Can working memory be unconscious?

Can Working Memory Be Unconscious? Master thesis by Carlo Rooth Student number: 10198350

Supervisor: Timo Stein Co-assessor: Yair Pinto Date: 24-07-2017 Abstract

Recent studies claimed that unconscious working memory (WM) underlies the maintenance of subjectively invisible stimuli. This is contrary to usual theories of WM in which

consciousness experience plays a key role for WM processes. In a study by Trübutschek et al. (2016) it was found that stimuli reported as invisible, can be recalled after several seconds above chance level. This is what the authors called a long-lasting blindsight effect. The current study tried to replicate this effect, and test if a single conscious process or a dual conscious-unconscious process underlies WM. Stimuli were shown around the threshold of perception in a delayed localization task in which the load of WM was manipulated with an additional task. Results show that the blindsight effect is replicated and that performance improves with a better visibility. However, a load of WM did not influence the maintenance of subliminal stimuli. It is discussed how these findings can be interpreted and how future research can improve research in this field.

Table of contents

Page number 1. Introduction

- Working Memory and conscious experience 3

- Unconscious Working Memory 5

- Limitations 6 - Current study 7 - Hypotheses 7 2. Methods - Subjects 8 - Design 8 - Materials 9 - Procedure 9 - Calibration task 10 - Experiment 2 11 - Data analysis 11 3. Results - Results experiment 1 12

- Motivation for experiment 2 13

- Results experiment 2 14

- Additional analyses 15

4. Discussion

- Experimental results 16

- Interpretation and implication 16

- Limitations and future perspectives 17

Introduction

Working memory and conscious experience

Working memory (WM) is the process that is involved in the storing, maintaining and manipulation of stimuli, and this underlies human thought processes (Baddeley, 2003). Other than iconic memory that has a high capacity but decays after a few hundred milliseconds, visual WM has a low capacity, but can be maintained for longer periods of time (Sligte, Vandenbroucke, Scholte, & Lamme, 2010). WM is used to plan and carry out behaviour. Generally, WM is viewed as the combination of multiple components working together, such as recall, attention and planning (Cowan, 2009).

Most studies on working memory found that conscious awareness is needed for WM (Baars & Franklin, 2003; Baddeley, 2003). Active elements of WM like perceptual input, rehearsal and recall are reportable and are therefore conscious (Baars, 2002; Baars & Franklin, 2003; Stein, Kaiser, & Hesselmann, 2016). When you try to remember a phone number, all processes are reportable: reading the numbers, rehearsing them and recalling them for use.

Figure 1. Input is perceived through conscious processes in the global workspace. The GW immediately recruits unconscious networks that carry out functions of WM, such as visuospatial automatisms. These functions are reportable when they are retrieved back in the global workspace. Only then, output can be derived.

The Global Workspace Theory (GWT) is based on the assumption that active functions of WM are reportable and therefore conscious. These functions are normally unconscious and distinct from each other. But through consciousness they are centralized in what is called the global workspace. Only items that are retrieved back in the global workspace can be used for WM functions. In that way, consciousness is needed for WM. So, according to GWT,

many specialized networks (Baars, 2005). Figure 1 describes a schematic process of the GWT. TOP-DOWN ATTENTION BOTTOM-UP STIMULUS STRENGTH Absent Present

WEAK OR INTERRUPTED Subliminal (unattended)

- little activation - no priming - no reportability Subliminal (attended) - strong feedforward activation - short-lived priming - no reportability STRONG Preconscious - intense activation - priming at multiple levels - no reportability Conscious - intense activation - stimulus maintaining in WM - conscious reportability

Table 1.Distinction between subliminal, preconscious and conscious processing. Derived from Dehaene et al. (2006).

