Working memory and consciousness: The effect of distracting information on metacognition Keijser. C. Mike
University of Amsterdam Brain and cognition department
Student: Keijser. C. Mike; Student nr: 10332324; Supervisor: dr Stein. Timo; Publishing date 21-07-2017; Research: Masterthesis; University of Amsterdam
Abstract
There are several theories that seek to explain the link between working memory and consciousness. This relationship can be investigated by measuring metacognition. One of the more novel theories states that working memory contents and their conscious experience can be dissociated. Key evidence for this theory comes from a single research, which demonstrated that, by the introduction of an invisible distracter, working memory performance and the metacognitive ability of tracking this performance could be dissociated. In our research we reinvestigated this theory with two experiments. In the first experiment we replicated the previous research for this key evidence. A delayed cue target discrimination task was used with raw confidence scores as a measure of metacognition. In the second experiment an improved measure for working memory was used to investigate if the link between working memory and metacognition depends on how working memory is measured. Besides raw confidence scores a signal detection type 2 analysis was executed to investigate the effect of the distracter on the ability to discriminate between correct and incorrect responses (metacognitive sensitivity). In both experiments no condition has been found that could indicate dissociation between working memory and the metacognitive ability of tracking this performance. Differences in the effects of visibility and orientation of the distracter on working memory accuracy and confidence indicate different underlying processes for different types of tasks. Metacognitive sensitivity was not influenced over all conditions in both experiments. We suggest that metacognition and its properties could give more insights in the relationship between working memory and consciousness and therefore should be considered as one of the key factors in this line of research.
Working memory and metacognition cannot be dissociated when metacognition is assessed through raw confidence scores.
Introduction
The last few decades, consciousness has been linked to a great variety of neurological processes. For most of these cognitive processes a state of consciousness is assumed to be necessary for its function. Two of the processes that are proposed to require a state of consciousness are working memory and metacognition.
Working memory (WM) is the active and temporary representation of information that is maintained for the short term. Working memory involves the temporary retention of information just experienced or just retrieved from the long-term memory. When information is maintained in working memory, it can be manipulated in ways that makes it useful for goal-directed behaviour (Cowan, 2008). An important feature is that WM can be consciously accessed, manipulated and examined (metacognition).
Metacognition can be defined as higher order thinking which involves the knowledge that someone has about its own cognitive performance, experience, processes and strategies “cognition about cognition” (Rademaker et al., 2012). Metacognition is the active control over cognitive processes involved during learning, and has been associated with intelligence (Daseking et al., 2015; Borkowski et al., 1987). Metacognition is commonly assessed with the use of confidence ratings, which are the subjective assessments of participants’ performance on memory tasks (Bona & Silvanto, 2014).
So why is metacognition thought to be linked to working memory?
Traditional models of the relationship between working memory and consciousness hold that the contents of working memory are continuously accessible to consciousness, and can be continuously manipulated and examined. Therefore, evaluating your own working memory performance based on the working memory content, seems to be linked by consciousness (Rademaker et al., 2012). However, this notion has been challenged by counterintuitive findings from recent studies, showing that under specific circumstances metacognitive judgments can be completely dissociated from working memory performance (Bona et al., 2013).
Are we always consciously aware of information that we try to retain in memory over short periods of time, or can we remember without consciousness? Currently there are four prominent cognitive theories concerning the relationship between working memory and consciousness. The dominant view in the literature is that consciousness and working memory
are very closely related, such that conscious experience reflects the contents of working memory, and the contents of working memory are consciously accessible. (Richet, 1886; Baars & Franklin, 2002; Baddeley, 1992). A second framework holds that consciousness is richer (i.e. has a higher capacity) than working memory, meaning that some conscious contents are residing outside working memory (Block, 2011; Lamme, 2003). In contrast to this view, a third theory states that conscious experience is only a part of the working memory, and that working memory can contain information that is only consciously experienced when explicitly attended (Cowan, 1988; Oberauer, 2002, 2009). Finally, a more radical, recently proposed fourth theory states that working memory contents and their conscious experience can be fully dissociated (Jacobs & Silvano, 2015), such that consciousness is always based only on a copy of working memory contents, and working memory contents are never consciously experienced directly. Key evidence for the novel theory holding that working memory and consciousness can be dissociated comes from a study by Bona and colleagues (Bona et al., 2013, demonstrating that, under specific circumstances, WM performance and one’s metacognitive ability of tracking this performance can be dissociated. Taking metacognitive ability as a measure of subjective awareness, this indicates that WM performance can be dissociated from subjective awareness, thereby challenging the dominant view in the literature that WM and consciousness are closely linked. This study forms the basis of the proposed theory.
The fourth theory as described above, challenges traditional views on the relationship between working memory and consciousness. Evidence for this theory comes from a double dissociation between working memory performance and metacognitive judgments about this performance. The key evidence for this theory was the existence of a condition in which WM was impaired whilst metacognition assessed through vividness ratings was unaffected.
