sequence will be used: an (1) the original orientation cue is shown, which is (2) masked, followed by a (3) delay. During the delay period an (4) additional orientation cue is presented. The delay period will end with the (5) memory test (left or right relative to the original orientation cue). After (6) subjective visibility of the original orientation cue is rated using the perceptual awareness scale (PAS), participants perform a (7) second
memory test regarding the additional orientation cue (the additional orientation cue has to be reproduced using the computer mouse). The additional orientation cue is showed for 200 ms, which means this cue is consciously perceived. For the subjective visibility the following ratings are possible: 1 did not see anything, 2 maybe saw something, 3 saw the stimulus but not its orientation and 4 saw the stimulus and its orientation.
This study has two possible outcomes that will be discussed. The first possible outcome (Soto’s approach) would support the idea that WM can operate on unconscious stimuli. This implies that WM content is made up of a conscious part, implying conscious WM content, and an unconscious part, implying
unconscious WM content. The conscious visual WM load, which is part of the additional dual task, is expected to affect only the conscious part of WM, meaning that the unconscious part stays untouched and thus performance on unseen trials will be above chance. Besides, affecting or in other words constraining of the conscious part as a result of the visual WM load will result in a decrease of performance on the memory test on conscious seen trials. So, dissociation is made between conscious and unconscious WM content (see Figure 1; Recent view). The second possible outcome (alternative approach) would support the idea that all WM content is conscious, assuming that the visual WM load will
affect performance on both conscious seen (visible) and non-‐conscious unseen (invisible) trials, the latter meaning non-‐conscious perception. So, independent of the subjective visibility rating, the visual WM load will reduce performance on the memory test (see Figure 1; Classical view). Figure 2 shows an overview of these possible expectations.
Figure 1. Visualizing WM approaches regarding either the overlap of or
dissociation between WM content and what we experience as the WM content. This figure shows on which parts of WM/consciousness the different memory cues load. In this example the original memory cue was not consciously perceived, so it’s labeled invisible. The additional memory cue is always
perceived consciously, so this cue always loads on the conscious part of WM. The latter does not always apply to the original memory cue; because this cue is masked and presented for only 17 ms, it is not always perceived consciously.
Figure 2. Possible outcomes regarding memory test performance (% correct) on
the cue-‐target orientation discrimination task. During single task trials only the original memory cue test is performed, while during dual task trials an
additional to-‐be-‐remembered memory cue is presented. Assuming horizontal 50 gridline represents chance line (left/right; 50%). So, the visual WM load induced during the dual task will affect performance on memory cue visible trials
assuming both approaches (Soto’s and alternative). But, on memory cue invisible trials Soto et al. argue that performance will be unaffected, while the alternative approach expects a decrease in performance. On the left side (green bars) we expect to replicate Soto’s effect (2011), namely above chance performance on the memory cue invisible trials and logically high performance on visible trials due to the absence of an interfering visual WM load (single task). Note. The performance on all the invisible trials is expected to be significantly above chance.
Method
Participants
In the experiment 17 subjects participated. Participants had to be between the age of 18 and 35 years old. Participants had to meet the screening criteria, which included having no eye disorders (corrections are allowed), not being colour blind and being able to see depth. Participants studying at
Universiteit van Amsterdam received two participation points as a reward.
Materials
On a computer, using Matlab Student Version 12, the trials were
presented. The arrow keys were used for indicating if the memory probe was tilted left or right compared to the memory cue. The number keys were used for rating the visibility of the memory cue. The computer mouse was used for reproducing the additional memory cue in the dual task trials. The software IBM
SPSS Statistics version 23 was used for analyzing the data.
Manipulation: Single vs. Dual Task
As mentioned earlier the experiments consist of two different tasks. De dual task trials consist of a grey circle with black gratings indicating the (original) orientation to be remembered. The mask forms a grey with black radial sinusoid. The additional orientation cue is the same as the original
memory cue, except that this grating has a greenish cover for a clear distinction of the original grating. The memory probe is identical to the memory cue, except that its orientation is tilted to the left or right relative to the memory cue. During
the dual task, the reproducing of the additional memory cue is done by a green line that has to be rotated by hand using the computer mouse. In the single task trials participants were instructed to ignore the additional memory cue. This cue is shown anyway in order to control for any unexpected influence regarding the additional green cue. For instance, to rule out any purely visual effect.