According to GWT conscious experiences play a key role in perception and WM. The Global Workspace operates as a central process that receives input and executes other

functions (Baars & Franklin, 2003). In line with the GWT, some researchers make predictions how unconscious stimuli are processed. This implicates a distinction between conscious and unconscious processing (Dehaene et al., 2006; Kouider & Dehaene, 2007). Three levels of processing are distinguished: subliminal, preconscious and conscious processing (see Table 1). During subliminal processing, the strength of a stimulus is very weak or the presentation of the stimulus is interrupted, for example by masking. In terms of GWT, the global

workspace is less activated. For that reason, subjects cannot report the stimulus, but activation can be sufficient to reach the semantic level. This priming effect results in above chance performance for stimuli that are reported as invisible. The priming effect only occurs when subject’s attention for the stimulus is present. However, the effect is weak and quickly decays after a second (Dehaene et al., 2006). This suggests that processing of subliminal stimuli is strong enough to evoke a priming effect on the short term, but that it is not strong enough to be maintained in WM. Moreover, it suggests that a conscious process of attention is needed. Without top-down attention there is no priming effect (Dehaene et al., 2006).

Unconscious working memory

Thus, both GWT and subliminal processing theory suggest a key role for consciousness and that stimuli cannot be maintained in WM without conscious awareness (Baars & Franklin, 2003; Dehaene et al., 2006). However, some argue that the content of WM can partly be unconscious (Bergström & Eriksson, 2015; Soto, Mäntylä, & Silvanto, 2011; Trübutschek et al., 2016). For instance, in a study subjects were presented with Gabor stimuli that were either tilted 30 degrees clockwise or counter-clockwise. The stimuli were directly masked after 17ms to prevent conscious perception. Subjects were asked to recall the orientation of these stimuli. It was found that subjects perform significantly above chance, when they report the stimulus as unseen. Since this effect lasted for more than 5 seconds, the authors claim that unconscious stimuli can be maintained in WM (Soto et al., 2011).

Accordingly, the ‘conscious copy model’ by Jacobs & Silvanto (2015) explains how we become aware of content in visual WM. The conscious domain is needed when the content in WM is introspected. It is proposed that the conscious domain does not operate on the original memory trace, but it requires a new representation. So, when content in WM is introspected, a ‘conscious copy’ of the original unconscious memory trace is made (Jacobs & Silvanto, 2015). In forced-choice tasks, the conscious domain is not needed. Content in WM is then retrieved on the original memory representation, which is unconscious (Jacobs & Silvanto, 2015; Soto & Silvanto, 2014). Thus, according to the conscious copy model, consciousness is not needed for WM.

A recent paper by Trübutschek et al. (2016) tested the reality of unconscious working memory with a delayed localization task. Stimuli were presented using a circular masking paradigm (CMP). A target was presented for 17ms on the screen followed by 20 spatial masks in a circle for 233ms. In that way, stimulus strength was too weak to perceive the target consciously, so subliminal processing was needed (Dehaene et al., 2006). After a variable delay period of 2.5 – 4.0 seconds participants had to localize the target by choosing one of 20 letters that replaced the masks, thereby making a guess if they reported that they did not see a target. This experimental design is shown in Figure 2a.

In the second experiment, they tested if the memory of subjectively visible and

subjectively invisible information gets impaired when the conscious WM is already loaded with another task. The task was a digit span test in which subjects had to remember 1 – 5 numbers in random order and recall them after the delay period.

Figure 2b shows the results of the experiments. It was found that subjects are able to localize targets above chance that are reported as invisible, even after 4 seconds. This ‘long-lasting blindsight effect’ was not impaired by the load of conscious working memory for both subjectively visible and subjectively invisible trials. Since conscious perception was not needed to localize a target, the authors conclude that working memory can exist without consciousness (Trübutschek et al., 2016).

Figure 1a. Design of the experiment by Trübutschek et al. (2016). Figure 2b. Results of the experiment (Trübutschek et al., 2016). Performance improves with better visibility, but subjects are able to maintain unseen locations above chance.

Limitations

However, the results of the study do not necessarily imply unconscious WM. The long-lasting blindsight effect could also be explained by a weak conscious process, which

eliminates the need for a separate unconscious process (Stein et al., 2016). Subjective ratings of visibility may evoke response biases, because subjects indicate stimuli as unseen, while they might have some weak experiences of it. Thus, the above-chance performance reflects weak conscious information that is maintained in conscious WM over the delay-period (Stein et al., 2016). That implicates there is only one conscious process that mediates the

performance on the localization task.