The previous study by Bona and colleagues (Bona et al., 2013) investigated the link between working memory accuracy and metacognition. This study is of particular relevance for this research, because it provides evidence for a double dissociation between working memory accuracy and metacognition, challenging traditional models of their relationship. A delayed cue-target orientation discrimination task like the one in Figure 1 was used to determine working memory accuracy and metacognition. This was a visual task where participants were asked to hold in mind the orientation of a grating (memory cue). The participants had to indicate if a memory probe was tilted to the right or the left relative to the memory cue. Distracters were found to influence the subjective experience. Memory cue, memory probe and the distracter were all similar gratings with different orientations. Besides
the effect of distracters on the subjective experience, subjective experience did not always accurately reflect the underlying working memory representation. Working memory was impaired for visible and invisible distracters, but only if the orientation differed by 40°. For smaller orientation differences, there was no difference in working memory accuracy. Vividness was reduced by distracters of all orientations, when they were judged to be invisible, relative to when they were judged to be visible. These results suggest that metacognition and working memory rely on separate representations. If metacognitive processes are directly linked to working memory, then manipulations of the working memory accuracy should have affected its metacognition. However, this was not the case. As described above, this seems inconsistent with the view that metacognition involves direct and accurate access to working memory. The main goal of our research is to re-investigate the relationship between the subjective experience (metacognition) of the memory representation and its objective accuracy (working memory accuracy) by replicating and adjusting their experiments. The double dissociation described above suggests that working memory and metacognition rely on separate representations. Because the whole theory, that working memory and consciousness can be dissociated, rests only on the findings described by Bona and colleagues (2013), a replication and an improved method will substantiate the reliability of this theory.
Concerning the assessment of metacognition, in the study by Bona and colleagues (2013), vividness was used to determine metacognition. Vividness, which is referred to as “how vivid the item held in memory was” reflects the richness of representations in working memory (Baddeley & Andrade, 2001). In our experiments, we used confidence scores as a measure of metacognition, because these confidence scores tend to be more evaluating one’s performance on the task, instead of the memory cue’s representation in working memory (Fleming & Lau, 2014). Confidence scores are measured by taking scores on an ordinal scale. Besides investigating these raw confidence scores we will use a signal detection type 2 analysis (SDT-2) to investigate the metacognitive sensitivity (i.e. metacognitive performance). This measure of metacognition provides an unbiased, criterion-free index of how well observer’s confidence scores discriminate between their own correct and incorrect responses (Fleming & Lau, 2014).
To determine if working memory contents and their conscious experiences can be dissociated we executed two experiments. In the first experiment, we investigated if metacognitive judgements (confidence ratings) track working memory performance accurately (even in the presence of invisible distracting information), or if metacognition can
be dissociated from objective working memory accuracy. And if results depend on how metacognition is measured, for example when using criterion-free indices further explained below. In the second experiment, we determined if the link between metacognition and working memory accuracy depends on how WM performance is measured. Using an improved measure for the working memory accuracy outcome.
We expect that for both experiments metacognition, as assessed through confidence ratings, will in generally track WM performance (Vandenbroucke et al,, 2014). However, this link breaks down when the subjectively invisible distracting information is introduced, such that confidence ratings are lower when distracting information is invisible, whereas WM performance is not affected by visibility of the distracting information (Silvanto and Soto, 2012; Bona et al., 2013). These findings may be specific for measuring metacognition through raw confidence ratings, but may differ when assessing metacognition by criterion-free type-II signal detection theory analyses assessed as metacognitive sensitivity (Fleming & Lau, 2014; Stein & Sterzer, 2014).
Materials and Methods
Experiment 1 Participants
In total 35 students of the University of Amsterdam with normal or corrected to normal vision participated in the experiment. All the participants were naive to the aims of the study and were given an elaborate instruction and were only allowed to participate if they fully understood the task. 31 participants were included in the data analysis (7 males, mean age 21.29 and 24 females, mean age 19.58). Four participants were excluded due to poor performance (overall accuracy below 50%). All participants provided informed consent and were given 1.5 research credit as a reward for their participation
Stimuli and Experimental procedure
Participants viewed the stimuli in a dark room on a 21” (1920 x 1080 pixels) luminance calibrated ASUS LCD monitor with a refresh rate of 60Hz. The screen presentation times were controlled using a photodiode measurement to make sure the programmed presentation times did not differ from the on-screen presentation. Stimuli and tasks were controlled by MATLAB using Psytoolbox (Brainard, 1997; Pelli, 1997; Kleiner et al., 2007). The task required the maintenance of a sinusoidal luminance-modulated Gabor Patch (memory cue) in VSTM while a masked distracter grating (also a Gabor Patch) was
presented on half of the trials during the delay period. The average brightness of the background was the average brightness of the gratings (0,1 Michelson contrast; spatial frequency 1 cycle/”; diameter 4° of visual angle from viewing distance of 72 cm) variables 4deg = 166 pixels (Bona et al., 2013).
On each trial (see Figure 1) subjects were shown a Black fixation cross in the middle of the screen (1000ms), followed by a blank screen (500ms), the memory cue (300ms), black mask (83ms), blank delay (1500ms), distracter or blank (17ms), black mask (83ms), blank delay (1500 ms), fixation cross (500ms), memory probe (300ms) and a blank delay (100ms). After this the participants were forced to response for the memory task, confidence task and visibility of the distracter.