Interference could originate from the green ‘flash’ alone, and not from the WM load formed due to the additional task.
Measurement: Perceptual Awareness Scale
Response accuracy regarding the original memory cue and probe is measured (left/right). Furthermore, subjective visibility of the original memory cue is rated using the perceptual awareness scale (PAS), developed by Ramsøy and Overgaard in 2004. It is believed that the PAS is among the best suggestions to measure subjective experience (Sandberg & Overgaard, 2015). Participants are shown the following question: ‘’Awareness of memory cue?’’ (see Figure 2; PAS). The following answers are possible: 1) did not see anything, 2) maybe saw
something, 3) saw the stimulus but not its orientation, 4) saw the stimulus and its orientation. It is important to emphasize that all trials in which the awareness of
the memory cue is rated with 1, these trials are considered invisible, meaning that the memory cue was not consciously perceived. All trials rated higher than 1 are considered visible.
For the reproducing task of the additional cue the error is measured: 0 error indicating a precise replication of the additional cue with a maximum error of 90o degrees.
Procedure
Participants were invited personally to participate in the experiment. After acceptation, participants were summoned to the lab at Roeterseiland Campus. Next, participants received an information brochure serving as an introduction to the research topic and including instructions regarding the experiment. After signing the informed consent, they were told that the experiment consists of four blocks: two dual tasks and two single tasks. Each block consists of 144 trials, which on there turn takes approximately 9 seconds per trial. So, the experiment takes 90 minutes average in total. During the dual task participants had to do two tasks. First, comparing the original memory cue to the original memory probe. Second, during the delay a second, additional memory cue was shown which orientation had to be reproduced in the end of each trial. Participants were told that the memory cue is absent in 50% of the trials. The memory cue is oriented at 10, 40 or 70 degrees and the probe offset is tilted 10 degrees to left or right compared to this memory cue. For the single task, participants were instructed to ignore the additional memory cue and no reproducing had to be done at the end of each trial. An example sequence is shown in Figure 3.
Figure 3. Example of the dual task trial. During single task response 2 is absent
and participants were instructed to ignore the additional cue.
Results Participants
The individuals that participated in this experiment were Dutch students studying at the following institutes: Universiteit van Amsterdam, Hogeschool van Amsterdam, Nyenrode New Business School Amsterdam, Vrije Universiteit Amsterdam and Inter College Business School, with the exception of one
professional tennis player and one Dutch soldier. De group consisted of 15 men and two women (N = 17) with an average age of 21 years old (M = 21.71, SD = 1.36). In some analysis’s participants had to be removed because of their inability to perceive the memory cue and thus rated all trials with 1 (did not see
anything). ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! 500!ms! 17!ms! 100!ms! 800!ms! 200!ms! 1000!ms! 200!ms! Response! PAS! Response!2! Fixation! Memory!cue! Mask! Delay! Additional!cue! Delay! Probe!
Cue Visibility: Effect of Cue Presence and Task
A 2x2 repeated measures ANOVA was conducted for inspecting the effect of cue presence and task (independent variables) on visibility (dependent
variable). These factors had two levels; the cue presence factor including present vs. absent and the task factor including single vs. dual. The analysis showed that the main effect of cue presence was significant, F(1,16) = 4.65, p = .047, logically meaning that the visibility was higher in memory cue present trials compared to absent trials. Also, the main effect of the task was significant, F(1,16) = 25.01, p < .001, indicating that during the single task visibility was higher compared to the dual task. Finally, a significant interaction effect was found between task and cue presence, F(1,16) = 6.07, p = .025, showing that the effect of the task is only visible in the memory cue present trials. This is favourable, because we expected participants to rate visibility with 1 during memory cue absent trials,
independent of the task. In Figure 1 an overview is shown.