Moreover, the digit span test is not sufficient to load spatial WM. The digit span test is a widespread test to measure verbal WM (Baddeley, 2003), whereas the localization task measures spatial WM. These are implemented by different neural structures (Baddeley, 2003;

Smith, Jonides, & Koeppe, 1996). Therefore, a test is needed that loads spatial WM to falsify the long-lasting blindsight effect.

The current research

The current study will test if subjectively invisible stimuli need to be accounted for by unconscious WM or whether it can be explained by weakly conscious WM. In that way, unconscious WM is not needed to account for the above chance performance on subjectively invisible trials. The main research question is: does WM rely on a single process or a dual-process? More specifically, does performance on the delayed localization task for subjectively invisible stimuli needs to be accounted for by unconscious WM or by weak-conscious WM?

To test our research question the concept of unconscious WM and weak-conscious WM needs to be specified. One implication of unconscious WM is a dual-process model with two processes: one conscious process and one unconscious process. The conscious and

unconscious processes rely on different resources, namely the unconscious original

representation and the conscious copy representation in WM (Jacobs & Silvanto, 2015). This will be tested in the current study.

The alternative we propose is that one weak conscious process underlies WM, comparable to the global workspace theory as proposed by Baars & Franklin (2003). The GW model implies that performance for subjectively invisible stimuli is mediated by the same conscious process as for subjectively visible stimuli. The implication is that they rely on the same resource.

In the current study subjects performed a single task and a dual-task to test if WM relies on a single process or on a dual-process. In the single task subjects performed the delayed localization task as in the experiment of Trübutschek et al. (2016). In the dual-task subjects needed to memorize four more targets. These were visibly presented in the delay-period. For more details of the tasks see the methods section under the header procedure.

Hypotheses

If WM relies on a single process, performance on the dual-task would be impaired

compared to performance on the single task, because WM is loaded and does not have enough capacity to perform well on both tasks (Cowan, 2009; Trübutschek et al., 2016). If WM relies on a dual-process with two resources, performance on the dual-task would not be impaired, because the unconscious process would account for the one task and the conscious process for the other task. The following hypotheses are derived.

1. According to the dual-process model, it is expected that locations of subliminal stimuli can be recalled above chance after several seconds, meaning that the long-lasting blindsight effect of Trübutschek et al. (2016) would be replicated. According to the single-process model, it is not expected to replicate the blindsight effect.

2. According to both the dual-process model and the single-process model, it is expected that performance is better for stimuli reported as seen than for stimuli reported as unseen.

3. According to both the dual-process model and the single-process model, it is expected that performance is better when WM was not loaded compared to when WM was loaded with an additional task.

4. Additionally, if WM relies on a single-process with one resource, WM-load would influence performance on trials rated both as visible and trials rated as invisible. But if WM relies on a dual-process with two resources, WM-load would only influence performance on trials rated as visible. There would be no differences on invisible trials between the no WM-load condition and WM-WM-load condition.

According to the single-process model, it is expected that performance for both

subjectively visible stimuli and subjectively invisible stimuli would be impaired, when WM is loaded with an additional spatial task. According to the dual-process model, it is expected that the conscious WM-load only influences performance on subjectively visible trials, and that performance would not be influenced by the conscious WM-load on subjectively invisible trials.

Methods

Subjects

In total, 43 participants were recruited from the University of Amsterdam (UvA) for experiment 1 (N = 12, Mage = 22.9, SDage = 4.8, 6 male, 8 right-handed) and experiment 2 (N = 31, Mage = 22.4, SDage = 3.0, 14 male, 27 right-handed). They earned 1.5 participation credits or €15 for their participation.

Design

Two experiments were conducted in a within-subjects design. The task that subjects had to perform was based on the delayed-localization task from Trübutschek et al. (2016). During the experiment, subjects were presented with a masked target for 17ms. After a delay period of 2 seconds, subjects were asked to recall the location by choosing 1 out of 20 possible

positions in a circle. The two dependent variables were the measures for performance and the measure for visibility. The independent variable was the manipulation of WM load.