The orientation of the memory cue (10°, 20°, 30°, 40°, or 50° to the left or right from vertical) had to be kept in mind by the subject. On half of the trials, a distracter grating (the spatial frequency, contrast, size, phase and location were the same as that of the memory cue) appeared during a 3.1 second maintenance interval. The orientation of the distracter was either
Figure 1| Timeline of experiment one. The orientation of the memory cue had to be kept in memory by the
participants. At the end of each trial, participants were asked if the shown memory probe was tilted more to the left or right relative to this memory cue. After the left right response the participants had to respond how confident they were about the left right answer on a 1-4 scale (Confidence in response?; 1:Very low, 2:Low, 3:High, 4:Very high). In the period between the memory cue and memory probe in 50% of the trials a distracter was presented (grating with the same size, contrast and spatial frequency as memory cue/probe) this distracter could be identical to the memory cue, or could differ 10°/40° . After the left right response and the confidence rating, participants were asked how visible the distracter was (Awareness of distracter; 1:Did not see the distracter, 2:Maybe saw something, 3:Saw the distracter but not its orientation, 4:Saw the distracter and its orientation).
identical (0°) to the memory cue, or orientated 10°/40° in a congruent way. The distracters were masked and presented for only 17 ms like Bona et al. (2013). At the end of the delay period a memory test probe was presented that was tilted 10° to the left or to the right relative to the memory cue, and participants were asked to answer if the memory probe was left or right orientated in contrast to the memory cue (left, right arrow-key). After that they had to answer how confident they were about their answer (Confidence in response?; 1:Very low, 2:Low, 3:High, 4:Very high) and rate the visibility of the distracter (Awareness of distracter; 1:Did not see the distracter, 2:Maybe saw something, 3:Saw the distracter but not its orientation, 4:Saw the distracter and its orientation). In the research by Bona et al they used a 1-9 vividness scale instead of a confidence rating, we chose to replace this by the more commonly used confidence ratings (Maniscalco & Lau, 2012; Malmberg, 2002), so that metacognition and metacognitive sensitivity could be acquired.
Due to the combination of: memory cue orientation, distracter presence, distracter difference relative to the memory cue and memory task response, a combination of 40 different trial types was constructed. The experiment consisted of 480 trials divided in six blocks of 80 trials with a 20 second break in-between blocks. In total the test consisted of: No distracter condition = 240 trials; 0° distracter condition = 80 trials; 10° distracter condition = 80 trials; 40° distracter condition = 80 trials. The 10°, 20°, 30°, 40°, and 50° left/right memory cue orientations were all evenly balanced over the different conditions.
Results Experiment 1
1: Accuracy, visibility and confidence outcomes
Figures 2A, 2B and 2C give an overview of the overall performance of the participants on the tasks. Figure 2A displays the relationship between WM accuracy and the confidence scores. There is a positive Pearson correlation between high WM performance and high confidence [Pearson’s r = 0.612, p= <0.01; 2-tailed]. Figure 2B displays the mean frequency of responses for each level of confidence rating. Figure 2C shows the proportions of responses for distracter visibility rating distributed as the ratings of distracter visibility.
Signal detection theory type 1 has been used to derive a measure of perceptual sensitivity. When participants reported a 1 on the distracter visibility response task it was interpreted as invisible and the responses 2,3 and 4 as distracter visible. A hit was defined as the condition in which the distracter was rated as visible (i.e. 2,3,4) when a distracter was presented. A false alarm was defined as the condition in which the distracter was rated as visible when there was
no distracter presented. This has been done for all the three orientations (0°, 10°, 40°). The mean d prime was 0.46 (SD = 0.38) but significantly different from chance level performance [t(92)=11.56, p=<0.001]. Individual d prime outcomes for all different orientations levels (0°, 10°, 40°) were compared with an ANOVA. There was no significant difference between the d prime outcomes for all different orientation levels.
2: Impact of distracter visibility and orientation on working memory accuracy
Figure 3A displays WM accuracy as a function of distracter visibility and orientation. A repeated measures ANOVA with only the distracter present trials, was used to assess the impact of distracter visibility and orientation on WM accuracy. Distracter visibility (visible, invisible) and orientation difference (0°, 10°, 40°) relative to the memory cue were used as main factors. This analysis revealed no significant effects of visibility [F(1,30)=<1], orientation [F(2,58)=<1] or an interaction between visibility and orientation [F(2,58)=<1] on WM accuracy.
Figure 2| (A) Mean WM accuracy as function of WM confidence (n=32). This figure displays the mean WM accuracy for all possible confidence ratings: 1=Very low, 2=Low, 3=High, 4=Very high. (B) The distribution of responses on the confidence scales (n=32). This figure indicates the proportion of responses for all confidence ratings. (C) Ratings of distracter visibility (n=32). Proportion of visibility responses for distracter absent and present trials.
* Error bars indicate ±1 SEM
A
C
Additionally, a separate set of t-tests to compare no distracter vs distracter conditions, resulted in no significant effects on WM accuracy for both visible and invisible distracters. In short, differences in distracter visibility/orientation had no significant effect on WM accuracy.