For the dual task trials mean visibility of memory cue was measured for memory cue present and absent trials. Using a paired sample t-‐test we found that mean visibility was higher in the memory cue present trials (M = 1.59, SD = .44) than absent trials (M = 1.13, SD = .18), t(16) = 4.91, p < .001. Logically, this was expected because memory cue absent trials should have been rated with 1, while during memory cue present trials higher ratings could be possible. The same was done for the single task trials and similar results were found; higher visibility (M = 1.71, SD = .53) was reported in memory cue present trials compared to
memory cue absent trials (M = 1.10, SD = .13), t(16) = 4.81, p < .001.
Finally, we conducted a paired sample t-‐test to compare mean visibility of memory cue present single task trials with memory cue present dual task trials.
Higher visibility of memory cue was reported during the single task trials (M = 1.71, SD = .53) compared to the dual task trials (M = 1.59, SD = .44), t(16) = -‐2.93,
p = .01. This suggests that the visual WM load, which is induced by the additional
memory cue in the dual task trials, is interfering with introspection of subjective visibility. In Table 1 and Figure 1 overviews are shown.
Table 1
Mean visibility, ranking from 1 to 4, on the single and dual task for memory cue present and absent trials.
Present Absent Single 1.71 (.53) 1.10 (.13) Dual 1.59 (.44) 1.13 (.18) Figure 1. Mean visibility on memory cue present and absent trials
WM Accuracy: Effect of Task on Cue Present Trials
We conducted a paired sample t-‐test to compare, during memory cue present trials, mean WM accuracy in dual vs. single trials. No significant different was found between single task trials (M = .67, SD = .15) and dual task trials (M = .64, SD = .13), t(16) = 1.57, p = .14. This suggests that the visual WM load, as a result of the additional task, does not interfere with WM accuracy on the
memory cue present trials during the dual task. However, this WM load did have effect on visibility; this dissociation was not expected. Furthermore, both dual task, t(16) = 4.63, p < .001, and single task, t(16) = 4.63, p < .001, did differ significantly from 50% chance. An overview is shown in Table 2.
Finally, the mean error on the additional memory cue reproducing task was measured (M = 19.46, SD = 7.01). The error includes the difference between the to-‐be-‐reproduced green line and the actual response of the participant in degrees. This indicates that participants performed fairly well on this task, given that the maximal possible error is 90o degrees. This also implies that the
additional dual task induced a visual load that depleted on the resources of WM.
Table 2
Mean WM accuracy on single and dual task on memory cue present trials, independent of visibility.
Present
Single 0.67 (.15)
Dual 0.64 (.13)
WM Accuracy: Effect of Task and Cue Visibility on Cue Present Trials A second 2x2 repeated measures analysis of variance was conducted for inspecting the effect of the task and visibility (independent variables) on WM accuracy (dependent variable). Both factors had two levels; the task factor including dual vs. single and the visibility factor including invisible trials (rating 1) vs. visible trials (rating 2-‐4). The analysis showed that the main effect of visibility was significant, F(1,14) = 31.30, p < .001, as we expected. No interaction effect was found and the analysis showed no significant main effect of the task. The latter was not expected; a larger decrease in WM accuracy was expected during the dual task due to the visual WM load. An overview is shown in Figure 2.
For memory cue present trials WM accuracy was measured depending on the subjective visibility rating. During the dual task, when visibility was rated higher than 1 – meaning that the participants were aware of the memory cue – WM accuracy was significantly higher (M = .76, SD = .14) compared to trials where the participants were unaware (rating 1) of the cue (M = .52, SD = .08),
t(14) = -‐6.19, p < .001. During the single task similar results were found; higher
WM accuracy was measured when the participants were aware (rating 2-‐4) of the cue (M = .73, SD = .19) compared to when the participants were unaware (rating 1) of the cue (M = .48, SD = .11).
Furthermore, during the memory cue present dual task trials, when visibility was rated with 1, WM accuracy did not significantly differ from 50% chance, t(16) = .94, p = .36. The same applies to memory present single task trials, t(16) = .92, p = .37. Finally, no significant different in WM accuracy was found between dual (M = .52, SD = .08) and single (M = .48, SD = .11) task trials
when the memory cue was present and participants were unaware of the memory cue, t(16) = 1.70, p = .11. An overview is shown in Table 3.
