BACHELORPROJECT BREIN & COGNITIE 2017 |BEGELEIDER: TIMO STEIN
Working memory:
unconscious or not?
JIKKE VAN DEN ENDE, 10787593
1
Abstract
This current study is based on a research of Soto and colleagues in 2011, which states that working memory can operate on unconscious input. They found above-chance performance on unaware cue-present trials in a delayed cue-target discrimination task. Hence, it could be possible that the effect they found wasn’t caused by unconscious memory, but by other perceptual processes. To test this, current study used a slightly different form of the main experiment of Soto and colleagues, and a perception task. Replicating the results of Soto and colleagues has failed, but a higher cue sensitivity and performance on the perception task was found, in comparison to the delayed cue-target discrimination task. These findings indicate that a weakly conscious guess is held in the working memory, which causes the above-chance performance. But, because no significant above-chance performance is found in this study, further research is required to determine if unconscious memory exists or not.
2
Although there consist a lot of theories about working memory (WM), it’s still not an exactly understood concept. Classically, it is described as the short-term storage and
transformation of information that is not currently present in the environment (Stein, Kaiser & Hesselmann, 2016). It plays an important role in, for example, reasoning, language
comprehension and learning (Baddeley & Hitch, 1974). The most widely used model for WM is the model of Baddeley & Hitch (1974). This model used to consist of three different
components, but a forth component has been added (Baddeley, 2000). First, there is the phonological loop, which holds speech-like representations (Morris & Jones, 1990). The second component is the visuospatial sketchpad, which holds all the imaginary
representations (Hanley, Young & Pearson, 1991). The third component is the episodic buffer, which combines the information from the phonological loop and the visuospatial sketchpad and information from the long-term memory into a so-called episode: a time-period of information (Baddeley, 2000). Lastly, the controlling component for WM is the central executive system (CES). The CES monitors the continuous flow of information to and from different short-term buffers (D’Esposito, Detre, Alsop, Shin, Atlas & Grossman, 1995). It directs the attention towards extern and intern information, and it coordinates the three other components of WM (Morris & Jones, 1990). The model of Baddeley & Hitch (1974) is based on the assumption that WM operates on consciously represented stimuli (Soto,
Mäntylä & Silvanto, 2011). Hence, about the link between consciousness and WM is still a lot unknown.
Consciousness is a difficult concept. In the cognitive neuroscience, there consist a lot of different representative models, each trying to explain how aspects of consciousness can be found in the brain (Atkinson, Thomas & Cleeremans, 2000). Hence, the search for a neural correlation between some brain-areas and consciousness is still ongoing. Therefore, the relation between consciousness and cognition, and how these two interact, is complex and hard to measure. The global workspace (GB) theory, worked out by Baars (1988), tries to connect cognition and consciousness. When perceptual information enters the brain, it’s first processed by small, localised brain areas (Baars, 2005). Then, all these signals will be broadcast to a wide neural network all around the cortex, which Baars (1988) calls the global workspace. It predicts different conscious aspects of all these different signals, and so we become conscious of the things around us (Baars, 2005).
3
There are different views on the relationship between WM and consciousness. First, there is the classical view that states that WM is the same as consciousness and
consciousness is the same as WM: WM operates on conscious input and we are conscious of the contents and operations of WM (Soto & Silvanto, 2014; Baars & Franklin, 2003;
Baddeley, 2003). This view is based on the fact that active components of WM are accurately reportable, and accurate report is also the standard operational index of consciousness. Secondly, there are also theories saying that more things than WM content can be conscious, but that WM is always conscious (Lamme, 2006; Block, 2007). This view states that phenomenally aware information may become consciously accessible through WM; before this information is conscious, it may be held in short-term memory. Thirdly, according to the model of WM of Cowan (1988) and Oberauer (2002), not all WM contents are
consciously accessibly: if you want to be aware of something, you must focus on it with internal attention. Thus, items can be in WM, and remain unconscious until you attend to it. Lastly, some models disassociate WM content and conscious experience completely.