3: Impact of distracter visibility and orientation on subjective working memory confidence Figure 3B displays subjective WM confidence as a function of distracter visibility and orientation. A repeated measures ANOVA with only the distracter present trials, was used to assess the impact of distracter visibility and orientation on WM accuracy. Distracter visibility (visible, invisible) and orientation difference (0°, 10°, 40°) relative to the memory cue were used as main factors. This analysis revealed that there was a significant effect of distracter visibility on WM confidence [F(1,30)=42.96 ,p=<0.001; ηp2=0.585], with subjective confidence significantly lower in the invisible condition, and a significant effect of distracter orientation difference [F(2,58)=4.354, p=0.022; ηp2=0.231]. There was no significant effect of interaction between visibility and orientation [F(2,58)=1.00, p=0.38] on WM confidence.
Figure 3| (A) Objective WM accuracy as a function of the distracter orientation and visibility (n=32). This figure displays the mean WM accuracy for distracter visibility (No distracter, unaware of distracter, aware of distracter) and orientation (0°, 10°, 40) for distracter present and
distracter absent trials. There are no effects of distracter orientation and visibility on WM accuracy.
(B) Subjective WM confidence as a function of distracter orientation and visibility (n=32). This figure displays the mean confidence rating for distracter visibility (No distracter, unaware of distracter, aware of distracter) and orientation (0°, 10°, 40) for distracter present and distracter absent trials. Subjective WM confidence was reduced when participants reported the distracter as invisible, and participants were overall more confident when they reported the distracter as visible in contrast to the no distracter condition.
* Error bars indicate ±1 SEM
A
Additionally, a separate set of t-tests to compare no distracter vs distracter conditions, resulted in a significant difference between the no distracter and all invisible distracter conditions [0°: t(30)=6.35, p=<0.001, 10°: t(30)=3.39, p=0.002, 40°: t(30)=3.70, p=0.001)] but not for all of the visible distracter conditions [0°: t(30)=-1.84, p=0.075; 10°: t(30)=-4.23, p=<0.001; 40°: t(30)=-3.37, p=0.002]. In short, subjective WM confidence was reduced when participants reported the distracter as invisible, and participants were overall more confident when they reported the distracter as visible in contrast to the no distracter condition.
4: Impact of distracter visibility and orientation on metacognitive sensitivity
The relationship between confidence ratings and correct/incorrect WM responses was used as a measurement of metacognitive sensitivity (Table 1), using type-2 signal detection theory analyses, such as type 2 ROC (receiver operating characteristic) (Fleming & Lau, 2014) hit and false alarm rates can be plotted as individual points on the ROC plot (hit rate vertical, false alarm rate horizontal). The area under this curve (AUROC) indicates the metacognitive sensitivity. If the AUROC is 0.5 the performance is at chance. This measure of metacognition provides an unbiased, criterion-free index of how well observer’s confidence scores discriminate between their own correct and incorrect responses. In our study, there were 7 conditions: a No distracter condition, and a visible and invisible condition for all three conditions (0°, 10°, 40°). For all visibility and orientation conditions the AUROC was above chance level.
Table 1: SDT type 2 outcomes
Accuracy Confidence
Low High
Incorrect Type 2 correct rejection Type 2 false alarm
Correct Type 2 miss Type 2 hit
Figure 4 displays the mean metacognitive sensitivity (mean AUROC) for all visible and invisible distracter orientations (and the no distracter condition). A repeated measures ANOVA on all AUROC conditions determined that there was no significant effect of distracter visibility [F(1,30)=2.01, p=0.167], orientation [F(2,58)=1.95, p=0.16] or an interaction of both [F(2,58)=<1] on metacognitive sensitivity.
Discussion
The aim of the first experiment was to replicate the findings of Bona et al. (2013), with a similar experiment. The overall performance of the participants on the task was similar to Bona et al. (2013) (see Figure 2A,B,C / Bona et al. (2013)). There was a positive correlation between WM accuracy and confidence and the distributions of confidence and visibility scores were comparable. The signal detection type 1 analysis resulted in a d’ of 0.43 which was lower than the one in Bona et al. (2013), but still significantly different from zero. In our research, we did not find any significant effect of visibility, orientation or an interaction between visibility and orientation on WM accuracy, and no significant differences between the no distracter/distracter conditions. Concerning the impact of distracter visibility and orientation on confidence, there was a significant effect of visibility and orientation but no significant effect of the interaction of both. When evaluating the no distracter versus the distracter condition, WM confidence was reduced when participants reported the distracter as invisible, and participants were overall more confident when they reported the distracter as visible in contrast to the no distracter condition. At last a signal detection theory type 2 analysis was performed to investigate if there was an effect of visibility and orientation on the metacognitive sensitivity, this resulted in no significant outcomes. The positive correlation between WM accuracy and confidence confirms that confidence tracks WM accuracy
Figure 4| The mean metacognitive sensitivity as a function of distracter orientation and visibility (n=32). This figure displays the mean metacognitive sensitivity scores as a function of distracter visibility (No distracter, unaware of distracter, aware of distracter) and orientation (0°, 10°, 40°) for distracter present and distracter absent trials.
accurately (Figure 2A), but neither an invisible nor visible distracter can influence WM performance. However, confidence on WM performance is influenced by the introduction of a distracter.
The double dissociation that has been found in the experiment by Bona and Colleagues (2013), where confidence was reduced by introducing the visible and invisible distracters for all orientations (0°, 10°, 40°), and that WM accuracy for the visible and invisible 40° distracter is reduced, has not been replicated in our study, because no effects of the introduction of a distracter have been found. Therefore, we cannot conclude that working memory contents and their conscious experience can be dissociated. The signal detection type 2 analysis’s metacognitive sensitivity outcomes were not affected by the introduction of a distracter and indicate that the introduction of a distracter has no effect on how well observer’s confidence scores discriminate between their own correct and incorrect responses.