Table 3
Mean WM accuracy on single and dual task for memory cue present trials only (!), depending on visibility.
Invisible (1) Visible (2-‐4) Single 0.48 (.11) 0.73 (.19) Dual 0.52 (.08) 0.76 (.14)
Figure 2. Mean WM accuracy on single and dual task both for invisible and visible
trials. Invisible meaning the participant was unaware of the memory cue (rating 1); visible meaning the participant was aware of the memory cue (rating 2-‐4). The 50% chance gridline is shown.
Discussion
Effect of Task and Visibility on WM Accuracy on Target-‐Present Trials In conclusion, some unexpected results were found. First, during the analysis of variance for performance on target-‐present trials (task vs. visibility), we found no significant main effect for the task. This suggests that the additional memory task, which had to induce a visual WM load on conscious working memory, did not have any effect. Besides, the mean WM accuracy on the dual task lies a bit higher than the mean accuracy on the single task, both during memory cue visible and invisible trials.
Furthermore, neither during the single nor dual task did participants performed above 50% chance on the invisible present memory cue trials. From a Soto and colleagues’ (2011) point of view we expected participants to perform both during single and dual task above 50% chance, and because an invisible memory cue is processed by unconscious WM we did not expect any interference during the dual task with the visual WM load that is processed by conscious WM; however no interference was observed at all (in the form of a decrease in WM accuracy during the dual task). The first contradicts the assumption that invisible information can be processed by unconscious WM. The second either contradicts the notion that conscious and unconscious WM can be dissociated or it suggests that the visual WM load was not sufficient to interfere with WM accuracy. The latter is more likely, because also during visible cue trials the visual WM load did not interfere with WM accuracy during the dual task.
To conclude, the present results do not support the notion of processing invisible visual information by unconscious WM. Further, the results emphasize
a methodological implication regarding the visual WM load that should have interfered with WM accuracy during the memory cue present dual task trials.
Effect of Visual WM Load on Visibility and WM Accuracy
The results of the first ANOVA showed that a main effect was found for the task (on visibility). This implies that during memory cue present trials participants were more likely to perceive the cue during the single task
compared to the dual task. So, contrary to the former conclusion, this suggests that the visual WM load interfered with introspection of the WM content. This could be seen as the subjective measurement.
However, the second ANOVA shows different results. Here, no main effect is found for the task (on WM accuracy), meaning that the visual WM load does not influence WM accuracy. In other words, the visual load is not interfering with WM accuracy; this could be seen as the objective measurement.
Thus, this suggests that dissociation could be made between the subjective and objective measure. Namely, as Bona et al. (2013) showed, a dissociation between actual WM content (objective; accuracy) and what we
experience as the WM content (subjective; visibility).
Evaluation of Visual WM Load: Dual Task
There is some uncertainty about the effect of the dual task, regarding the visual WM load. There are alternative explanations for the dissociation that is mentioned above. First, as mentioned earlier, during the single trials the green additional memory cue was still presented to control for any purely visual effect. There is a possibility that this ‘green flash’ alone, presented during the delay,
interfered with WM accuracy. So, if this was true, the actual visual load was present in both dual and single trials. This could explain why no significant difference was found between WM accuracy in dual versus single trials; because during both tasks the additional memory cue could have interfered with WM accuracy, even if participants were instructed to ignore this additional cue during the single trials. This could have been tested if trials were added where nothing is presented during the delay.
This brings us to the next question, namely: was this additional memory cue actually depleting on WM or is there maybe another type of memory
involved? A clear distinction has to be made between visual WM (VWM) and other visual short-‐term memory (VSTM) stores. Sperling (1960) for instance proposed an iconic memory: a high-‐capacity store, reflecting an entire
representation of the visual field, which decays rapidly (<1000 ms). Attention allows a few items to be selected for a more durable and robust representation in VWM.
Sligte and colleagues (2010) proposed another VSTM store, namely fragile visual short-‐term memory (fragile VSTM). Fragile VSTM supposedly has a lower capacity than iconic memory, but stores high-‐resolution representations. Subsequent, VWM stores only one or a few high-‐resolution representations. In Soto’s experiment (2011) a to-‐be-‐remembered orientation cue had to be maintained until the end of the trial where it was compared with a target stimulus. However, no manipulation was involved which excludes the use of VWM but rather some kind of VSTM (Persuh et al., 2017).