Looking at the third view on WM content and conscious experience, there is an even more radical vision by Soto & Silvanto (2014). They state that even if some items in WM are attended, they still can remain unconscious. This means that the input of WM can be
unconscious, it can be unconsciously maintained and later be used for retrieval. They got this conclusion from their research in 2011, which was based on research of Rosenthal, Kennard & Soto (2010) and Levy, Goldstein, Mandel, Maril & Hassin (2012) that both had found evidence that working operations may occur independently of conscious awareness of the critical information. Soto, Mäntyla and Silvanto (2011) tested whether WM content can indeed be unconscious with a delayed cue-target discrimination task. Participants were presented with a brief masked subliminal orientation cue (the memory cue). The mask was a black circle, the same size as the cue and showed on the exact same place as the memory cue. After the masked memory cue, there was a short delay, and then a test-cue appeared. The participants had to determine if the test-cue was turned left or right in comparison to the memory cue. Afterwards, they had to give a subjective rating on how aware they were of the memory cue. The experiment used an awareness-scale of 1-4, in which 1 meant ‘Did not see anything’; 2, ‘Maybe saw something’; 3, ‘Saw the stimulus but not the orientation’ and 4, ‘Saw the stimulus and its orientation’. The participants were told to attend and hold the
4
memory cue in memory even if they could not consciously perceive it and perform explicit cue-target orientation discrimination. In 50% of the trials the memory cue was absent. In the analysis of the results only the cue-present-trials in which the participants were fully
unaware of the cue (they rated ‘1’ on the awareness-scale) were included. The results showed that discrimination performance on unaware cue-present trials was above chance. Soto et al. (2011) concluded that WM could also operate on unconscious represented stimuli.
This research was almost completely replicated by Dutta, Shah, Silvanto and Soto (2014). They also included an fMRI and tDCS study, to determine if there was a correlation between awareness of the stimuli and some brain-areas, namely the PFC. They didn’t found any interesting correlations, so there isn’t any information about unconscious WM in the brain. They did find the same results about discrimination performance on unaware trials, which was above chance.
There are some things about this researches of Soto et al. (2011) and Dutta et al. (2014) and their view on WM content and conscious experience that aren’t correct. For instance, in both articles the researchers did a psychophysical analysis of perceptual sensitivity, to measure the sensitivity the subjects had on the cue. They found pour
sensitivity, and concluded that memory discrimination could be dissociated from perceptual awareness. They both state they used the Signal Detection Theory (SDT) for this analysis, but according to Stein et al. (2016) they calculated cue sensitivity (d’) wrong. They only used the hits and false alarms to derive a measure of d’, but they didn’t include the correct rejections and misses (the trials where people rated they were aware of the memory cue, when the memory cue was present or absent respectively). For measuring d’ correctly, the misses and correct rejections should be concluded in the calculation. For example, Soto et al. (2011) and Dutta et al. (2014) calculated the hit-rate as the number of hits divided by the number of hits plus the number of false alarms. According to SDT you should calculate the hit-rate as the number of hits divided by the number of hits plus the number of misses. So, Stein et al. (2016) argue that both researches came to the incorrect conclusion that memory discrimination can be dissociated from perceptual awareness.
In a response on Stein et al. (2016), Soto & Silvanto (2016) say they agree with Stein et al. (2016) that the calculated d’ in their prior studies wasn’t a pure sensitivity measure.
5
They did a reanalysis of the data from Soto et al. (2011) with the bias-free sensitivity
approach proposed by Stein et al. (2016), but they again didn’t find the correlation between perceptual sensitivity and WM performance. In this research, this correct analysis of
sensitivity will also be performed.