The overall similar outcomes in distributions of visibility and confidence scores indicate that we achieved replicating Bona et al. (2013) methodologically, but the results of our analysis differed in a lot of ways. In the first place the difference in mean d’ between Bona et al. (2013) (d’=1.52; SD=0.61) and our study (d’=0.43; SD=0.38). In our experiment the subjects performed worse on the task than the participants in Bona et al. (2013), suggesting that the introduced distracter was less visible in our study than in Bona et al. (2013). This could explain the difference in overall outcomes concerning the impact of orientation and visibility on WM accuracy and WM confidence, and explain why there is no general effect of distracter orientation on WM accuracy (Rademaker et al., 2015). Secondly in Bona et al. (2013) research they used a vividness scale from 1-9 (Baddeley & Andrade, 2001) instead of the confidence scale of 1-4 that we have used. This vividness scale is based on how vivid the memory item in memory was for which 1 means the absence of a mental image and 9 refers to as clear and vivid as visual perception of the grating. Our confidence ratings were more directed towards evaluating their own performance on the WM task then evaluating how vivid the image of the WM task was in their mental image. This difference in method could also have resulted in slightly different outcomes on the effect of the distracter on confidence. Finally, the number of participants in our study was twice as much as the number of participants in Bona et al. (2013). This difference indicates that the statistical power of the Bona et al. (2013) research was lower than in our experiment. This could have resulted in false positives due to type 2 errors (Cohen, 1992).
Because our research was not only about replicating Bona et al. (2013), but also about adding some new proof that could support the theory that WM contents and their conscious
experience can be dissociated, we designed a second experiment that could give more insights in the link between WM accuracy and confidence. The WM task that we used in the first experiment was the same as in Bona et al. (2013), but the question if the memory probe was tilted more to the left or to the right in contrast to the memory cue seemed to be a difficult task, and only resulted in a right/wrong possibility at which the participants had a 50% chance of guessing the right answer. for the second experiment, we created a more precise measure to investigate if the results that we’ve found depend on the way they were measured. The new WM accuracy measure that we will used is similar to the one used by Rademaker and colleagues (2015), which they used to determine the impact of interference on short-term memory for visual orientation. The Rademaker et al. (2015) research indicates that larger relative orientation differences between target and distracter lead to bigger performance decrements, compared with relative smaller differences. The effect of attention and awareness for the distracter has also been investigated and resulted in reduced effects of interfering information on memory performance when the distracting information was made task relevant like in our project. By using the adjustment-bar method (described below) we will investigate if this different approach of assessing working memory performance will result in different outcomes compared with Bona et al (2013) and our first experiment.
Materials and Methods Experiment 2
Participants
In total 25 students of the University of Amsterdam with normal or corrected to normal vision participated in the experiment. All the participants were naive to the aims of the study and were given an elaborate instruction and were only allowed to participate if they fully understood the task. 24 participants were included in the data analysis (9 males, mean age 23.33 and 15 females, mean age 21.4). One participant was excluded due to poor performance (mean error above 25°). All participants provided informed consent and were given 1.5 research credit or 15€ as reward for their participation.
Stimuli and Experimental procedure
The second experiment did not differ much from the first experiment (see Figure 5), except that the memory probe and left right response were replaced by a mouse probe consisting of a centrally presented white bull’s eye fixation (0.5° of visual angle in diameter) that was interrupted by a white line and divided the line in two parts (0.025° wide and 0.125° long, spaced 3° apart) (Rademaker et al., 2015). For the memory cue orientations, a set of
random orientations was picked, and were randomly presented in all experimental conditions. All other presentation screens and times were identical to the ones in the first experiment. By moving the mouse, the interrupted white line rotated around the fixation point. This was used to replicate the orientation of the memory cue that was kept in mind by method-of-adjustment. This method gives a working memory score in a range of degrees of error instead of a right/wrong accuracy outcome. The maximum error that could have been made was 45°.
The Experiment consisted of 432 trials divided in six blocks of 72 trials with a 10 second break in-between blocks. The test consisted of: No distracter condition = 108 trials; 0° distracter condition = 108 trials; 10° distracter condition = 108 trials; 40° distracter condition = 108 trials. The 10°, 20°, 30°, 40°, and 50° left/right memory cue orientations were all evenly balanced over the different conditions
Due to the measure used in the second experiment, WM error is measured instead of WM accuracy (Rademaker et al., 2015). Both are measuring WM performance for which WM accuracy can be interpreted as percentage right, and WM error (if converted) can be interpreted the same (0° error = accuracy of 1.0; 90° error = accuracy of 0.0). The smaller the WM error, the better WM accuracy was.