Performance was measured by the correctness of the response and the precision of the response. A response is correct if it is within two positions of the target, so for every trial 25% of the positions was counted as correct (5/20). Also, the precision of the response was

measured by the deviation from the correct location. For instance, a mean distribution of 2.5 indicates that a subject was on average 2.5 locations away from the correct location.

Visibility served as an indication of awareness. It was measured by the score on the Perception Awareness Scale (PAS). This is a scale from 1 to 4, with the values 1: no experience of the target, 2: brief glimpse, 3: almost clear experience, 4: clear experience (Sandberg, Timmermans, Overgaard, & Cleeremans, 2010). A target was rated as unseen for value 1, and seen for values 2 – 4.

The WM load was used to manipulate a load on the conscious working memory. This manipulation was done to test if a load on conscious spatial WM interferes with the

maintenance of subjectively visible and subjectively invisible stimuli. In both experiments, there was a non-WM load and a WM load condition. The WM load was carried out by a dual-task in which subjects had to recall four additional targets.

Materials

The experiments took place in a cubicle in UvA LAB. The experiments were installed on a computer running on Windows 7. It was presented on a 21” monitor with a refresh rate of 60Hz. The experiments were programmed in MATLAB with the use of the toolboxes

Psychtoolbox and Palamedes. Psychtoolbox is a toolbox extension in MATLAB to accurately show controlled visual stimuli. Palamedes is used to measure subjects sensitivity for stimuli (Prins & Kingdom, 2009). More details about that procedure are provided in the next section. To avoid reflections on the screen, the light in the cubicle was dimmed to the lowest level.

Procedure

Two experiments were conducted. The procedure of experiment 1 can be seen in Figure 3. The experiment consisted of two conditions. In the ‘single condition’ was no load of working memory. Subjects only needed to remember the first target and were instructed to ignore the stimuli in the delay-period. In the ‘dual condition’ was a load of conscious working memory. WM was loaded by instructing subjects to memorize both the first target and the four targets in the delay-period.

Each trial had the following procedure. First a fixation cross was presented for one second. A small grey square served as the target in the experiment. It was shown for 17ms on one out of 20 possible locations in a circle. All 20 locations were immediately masked for 200ms. The 20 mask elements consisted of four small squares that surrounded the target square. In that way, subjects could hardly perceive the target consciously. A delay-period of three seconds followed. During this delay-period four other squares were flashed. Because these were not masked, they were perceived consciously. Next, the 20 locations were replaced by letters. Subjects were asked to recall the location of the first target by pressing the corresponding letter on the keyboard. After that they had to rate the visibility of the target on the PAS.

Only in the dual condition, subjects were asked to recall the four other targets by pressing the corresponding letters. In the single condition subjects were instructed to ignore the four other targets and did not need to recall them.

Each condition block consisted of 80 trials on which the target was absent in 25% of the trials. Subjects alternately participated twice in both conditions.

Figure 3. Design of the current study.

Calibration task

Prior to the experimental conditions, a calibration task was performed to estimate subjects’ perceptual threshold. Subjects completed 120 trials in which they had to recall the location of

the subliminal target and rate the visibility of the target. The luminance of the mask changed according to the visibility responses of the subjects. The contrast between the target and the mask increased when subjects rated the target as invisible (PAS rating 1). This was done by lowering the luminance of the mask elements. The mask contrast was decreased by increasing the luminance of the masks when they rated it as visible (PAS rating 2, 3 or 4). This is called a up/down procedure and was carried out by an adaption function in Palamedes toolbox (Prins & Kingdom, 2009).

At the end of the calibration task individual thresholds were calculated by averaging the mask contrasts from the last four switches from visible to invisible, and the last four switches from invisible to visible. The luminance of the average mask contrast was used in the

experimental conditions.

Experiment 2

The design of experiment was similar to experiment 1, but differed slightly in both conditions. In the ‘single condition’ in experiment 2 a blank screen was presented. So, no additional targets were flashed. In the ‘dual condition’ in experiment 2 four targets were sequentially flashed for 17ms. So, the targets were not flashed all at once. To estimate subjects’ perceptual threshold, the same calibration tasks were performed as in experiment 1.