Secondly, Stein et al. (2016) note that the findings of Soto et al. (2011) could also be accounted by perceptual processes. Participants can be unconscious or weakly conscious of the perceptual representations of the masked memory cue, that makes a conscious guess which is held in WM during the delay period. This would mean that the above-chance performance on the unaware cue-present trials in the research of Soto et al. (2011) wasn’t caused by unconscious WM, but by weakly conscious WM. To test this hypothesis, a perception task should be included in this research. This perception task shouldn’t contain the test cue, but the task for the subjects should only be to determine if the memory cue is left or right orientated. This perception task is less complex than the delayed cue-target discrimination task. If the assumption proposed by Stein et al. (2016) is true and because of the shorter delay between the memory cue and the test question, the weakly conscious guess held in WM during the perception task is stronger because there’s less time for it to go into decline. If cue visibility is higher on the perception task, the effect found by Soto et al. (2011) could be assigned to other perceptual processes instead of unconscious WM.
Thirdly, the awareness scale Soto et al. (2011) and Dutta et al. (2014) used in their experiments isn’t a very good subjective measure of awareness. They said they used a version of The Perceptual Awareness Scale (PAS). The PAS usually is a four-scale measure, covering: ‘no experience’, ‘vague experience’, ‘almost clear experience’ and ‘clear
experience’ (Overgaard, Timmermans, Sandberg & Cleeremans, 2010). Hence, the awareness scale Soto et al. (2011) used didn’t contain the term experience. Instead they used: ‘did not see anything’, ‘maybe saw something’, ‘saw the stimulus but not the orientation’ and ‘saw the stimulus and its orientation’. With this scale, you ask about the visibility of the stimulus, not about the experience of the stimulus. Moreover, their scale did not have the option ‘saw the orientation, but not the stimulus’ or in a better way ‘almost clear experience of the orientation’. It could be possible that the orientation was clear for awareness, but the total stimulus was not. Concluding, the awareness-scale Soto et al.
6
(2011) used wasn’t complete enough. Therefore, in this research the original PAS will be used, to see if the same results could be obtained with a more complete scale.
Finally, the mask Soto et al. (2011) used in their experiment may not be very strong. Because of the strong contrast between the white stripes and the black circle, sometimes the orientation of the memory cue could still be seen right after the mask appeared. This is because of an optical illusion, when two opposite colours (black and white) follow directly after each other (Li & Guo, 1995). It’s important that subjects are unaware of the memory cue, but when the orientation can be seen for a longer time, this might cause much more aware trials. Therefore, current study will use another mask. This mask will be a circular one, consisting of botch white and black colours. The assumption is that the contrast will be less strong, so that the mask completely covers the memory cue.
Concluding, this research will try to replicate the findings of Soto et al. (2011), but then including the PAS instead of the awareness-scale Soto et al. (2011) used. Also, a
different mask will be used. Most importantly, with the extra perception task it is possible to see if cue visibility is higher than in the delayed cue-target discrimination task. That would mean that it isn’t unconscious WM that is causing the effect Soto et al. (2011) found, a weakly conscious guess that is held in WM during the delay.
7
Figure 2. One trial of the perception task.
Box 1.
Delayed cue-target discrimination task Participants were presented with a brief (16.7 ms)
and masked subliminal orientation cue (Figure 1). After a short delay (2 seconds) a test cue appeared, with a different orientation than the memory cue. The participants had to
determine if the test cue was turned left or right in comparison to the memory cue, with the left or right key. Afterwards, they had to rate their awareness on the PAS (1 = no experience, 2 = vague experience, 3 = almost clear experience, 4 = clear experience). In 50% of the trials, the memory cue was absent.
Perception task Participants were presented with a brief (16.7 ms) and masked subliminal
orientation cue (Figure 2). After a short delay (200 ms) the participants had to determine if the memory cue was left or right orientated, with the left or right key. Afterwards, they had to rate their awareness on the PAS. Again, in 50% of the trials the cue was absent.
Fifteen participants were included in this experiment. There were nine women and six men, and the mean age was 21,3. Before the experiment, the participants were explicitly told to attend and hold the memory cue in memory even if they could not consciously perceive it and perform explicit cue-target orientation discrimination. They were also informed with the fact that in 50% of the trials the memory cue was absent, in both tests they would perform. First, they did one block of the delayed cue-target discrimination task (Figure 1). This block consisted of 144 trials. There was a short break after 72 trials. After that, they did one block of the perception task (Figure 2), which also consisted of 144 trials, again including a break. Then they had to perform both tasks again, so in total all the
8
participants performed four blocks (both tests twice). The orientation of the memory cue and the test cue was random, so the blocks were never the same.