Figure 5| Timeline of experiment two. The orientation of the memory cue had to be kept in memory by the participants. At the end of each trial, participants were asked to replicate this orientation with an adjustment bar that could turn 360° by the use of a mouse, and click if they thought it was the same as the memory cue. After the left right response the participants had to respond how confident they were about the left right answer on a 1-4 scale (Confidence in response?; 1:Very low, 2:Low, 3:High, 4:Very high). In the period between the memory cue and memory probe in 50% of the trials a distracter was presented (grating with the same size, contrast and spatial frequency as memory cue/probe) this distracter could be identical to the memory cue, or could differ 10°/40° . after the left right response and the confidence rating, participants were asked how visible the distracter was (Awareness of distracter; 1:Did not see the distracter, 2:Maybe saw something, 3:Saw the distracter but not its orientation, 4:Saw the distracter and its orientation).
Results Experiment 2
1: Accuracy, visibility and confidence outcomes
As for the first experiment, we plotted the overall outcomes and performance distributions of the second experiment in figures 5A, 5B and 5C. Because we used a continuous gradient scale in the second experiment instead of a right/wrong possibility outcome, accuracy for the second experiment is now distributed as the mean degrees of error with a minimum of 0° (adjustment bar set exactly as memory cue) and a maximum of 90°(adjustment bar set perpendicular on memory cue orientation). Due to the measure used in the second experiment, WM error is measured instead of WM accuracy. In figure 6A is the mean degrees of error distributed for all four confidence ratings (Confidence in response; 1:Very low, 2:Low, 3:High, 4:Very high). There was a negative correlation between WM error and high confidence [Pearson’s r = -0.407, p = 0.048; 2-tailed]. Better performance (low error) resulted in higher confidence and worse performance (high error) resulted in low confidence. Figure 6B displays the distribution of responses on the confidence scale for the second experiment, and indicates that participants in the second experiment were more confident in comparison with the first experiment. Figure 6C displays the ratings of distracter visibility in proportion of responses for all conditions. Like in the first experiment, signal detection theory type 1 has been used to derive a measure of perceptual sensitivity. Hit and false alarm rates were derived in the same way as described in the first experiment. The mean d’ score was 0.70 (SD = 0.56) and was significantly different from chance level performance [t(73)=10.86, p=<0.001]. Individual d’ prime outcomes for all different orientations levels (0°, 10°, 40°) were compared with an ANOVA, which resulted in no significant differences.
2: Impact of distracter visibility and orientation on working memory accuracy
Figure 7A displays WM error in degrees as a function of distracter visibility and orientation. Like in experiment one a repeated measures ANOVA was executed with only the distracter present trials. This was used to assess the impact of distracter visibility and
orientation on WM accuracy. The repeated measures ANOVA revealed that there was a significant effect of distracter orientation on WM error [F(2,44)=6.52, p=0.006; ηp2=0.372]. Pair wise comparisons on the factor orientation difference, resulted in a significant difference between the 40° vs 0° condition [t(23)=1.51, p=0.004; Bonferroni corrected]. The 40° vs 10° and 0° vs 10° conditions were not significantly different. However, neither a significant effect of visibility [F(1,24)=3.07, p=0.093] (marginally significant; invisible distracter condition leans towards higher error) and interaction between visibility and orientation [F(2,44)=1.40, p=0.27] on WM error has been found.
A B
C Figure 6| (A) Mean WM error as function of
WM confidence (n=24). This figure displays the mean WM accuracy for all possible confidence ratings: 1=Very low, 2=Low, 3=High, 4=Very high. As expected there was a significant negative correlation between WM error and WM confidence. (B) The distribution of responses on the confidence scales (n=24). This figure indicates the proportion of responses for all confidence ratings. (C) Ratings of distracter visibility (n=24). Proportion of visibility responses for distracter absent and present trials.
Additionally, a separate set of t-tests to compare no distracter vs distracter conditions, resulted in a significant difference between the no distracter and the 40° invisible distracter condition [t(23)=2.71, p=0.033] but not for any of the visible distracter conditions. In short, subjective WM error was higher when participants reported the distracter as invisible in the 40° conditions.
3: Impact of distracter visibility and orientation on subjective working memory confidence Figure 7B displays subjective WM confidence as a function of distracter visibility and orientation. A repeated measures ANOVA with only the distracter present trials, was used to assess the impact of distracter visibility and orientation on WM accuracy. Distracter visibility (visible, invisible) and orientation difference (0°, 10°, 40°) relative to the memory cue were used as main factors. This analysis revealed that there was a significant effect of distracter visibility on WM confidence [F(1,23)=5.02, p=0.035; ηp2=0.179], with subjective confidence significantly lower in the invisible condition. No significant effects of distracter orientation [F(2,44)=<1] and interaction between visibility and orientation [F(2,44)=<1] have been found.
Figure 7| (A) Objective WM error as a function of the distracter orientation and visibility (n=24). This figure displays the mean WM error for distracter visibility (No distracter, unaware of distracter, aware of distracter) and orientation (0°, 10°, 40°) for distracter present and distracter absent trials.
(B) Subjective WM confidence as a function of distracter orientation and visibility (n=24). This figure displays the mean confidence rating for distracter visibility (No distracter, unaware of distracter, aware of distracter) and orientation (0°, 10°, 40°) for distracter present and distracter absent trials. * Error bars indicate ±1 SEM
A
Additionally, a separate set of t-tests to compare no distracter vs distracter conditions, resulted in a significant difference between the no distracter two out of three invisible distracter conditions [0°: t(23)=2.26, p=0.034; 40°: t(23)=2.51, p=0.020] but not for the visible distracter conditions. In short, subjective WM confidence was reduced when participants reported the distracter as invisible for the 0° and 40° conditions, and for the visible distracter condition no effects have been found.