Data analysis

Data was collected in MATLAB and SPSS was used for statistical analysis. Unfortunately, the data of one participant in experiment 1 and three participants in experiment 2 could not be included in further analysis because they responded on all trials visible or on all trials

invisible. That means no differences between seen and unseen trials could be analyzed, and further analyses are therefore performed on 11 subjects in experiment 1 and on 28 subjects in experiment 2.

For both experiments the four hypotheses were tested. This was performed with a repeated measures ANOVA, in which the condition and visibility served as the factors and the

accuracy and precision served as the measures. Per hypothesis it was checked which parts of the ANOVA were needed to reject or meet the hypothesis. All hypotheses were tested against an alpha level of .05.

To test the first hypothesis, performance was analyzed for subjectively unseen trials, so trials in which subjects responded 1 on the PAS. This was tested with a t-test in which only

the measure of accuracy was tested against the chance level of .25. If performance was above chance, the first hypothesis would be met for the dual-process model.

The second hypothesis was tested by comparing performance in invisible trials (PAS: 1) with visible trials (PAS: 2,3,4) in both conditions, so a main effect of visibility was tested. This was done for both measures of accuracy and precision. If performance for visible trials was significantly better than invisible trials, the second hypothesis would be met.

The third hypothesis was tested by comparing performance in the non-WM load condition with the WM-load condition in both visible and invisible trials, so a main effect of condition was tested. This main effect was also tested on both measures. The third hypothesis would be met, if performance was better on the blocks without a WM-load.

The fourth hypothesis was tested by comparing the effect of WM load between invisible and visible trials. It was tested if WM-load has a different effect on invisible trials compared to visible trials. So, the interaction effect between condition and visibility was tested. For the single-process model the hypothesis would be met, if the effect of WM-load would be the same for both visible and invisible trials. That means performance in the ‘dual condition’ is impaired compared to the ‘single condition’. For the dual-process model the hypothesis would be met, if the WM-load would only influence visible trials, but it would not influence

invisible trials. That means there would be a significant interaction effect of visibility and condition.

Results

Results experiment 1

The first hypothesis stated that locations of subliminal stimuli can be recalled above chance. This chance level was .25, since for the twenty possible locations two places around the target were also counted as correct. This long-lasting blindsight effect was confirmed in both conditions. Participants performed above the chance level of .25 on invisible trials in both the no WM-load condition, t(10) = 5.39, p = .001, and the WM-load condition, t(11) = 4.09, p = .002. That means the first hypothesis is met.

Next, the second hypothesis stated that performance for seen stimuli is better than for unseen stimuli. This hypothesis was also met, since an overall main effect of visibility was found. This effect existed in both the accuracy measure, F(1,10) = 75.71, p < .001, and the precision measure, F(1,10) = 65.37, p < .001. Subjects performed better for stimuli that were seen than unseen. That means that the second hypothesis is met.

However, the third hypothesis could not be met. The main effect of condition was not found, in both accuracy, F(1,10) = 1.78, p = .211, and precision, F(1,10) = 2.02, p = .186. Subjects did not perform worse when their WM was loaded with a dual-task.

To test the fourth hypothesis, the effect of WM-load needed to be analyzed further. It was found that performance was impaired for invisible trials, when WM was loaded with an additional task, t(10) = 2.33, p = .042, but that this effect was not found for visible trials, t(10) = .07, p = .94. So, that means an interaction effect was found, but it was not caused by a dual-process of WM.

Figure 4. Plots of the mean accuracy and precision in experiment 1 with error bars. The blue lines represent the ‘single condition’ without WM-load, the red lines represent the ‘dual condition’ with the WM load, and the dashed line represents the chance level of .25. The two conditions showed no statistical significant difference in accuracy and precision.

Motivation for experiment 2

The first experiment did not found a main effect of condition, which makes it hard to speculate if WM relies on a single-process with one resource, or on a dual-process with two resources. A possible explanation why the main effect was not found might be that the conditions were not different enough. Subjects might have attended the four additional targets in the single condition, though they were instructed to ignore them. Testing this hypothesis is crucial to answer the research question of this study.