Figure 3. Results of the repeated-measures ANOVA to test the accuracy on the cue-present trials depending on
task and awareness.
First of all, a repeated-measures ANOVA was used to test the discrimination and determination accuracy on the cue-present trials depending on the awareness; 1 on PAS or 2,3 or 4 on PAS, and on the task; delayed cue-target discrimination task or perception task (Figure 3). The assumptions for this repeated-measures ANOVA were assumed. A main effect was found for awareness (F(1, 12) = 7.918; p = 0.16). This means that the accuracy was higher when the subjects were aware of the memory cue. Also, a main effect for task was found (F(1, 12) = 67.418; p < 0.001). Overall performance was higher on the perception task: for both the unaware and aware trials there was a higher accuracy then in the delayed cue-target discrimination task. No interaction effect between awareness and task was found (F(1, 12) = 1.060; p = 0.324).
9
The same analysis Soto et al. (2011) did was performed in this study. Two
independent t-tests were used to test the difference between the performance on both tests on the unaware cue-present trials, and the level of chance (50%). The assumptions for this test were assumed. The performance on the unaware cue-present trials on the delayed cue-target discrimination task wasn’t significantly above chance (t(14) = 2.07; p = 0.058). Hence, the performance on the perception task on the unaware cue-present trials was significantly above chance (t(14) = 5.02; p < 0.001).
Another repeated-measures ANOVA was used to test the mean visibility ratings depending on the cue presence; present or absent, and on the task; the delayed cue-target discrimination task or perception task. The assumptions for this repeated-measures ANOVA were assumed. A main effect was found for presence (F(1, 14) = 14.211; p = 0.002). This effect shows that when the cue was present, the mean visibility rating was higher in comparison to the trials where the cue was absent. No main effect was found for task (F(1, 14) = 0.681; p = 0.423), and also no interaction effect (F(1, 14) = 1.094; p = 0.313).
Figure 4. The proportion of the pressed subjective rating on the PAS in the Delayed cue-target discrimination
task, depending on the memory cue being absent or present.
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 1 2 3 4 Pr opo rt io n Rating on PAS
Delayed cue-target discrimination task
10
Figure 5. The proportion of the pressed subjective rating on the PAS in the Perception task, depending on the
memory cue being absent or present.
Continuing on this, a manipulation check was done to test whether the subjects performed the two tasks correctly. It’s necessary to check if they completed the test seriously. One way to verify this, is looking at the proportions of each of the possible subjective ratings on the PAS (1-4) for both tests, depending on the memory cue being absent or present (Figure 4; Figure 5). The expectation is that the subjects will press significantly more 1 on the PAS when the memory cue is absent, instead of when it’s present. That would indicate that they were concentrated during the test, because they sometimes could detect the absent trials. The proportion of the rating 1 on the cue-present trials on the delayed cue-target discrimination task differs significantly from the proportion of the rating 1 on the cue-absent trials (t(14) = -3.6; p = 0.003). This is also the case for the proportions of rating 2 and 3 (t(14) = 3.01; p = 0.009; t(14) = 2.7; p = 0.018). Only for the proportion of the rating 4 there isn’t a significant difference (t(14) = 1.19; p = 0.255). Hence, the proportions of the rating 4 were extremely low (in both aware and unaware trials less than 1%). The proportion of the rating 1 on the cue-present trials on the perception task differs significantly from the proportion of the rating 1 on the cueabsent trials (t(14) = -3.475; p = 0.004). This is also the case for the proportions of rating 2 and 3 (t(14) = 2.676; p = 0.018; t(14) = 2.231; p = 0.043). Again, only for the proportion of the rating 4 there isn’t a
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 2 3 4 Pr opo rt io n Rating on PAS
Perception task
11
significant difference (t(14) = 1.585; p = 0.135). Hence, the proportions of the rating 4 were extremely low (in both aware and unaware trials less than 1%).