4: Impact of distracter visibility and orientation on metacognitive sensitivity
The same type of SDT-2 analysis as in experiment one has been performed. For all visibility and orientation conditions the AUROC was above chance level. Figure 8 displays the mean metacognitive sensitivity (mean AUROC) for all visible and invisible distracter orientations (and the no distracter condition A repeated measures ANOVA on all AUROC conditions determined that there was no significant effect of distracter visibility [F(1,23)=<1], orientation [F(2,44)=<1] or an interaction of both [F(2,44)=1.67, p=0.21] on metacognitive sensitivity
Discussion
The aim of the second experiment was like in the first experiment to investigate if metacognitive judgements track working memory performance accurately (even in the presence of invisible distracting information), if metacognition can be dissociated from
Figure 8| The mean metacognitive sensitivity as a function of distracter orientation and visibility (n=24). This figure displays the mean metacognitive sensitivity scores as a function of distracter visibility (No distracter, unaware of distracter, aware of distracter) and orientation (0°, 10°, 40°) for distracter present and distracter absent trials.
objective working memory accuracy and if the outcomes are method specific (in contract to experiment one and Bona et al. (2013)).
Performance in the second experiment was similar to experiment one and Bona et al. (2013) (see Figure 6A,C / (Bona et al., 2013) ). There is a negative correlation between WM error and high confidence, which is the same as a positive correlation between WM accuracy and high confidence. The signal detection type 1 analysis resulted in a d’ of 0.70 which was lower than the one in Bona et al. (2013), but higher in comparison with experiment one and still significantly different from zero. A notable difference in comparison with experiment one and Bona et al. (2013) generally scored higher on confidence for the task (Figure 6B). In contrast with the first experiment, an effect of orientation on WM error has been found, and WM error increased when a distracter was introduced. The 40° invisible distracter seems to contribute the most in this effect, but is only significant compared to the influence of the 0° distracter on the overall effect. An increase in WM error has been found for the 40° invisible distracter in comparison with the no distracter condition. Additionally, there was a significant effect of visibility on confidence, with lower confidence scores for invisible distracters, and confidence on WM performance is reduced when the 40° invisible distracter is introduced. Likewise, in the first experiment, the introduction of a distracter resulted in no effects on metacognitive sensitivity.
The negative correlation between WM error and confidence (Figure 6A) confirms that confidence tracks WM accuracy accurately, and that in case of this WM task (with the adjustment bar) the distracter can influence WM performance negatively. The no distracter vs distracter outcomes suggest that the effect of the distracter is the largest for the 40° invisible distracter. In short, these results suggest that that the introduction of a 40° distracter causes a drop in WM accuracy. The introduction of 40° distracter not only influenced WM accuracy but also confidence, which is not in line with the hypothesis that the link between WM and confidence can be broken by the introduction of a distracter. Metacognitive sensitivity is like in the first experiment not affected by the introduction of a distracter and indicates that the introduction of a distracter has no effect on how well observer’s confidence scores discriminate between their own correct and incorrect responses.
Like in our first experiment the signal detection type one analysis resulted in a lower d’ compared with Bona et al. (2013), but higher than in our first experiment, which suggests that the participants in the second experiment performed worse on the WM accuracy than the ones in Bona et al. (2013), but better in the second experiment compared with the first. An
explanation for this difference in comparison with Bona et al. (2013) can be that the distracter is still less visible, but cannot explain the difference in performance compared with the first experiment (conditions were completely identical). The difference between performance in experiment one and two can be explained by using a different method to determine WM performance. The overall difference in confidence can be explained by the fact that participants in the second experiment were evaluating the action of adjusting the adjustment-bar instead of evaluating a right/wrong answer like in experiment one. This indicates that there are differences in metacognitive preference for different kind of tasks and may have resulted in response bias. The influence of the distracter on WM performance is not completely in line with the previous research of Rademaker and colleagues (2015). The amount of orientation difference in contrast to the memory cue, resulted in larger negative effects on WM performance which was similar, but in contrast to Rademaker. (2015) awareness of the distracter was not necessary to influence the representation of a grating in memory. An explanation for this difference in effect of the distracter orientation on WM accuracy could be explained by the fact that in the Rademaker. (2015) experiment binocular rivalry was used to measure the effect of the distracter and we used backward masking. These differences in tasks could explain the different outcomes because the binocular rivalry method is a more specific measure of assessing the influence of the unconscious processing of visual information. The differences in results of the first and second experiment suggest that the outcomes in this line of research are method specific.
Conclusion
Based on this research we can conclude that the link between working memory and consciousness (assessed through metacognition) cannot be dissociated when raw confidence scores are used. The theory that “working memory contents and their conscious experience can be fully dissociated, such that consciousness is always based only on a copy of working memory contents, and working memory contents are never consciously experienced directly” (Jacobs & Silvano, 2015; Bona et al, 2013) seems to be not a theory that can be generalized when raw confidence scores are used as a measure of confidence in contrast to vividness. Both confidence and vividness are assumed to measure metacognition, but result in different outcomes on a similar task. When reappraising the other theories, our findings of differences in effect of the distracter on metacognition and working memory tend more to the theory that consciousness is richer than working memory, meaning that some conscious contents are
residing outside working memory (Block, 2011; Lamme, 2003). This is only a speculation and depends on the assumption that the raw confidence scores measure consciousness.