Therefore, the experiment was modified in experiment 2. First, a blank screen appeared without any additional targets during the delay-period in the single condition. Second, four additional targets were sequentially flashed in the dual-condition, not simultaneously. With these modifications, it was expected that a main effect of condition might be found and the research question could be answered.

Results experiment 2

The hypotheses of experiment 1 were also applied in experiment 2. Also in experiment 2, the blindsight effect was confirmed, since subjects could recall the location of subjectively invisible stimuli for both the no WM-load condition, t(27) = 5.16, p < .001, and the WM-load condition, t(27) = 4.43, p < .001.

Next to that, a main effect of visibility was found in both the accuracy measure, F(1,27) = 86.50, p < .001, and in the precision measure, F(1,27) = 93.45, p < .001. So, both the first and second hypotheses were met in the second experiment. Subjects can recall locations of unseen stimuli above chance and they perform better when they perceive the stimulus stronger.

The third hypothesis could only partly be confirmed. The main effect of condition was not found on the accuracy measure, F(1,27) = 2.85, p = .103, but a small effect was found on the precision measure, F(1,27) = 4.57, p = .042. That means when subject’s WM was loaded with an additional task, their answers were less precise, but within the range of correctness.

Again, the effect of WM needed to be analyzed further to test the fourth hypothesis to test if WM relies on a single-process or on a dual-process. It was found that performance on invisible trials was neither impaired on the accuracy measure, t(27) = 1.68, p = .105, nor on the precision measure, t(27) = 1.90, p = .068. This effect was also found for visible trials on the accuracy measure, t(27) = .56, p = .578, and the precision measure, t(27) = 1.34, p = .191. Overall, the study failed to find an effect of WM-load on the performance in the delayed-localization task. Therefore, it is hard to say if WM relies on a single-process or on a dual-process, since there are no dissociations found between the two conditions.

Figure 5. Plots of the mean accuracy and precision of experiment 2. The blue lines represent the ‘single condition’ without WM-load, the red lines represent the ‘dual condition’ with the WM load. The two conditions showed no statistical significant difference in accuracy and precision.

Additional analyses

After analyzing the results, the data was inspected again. The current study counted an answer correct if it was within two places of the target. However, the results seem to shift, when the criterium for a correct answer was more conservative. If only the exact place of the target was counted as correct, a main effect of condition is found indeed.

Subjects do perform better in the single condition compared to the dual condition for trials rated as invisible, t(28) = 2.319, p = .028. This effect also existed in trials rated as visible,

t(28) = 4.426, p < .001. Moreover, no interaction effect between condition and visibility was

found, F(1,27) = .185, p = .671.

So, if performance is measured more conservative, a main effect of condition does exist. Subjects perform worse when their WM is loaded with an additional spatial task. This effect was the same for both invisible trials as visible trials. This result is line with our predictions and leans towards the view that WM relies on a single-process with one resource.

Since the effect is found by altering the dependent variable after data collection, these results should be considered with caution. Future research needs to study this effect better and form hypotheses a priori to confirm the effect of WM-load.

Figure 6. Additional plot of the mean accuracy in experiment 2. Answers were counted as correct when it corresponds with the exact place of the target. Blue lines represent the single condition, red lines represent the dual condition.

Discussion

Experimental results

This study tried to examine if WM relies on single process or on a dual process. This was done by manipulating a load on WM. If WM would rely on a dual process, a load on

conscious WM would not influence the performance for subliminal stimuli, while it should have an effect on performance for stimuli that are perceived on a conscious level. In the current study, this was manipulated by showing stimuli around the threshold of conscious perception. Two experiments were conducted, but in both experiments the load on conscious WM did not influence performance differently. Therefore, this study could not answer the proposed research question. However, when taking a more conservative approach of measuring accuracy, a main effect of condition was found. This is line with the fourth hypothesis, since the effect did not differ across the levels of visibility. It seems that WM relies on a single process with one resource. This result was found after exploratory analyses and was not hypothesized prior to the study. So, this effect should be considered with caution.

Also, the current study found a strong effect of visibility on the performance to maintain locations in WM. Subjects were significantly better in recalling the location of a target, when they claimed to be aware of it. This result implicates a single process. It seems that conscious perception is needed to correctly recall locations, and that there is no dual-process assisting for subliminal stimuli.