Finally, an analysis of cue sensitivity was also calculated in this research. An independent t-test was used to compare the calculated d’ of the delayed cue-target
discrimination task and the perception task. No difference was found between the two (t(14) = -1.650; p = 0.121).
Concluding, this research failed to replicate the findings of Soto et al. (2011). The performance on the unaware cue-present trials wasn’t significantly above chance. More importantly, a higher accuracy performance was found in the perception task in comparison with the delayed target discrimination task. Also, higher cue sensitivity on the cue-present trials was found, which leads to the conclusion that it could be that the above-chance performance isn’t caused by unconscious WM, but by other perceptual processes. Weakly conscious awareness of the memory cue could have led to a conscious guess, which was hold in WM during the delay period in the cue-target discrimination task. For the subjects it feels as a guess, but it leads in more than 50% of the cue-present trials to a correct answer. The performance on the perception task on the unaware cue-present trials in comparison with the delayed cue-target discrimination task could have been higher, because of the guess that was stronger because the task was less complex and the delay was shorter. The guess was kept in WM for a shorter time then in the delayed cue-target
discrimination task, where it became weaker. In Hence, because this current study couldn’t replicate the findings of Soto et al. (2011), further research is required.
One important ant critical thing about this current study is that the failure of
replicating Soto et al. (2011) depends on the interpretation of the test. The p-value that was found (p = 0.058) isn’t significant, but because it was tested two-sided, and beforehand the direction of the searched effect was clear, it’s valid to divide this value by two. Than the value would have been 0.029, and that is significant. It’s possible to state that current study did replicate Soto et al. (2011), and taken this into account, further research can continue with this experimental design. But there are also some aspects of current study that could be improved.
12
It is hard to measure consciousness, and the results of such a test are never fully reliable. Even Overgaard et al. (2010) state in their article that there are several problems with using the PAS for measuring consciousness. Their conclusion is that there is enough reason to believe that confidence ratings may not be the best way to gain knowledge about conscious experience, and that participants tend to guess. Overgaard et al. (2010) suggests using wagering, because one is interested in whether people know on what knowledge they based their decision, without the experimenter giving structural knowledge. If you gave them a risk-factor, to win money when they made the correct decision (saw something, when there was something) or lose when incorrect, you could get more reliable results than when using subjective consciousness ratings.
Furthermore, it is known that the WM capacity is limited. Since Soto et al. (2011) and Dutta et al. (2014) propose to have found unconscious WM, it should be interesting to include a dual WM task in their original experiment. Just before the masked memory cue a second WM task could be included, that would also use some of the capacity of WM. If the results show that the information of the two WM tasks interfere, these two tasks depend on the same WM content. Knowing that, it is possible to determine if the delayed cue-target discrimination task was using unconscious WM content. When the two tasks interfere, the delayed cue-target discrimination task from Soto et al. (2011) wasn’t based on unconscious WM, since unconscious WM should have its own capacity. If it had his own capacity, it should not interfere with a conscious WM task.
Also, there isn’t found a significant difference between the performance on the delayed cue-target discrimination task for the target-present trials when the memory cue was invisible, and a value of 0,5. According to that, the participants scored at chance level on these trials, so this research failed to replicate the results of Soto et al. (2011). It could have been that the cue was too invisible. There are all kind of circumstances that could have been affected the visibility of the cue, for instance the luminance level, or the presenting-time. Hence, this was the same in the experiment of Soto et al. (2011). One possible explanation is that the mask could have been too strong. We changed the mask of Soto et al. (2011) into a circular mask. Because of the round orientation, it possibly could have blocked the memory cue, which made it harder to see. In the perception task, there was a significant difference between 0,5 and the performance on the target-present trials when the memory cue was invisible. Maybe because the time between the awareness rating and the memory cue was
13
shorter in the perception task, although the mask blocked the visibility of the memory cue very strong, it was still easier to see and ‘remember’ it than in the delayed cue-target discrimination task.