Besides, the difference in confidence outcome between the first and second experiment, our results might be even more interesting regarding the “consciousness is richer than working memory” theory because of the difference in metacognitive evaluating between different types of tasks and relevance of methods used for different outcomes. The difference in metacognitive evaluating for different tasks might be an interesting subject for future studies and can possibly give more insights in the difference between findings of multiple researches including ours. Because this study illustrates that the methods used to evaluate the link between working memory and consciousness can influence its outcomes further research should try to figure out if these effects are method specific or that different psychological processes could underlie these differences in effects.
Another conclusion we can make is that even when confidence and WM performance on a task are influenced by the introduction of a distracter, the ability to discriminate between your own correct and incorrect responses (metacognitive sensitivity) is not affected.
We propose that working memory and consciousness may overlap on some parts, but that a part of consciousness works past the working memory boundaries. We suggest that metacognition and its properties could give more insights in the relationship between working memory and consciousness and therefore should be considered as one of the key factors in this line of research. However, the methods of measuring metacognition and the effects of different tasks on confidence ratings have to be further evaluated.
References
Awh, E., Vogel, E. K., & Oh, S. H. (2006). Interactions between attention and working memory. Neuroscience, 139(1), 201-208.
Baars, B. J., & Franklin, S. (2003). How conscious experience and working memory interact. Trends in cognitive
sciences, 7(4), 166-172.
Baddeley, A. (1992). Working Memory and Conscious Awareness. Theories of memory, 11-20.
Baddeley, A. D., and Andrade, J. (2001). Working memory and the vividness of imagery. J. Exp. Psychol. Gen. 129, 126–145.
Block, N. (2011). Perceptual consciousness overflows cognitive access. Trends in cognitive sciences, 15(12), 567-575.
Bona, S., Cattaneo, Z., Vecchi, T., Soto, D., & Silvanto, J. (2013). Metacognition of visual short-term memory: dissociation between objective and subjective components of VSTM. Frontiers in psychology, 4, 62.
Bona, S., & Silvanto, J. (2014). Accuracy and confidence of visual short-term memory do not go hand-in-hand: behavioral and neural dissociations. PloS one, 9(3), e90808.
Borkowski, J. G., Carr, M., & Pressley, M. (1987). “Spontaneous” strategy use: Perspectives from metacognitive theory. Intelligence, 11(1), 61-75.
Cohen, J. (1992). Statistical power analysis. Current directions in psychological science, 1(3), 98-101.
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.
Cowan, N. (2008). What are the differences between long-term, short-term, and working memory?. Progress in
brain research, 169, 323-338.
Daseking, M., Melzer, J., Rißling, J., & Petermann, F. (2015). Relationship between Intelligence and Executive Function. Gesundheitswesen (Bundesverband der Arzte des Offentlichen Gesundheitsdienstes (Germany)),
77(10), 814-819.
Fleming, S. M., Weil, R., Nagy, Z., Dolan, R. J., & Rees, G. (2010). Relating introspective accuracy to individual differences in brain structure. Science, 329, 1541–1543
Fleming, S. M., & Lau, H. C. (2014). How to measure metacognition. Frontiers in Human Neuroscience, 8, 443. Lamme, V. A. (2003). Why visual attention and awareness are different. Trends in cognitive sciences, 7(1), 12-18.
Malmberg, K. J. (2002). On the form of ROCs constructed from confidence ratings. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 28(2), 380.
Maniscalco, B., & Lau, H. (2012). A signal detection theoretic approach for estimating metacognitive sensitivity from confidence ratings. Consciousness and cognition, 21(1), 422-430.
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. (2009). Interference between storage and processing in working memory: Feature overwriting, not similarity-based competition. Memory & Cognition, 37(3), 346-357.
Rademaker, R. L., Tredway, C. H., & Tong, F. (2012). Introspective judgments predict the precision and likelihood of successful maintenance of visual working memory. Journal of Vision, 12(13), 21-21.
Rice, M. E., & Harris, G. T. (2005). Comparing effect sizes in follow-up studies: ROC Area, Cohen's d, and r.
Law and human behavior, 29(5), 615.
Richet, C. (1886). Les origines & Les modalités de la mémoire: Essai de psychologie générale. Revue
Philosophique de la France et de l'Étranger, 21, 561-590.
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.
Stein, T., & Sterzer, P. (2014). Unconscious processing under interocular suppression: getting the right measure.
Frontiers in Psychology, 5, 387.
Vandenbroucke, A. R., Sligte, I. G., Barrett, A. B., Seth, A. K., Fahrenfort, J. J., & Lamme, V. A. (2014). Accurate metacognition for visual sensory memory representations. Psychological science, 25(4), 861-873.
Software
We wrote our experiments in Matlab, using the Psychophysics Toolbox extensions (Brainard, 1997; Pelli, 1997; Kleiner et al, 2007).
MATLAB and Statistics Toolbox Release 2016b, The MathWorks, Inc., Natick, Massachusetts, United States. Guide, M. U. S. (1998). The mathworks. Inc., Natick, MA, 5, 333.