To conclude, if assumed that the additional memory cue used in this study depletes on some kind of VSTM, this could explain why the additional memory
cue interferes in both the single and dual trials. To be more specific, this would mean that the original memory cue is held in this VSTM and that, during both single and dual trials, the additional memory cue is interfering with this VSTM. Consecutive, this would result in similar accuracy during both single and dual trials.
Subliminal Processing: Priming Effects
We have to take into account different types of non-‐conscious processing, which are subliminal and preconscious processes. Kouider and Dehaene (2007) assume that during subliminal processing the strength of the stimulus is too weak and insufficient to induce global ignition, in other words large-‐scale reverberation, which means that the information cannot enter the global workspace and thus consciousness. Respectively, during preconscious
processing the inability of the information to enter consciousness is the result of a lack of top-‐down attention rather than bottom-‐up strength. So, the information is accessibly but not accessed due to the absence of attentional amplification. According to Kouider and Dehaene masking prevents the bottom-‐up stimulus activation and therefore masking makes it impossible for information to enter global workspace and thus (‘unconscious’) WM. In that way, masked stimuli reflect behavioural priming effects. This suggests that the masked cue used in Soto’s experiment (2011) is processed subliminally resulting in priming effects rather than the use of an unconscious WM process.
Conclusion
Taking together the results outlined above, we can conclude that we failed to obtain robust evidence supporting the notion that unconsciously perceived visual information could be maintained in WM to be later used in an explicit discrimination task. However, the observed dissociation between WM content and the introspection of this WM content is promising for future
research into the possible ability of WM operating on unconscious information.
References
Atkinson, R. C., & Shiffrin, R. M. (1968). Human memory: A proposed system and its control processes. Psychology of learning and motivation, 2, 89-‐195. Baddeley, A. (2000). The episodic buffer: a new component of working
memory?. Trends in cognitive sciences, 4(11), 417-‐423.
Cowan, N. (1988). Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human information-‐processing system. Psychological bulletin, 104(2), 163.
Dehaene, S. (2014). Consciousness and the brain: Deciphering how the brain codes
our thoughts. Penguin.
Kouider, S., & Dehaene, S. (2007). Levels of processing during non-‐conscious perception: a critical review of visual masking. Philosophical Transactions
of the Royal Society B: Biological Sciences, 362(1481), 857-‐875.
Nieuwenhuis, S., & de Kleijn, R. (2011). Consciousness of targets during the attentional blink: a gradual or all-‐or-‐none dimension?. Attention,
Perception, & Psychophysics, 73(2), 364-‐373.
Oberauer, K. (2002). Access to information in working memory: exploring the focus of attention. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 28(3), 411.
Oberauer, K., & Hein, L. (2012). Attention to information in working
memory. Current Directions in Psychological Science, 21(3), 164-‐169. Schmidt, T. (2015). Invisible Stimuli, Implicit Thresholds: Why Invisibility
Judgments Cannot be Interpreted in Isolation. Advances in Cognitive
Sligte, I. G., Vandenbroucke, A. R. E., Scholte, H. S., & Lamme, V. A. F. (2010). Detailed sensory memory, sloppy working memory. Frontiers in Psychology, 1, 175-‐175.
Soto, D., Mäntylä, T., & Silvanto, J. (2011). Working memory without consciousness. Current Biology, 21(22), R912-‐R913.
Sperling, G. (1960). The information available in brief visual presentations. Psychological monographs: General and applied, 74(11), 1-‐29.
Stein, T., Kaiser, D., & Hesselmann, G. (2016). Can working memory be non-‐ conscious?. Neuroscience of Consciousness, 2016(1), niv011.
Trübutschek, D., Marti, S., Ojeda, A., King, J. R., Mi, Y., Tsodyks, M., & Dehaene, S. (2016). A theory of working memory without consciousness or sustained activity. bioRxiv, 093815.
Persuh, M., LaRock, E., Berger, J. (2017). Working memory and consciousness: The current state of play.