Finally, it would be interesting to determine if the same effect can be realised when using a different stimuli or different masks. The contrast and luminance of the trials can differ between trials, and can make the memory cue more or less visible. But it is also interesting to use stimuli that are hard to verbalize and that are less likely to be maintained through a ‘conscious guess’. The visibility of the memory cue also depends on the kind of mask that follows. When using a mask, with a striped pattern instead of the black mask Soto et al. (2011) used, this could make the memory cue less visible because the orientation will be partly overlapped by the mask.
14
References
Atkinson, A.P., Thomas, M.S.C. & Cleeremans, A. (2000). Consciousness: mapping the theoretical landscape. Trends in Cognitive Sciences, 4(10), 372-382.
Baars, B.J. & Franklin, S. (2003). How conscious experience and WM interact. TRENDS in
Cognitive Neuroscience, 7(4), 166-172.
Baddeley, A. (1992). WM and Conscious Awareness. Theories of memory, 11-20.
Baddeley, A. (2003). WM: looking back and looking forward. Nature Reviews Neuroscience,
4, 929-839.
Baddeley, A.D. & Hitch, G. (1974). WM. Psychology of Learning and Motivation, 8, 47-89. Block, N. (2007). Consciousness, accessibility, and the mesh between psychology and neuroscience. Behavioural Brain Science, 30, 481-499.
Cowan, N. (1988). Evolving conceptions of memory storage, selective attention, and their mutual constraints within the human information-processing system. Psychological Bulletin,
104, 163-191.
Dutta, A., Shah, K., Silvanto, J. & Soto, D. (2014). Neural basis of non-conscious visual WM.
NeuroImage, 91, 336-343.
D’Esposito, M., Detre, J.A., Alsop, D.C., Shin, R.K., Atlas, S. & Grossman, M. (1995). The neural basis of the central executive system of WM. Nature, 378, 279-281.
Hanley, J.R., Young, A.W. & Pearson, N.A. (1991). Impairment of the visuo-spatial sketch pad.
The Quarterly Journal of Experiment Psychology Section A, 43(1), 101-125.
Lamme, V.A. (2006). Towards a true neural stance on consciousness. TRENDS in Cognitive
Neuroscience, 10, 494-501.
Li, C. & Guo,, K. (1995). Measurements of geometric illusions, illusory contours and stereo-depth at luminance and colour contrast. Vision Research, 35(12), 1713-1720.
Morris, N. & Jones, D.M. (1990). Memory updating in WM: The role of the central executive.
Brittish Journal of Psychology, 81, 111-121.
Rosenthal, C.R., Kennard, C. & Soto, D. (2010). Visuospatial sequence learning without seeing. PLoS One, 5(2010), e11906.
Sklar, A.Y., Levy, N., Goldstein, A., Mandel, R., Maril, A. & Hassin, R.R. (2012). Reading and doing arithmetic nonconsciously. Proceedings of the National Academy of Sciences, 109(48) , 19614-19619.
Soto, D., Mäntylä, T. & Silvanto, J. (2011). WM without consciousness. Current Biology,
21(22), R912 – R913.
Soto, D. & Silvanto, J. (2014). Reappraising the relationship between WM and conscious awareness. TRENDS in Cognitive Sciences, 18(10), 520-525.
15
Soto, D. & Silvanto, J. (2016). Is conscious awareness needed for all WM processes?
Neuroscience of Consciousness, 2016(1), niw009.
Stein, T., Kaiser, D., & Hesselmann, G. (2016). Can WM be non-conscious? Neuroscience of
Consciousness, 2016(1), niv011.
Schmidt, T. (2015). Invisible stimuli, implicit thresholds: Why invisibility judgments cannot be interpreted in isolation. Advances in Cognitive Psychology, 11(2), 31-41.
Oberauer, K. (2002). Access to information in WM: exploring the focus of attention. Journal
of Experimental Psychology: Learning, Memory and Cognition, 28(3), 411.
Overgaard, M., Timmermans, B., Sandberg, K. & Cleeremans, A. (2010). Optimizing subjective measures of consciousness. Consciousness and Cognition, 19(2), 682-684.