Moreover, the current study replicated the long-lasting blindsight effect in a simple

behavioral design. Though subjective awareness is important for recall, it seems that subjects maintain the location of unconsciously perceived stimuli for several seconds above chance. This might suggest the existence of unconscious WM, but the effect may be caused by other factors.

Interpretation and implication

A possible explanation for the observed blindsight effect might be the subjective visibility reports. Two problems may arise with the use of subjective reports. Subjects may categorize a target as unseen (e.g. response PAS: 1) while they may have had a weak awareness for the target (e.g. response PAS: 2). Though subjects were clearly instructed to prevent this miss-categorization, it may have occurred nevertheless. If this was the case during the calibration task, then the estimated thresholds for subjective awareness were not accurate. The contrast between the mask and target was too high. This makes subjects too sensitive to press 1, while

they did perceive the target clearly. That would mean the blindsight effect is caused by miss-categorization.

However, if the observed blindsight effect is an actual phenomenon, it does not necessarily imply the existence of unconscious WM. When presented with a subliminal target, subjects may have form a guess of the location. This guess could have entered awareness through several conscious strategies. For example, one subject tried to remember the location by saying the corresponding time of the clock, while another subject tried to remember the guess by pointing her finger on the screen. The maintaining process would then reflect usual

conscious WM on a guess, instead of unconscious WM on a subliminal target.

Concluding, the current study did not found evidence for the existence of unconscious WM. The observed blindsight effect can be explained by usual theories of WM, in which consciousness serves an important role (Baars & Franklin, 2003). It seems that the Global Workspace Theory best serves the experimental findings, since performance improves with visibility. The unconscious blindsight effect might function as a quick, but less precise route of one conscious process.

Limitations and future perspectives

The study did not found a main effect of condition. Subjects did not perform worse on the task when their WM was loaded with a dual-task. One explanation for the null-result is the measure of performance. Additional analyses showed an effect of WM-load, when the accuracy was measured more conservative. Though this effect was found only after altering the dependent variable after data collection, the results are promising and future research should study this better.

The current paradigm only tested the maintenance process of WM. Previous research found that the manipulation process of WM has different neural substrates (Glahn et al., 2002). An implication could be that the WM-load does not influence the maintenance process of WM, but that it influences the manipulation process of WM. It might be interesting for future research to focus on this distinction.

Furthermore, it is unclear why subjects sometimes fail to detect the target (PAS rating: 1). Failed detection of a target might happen in an early stage or in a later stage of processing. In the early stage subjects fail to perceive the target, since the strength of the stimulus is weak. That implicates subliminal processing during that trial. In a later stage subjects do perceive the target but fail to report the target, since their attention is absent. That implicates

2010). The different ways of processing lead to different activation patterns and therefore different performance. This failed detection differs per trial and per subject and might underlie the results of this study. Future research should take these considerations in account.

Lastly, the current study tried to fit the experimental results in the conscious copy model by Jacobs & Silvanto (2015). However, this model might not be a good model to use. The model is mainly based on content and features of stimuli, while our paradigm is mainly based on spatial locations of stimuli. These are processed in different pathways, also known as the ‘what’ and ‘where’ -stream of visual perception (Goodale & Milner, 1992; Goodale & Westwood, 2004; Mishkin, Ungerleider, & Kathleen, 1983). Nevertheless, it was decided to use this model, since the conscious copy model is one of the few models concerning

unconscious WM. Still, future research should develop a new model for unconscious visual processing for spatial stimuli, so that the existence of unconscious WM can be studied better.

References

Baars, B. J. (2002). The conscious access hypothesis: origins and recent evidence. Trends in

Cognitive Sciences, 6(1), 47–52.

Baars, B. J. (2005). Global workspace theory of consciousness: Toward a cognitive neuroscience of human experience. Progress in Brain Research, 150, 45–53. https://doi.org/10.1016/S0079-6123(05)50004-9

Baars, B. J., & Franklin, S. (2003). How conscious experience and working memory interact.

Trends in Cognitive Sciences, 7(4), 166–172.

https://doi.org/10.1016/S1364-6613(03)00056-1

Baddeley, A. (2003). Working memory: looking back and looking forward. Nature Reviews

Neuroscience, 4(10), 829–839. https://doi.org/10.1038/nrn1201

Bergström, F., & Eriksson, J. (2015). The conjunction of non-consciously perceived object identity and spatial position can be retained during a visual short-term memory task.

Frontiers in Psychology, 6, 1–9. https://doi.org/10.3389/fpsyg.2015.01470

Cowan, N. (2009). What are the differences between long-term, short-term, and working memory? NIH Public Access, 6123(7), 323–338. https://doi.org/10.1016/S0079-6123(07)00020-9.What

Dehaene, S., Changeux, J. P., Naccache, L., Sackur, J., & Sergent, C. (2006). Conscious, preconscious, and subliminal processing: a testable taxonomy. Trends in Cognitive

Sciences, 10(5), 204–211. https://doi.org/10.1016/j.tics.2006.03.007

Glahn, D. C., Kim, J., Cohen, M. S., Poutanen, V., Therman, S., Bava, S., & Erp, T. G. M. Van. (2002). Maintenance and Manipulation in Spatial Working Memory : Dissociations in the Prefrontal Cortex, 213, 201–213. https://doi.org/10.1006/nimg.2002.1161

Goodale, M. A., & Milner, A. D. (1992). Separate visual pathways tor perception and action,

15(1), 20–25.

Goodale, M. A., & Westwood, D. A. (2004). An evolving view of duplex vision : separate but interacting cortical pathways for perception and action. Current Opinion in

Neurobiology, 14, 203–211. https://doi.org/10.1016/j.conb.2004.03.002

Jacobs, C., & Silvanto, J. (2015). How is working memory content consciously experienced? The “conscious copy” model of WM introspection. Neuroscience and Biobehavioral

Reviews, 55, 510–519. https://doi.org/10.1016/j.neubiorev.2015.06.003

Kanai, R., Walsh, V., & Tseng, C. (2010). Subjective discriminability of invisibility : A framework for distinguishing perceptual and attentional failures of awareness, 19, 1045– 1057. https://doi.org/10.1016/j.concog.2010.06.003

Kouider, S., & Dehaene, S. (2007). Levels of processing during non-conscious perception: a critical review of visual masking. Philosophical Transactions of the Royal Society of

London. Series B, Biological Sciences, 362(1481), 857–75.

https://doi.org/10.1098/rstb.2007.2093

Mishkin, M., Ungerleider, L. G., & Kathleen, A. (1983). Object vision and spatial vision : two cortical p hways. Trends in Neurosciences, 6, 414–417.

Prins, N., & Kingdom, F. A. A. (2009). Palamedes: Matlab routines for analyzing psychophysical data.

Sandberg, K., Timmermans, B., Overgaard, M., & Cleeremans, A. (2010). Measuring consciousness: Is one measure better than the other? Consciousness and Cognition,

19(4), 1069–1078. https://doi.org/10.1016/j.concog.2009.12.013

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(OCT), 1–10. https://doi.org/10.3389/fpsyg.2010.00175

Smith, E., Jonides, J., & Koeppe, R. a. (1996). Dissociating Verbal and Spatial Working.

Soto, D., Mäntylä, T., & Silvanto, J. (2011). Working memory without consciousness.

Current Biology, 21(22), R912–R913. https://doi.org/10.1016/j.cub.2011.09.049

Soto, D., & Silvanto, J. (2014). Reappraising the relationship between working memory and conscious awareness. Trends in Cognitive Sciences, 18(10), 520–525.

https://doi.org/10.1016/j.tics.2014.06.005

Stein, T., Kaiser, D., & Hesselmann, G. (2016). Can working memory be non-conscious ?

Neuroscience of Consciousness, 2(2), 1–3.

Trübutschek, D., Marti, S., Ojeda, A., King, J., Mi, Y., Tsodyks, M., & Dehaene, S. (2016). A theory of working memory without consciousness or sustained activity. bioRxiv, 93815. https://doi.org/10.1101/093815