Link To Publication in University of Groningen/UMCG Research Database
Link To Publication in University of Groningen/UMCG Research Database
Link To Publication in University of Groningen/UMCG Research Database
Published in:
Proceedings of the 13th International Conference on Cognitive Modeling.
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from
it. Please check the document version below.
Document Version
Early version, also known as pre-print
Publication date:
2015
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the
author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the
number of authors shown on this cover page is limited to 10 maximum.
252
and Bourne Jr. (2011) modelled the effects of fatigue on a
data entry task as a reduction in motivation in combination
with a reduction in attentional control. This attentional con-
trol parameter affects the activation of different pieces of in-
formation, and the larger this parameter is, the better these
pieces of information can be distinguished.
A previous ACT-R model of a sustained attention task that
is often used in mind-wandering studies (Peebles & Both-
ell, 2010) focused primarily on explaining response times
decrease just preceding an attentional lapse (as reflected in
an error). They produced this phenomenon by a competition
Figure 2: Model simulation of performance in the SART
between two response strategies: one strategy is responding
task described by Mrazek and colleagues (2012). The model
whenever a stimulus is detected, which is very fast, while an
(blue) captures both the number of SART errors and variabil-
alternative strategy first checks the stimulus before respond-
ity in response time observed empirically (red).
ing. When the fast strategy fails then ACT-R will switch to
the most costly slow strategy. Note that this model does not
implement an explicit cognitive mechanism for what happens goal is activated because it has been retrieved from episodic
during distraction. Here we intend to build on that previous memory. This goal then decays over time, and at some point
model by implementing a competition between a “distracted” the “distraction” goal becomes stronger (Figure 1). When the
and an “attentive” model, where the distracted model makes “distracted” goal is retrieved by this checking production, the
the mind-wandering process explicit. mind-wandering model commences.
Mind-wandering consists of a continuous retrieval of
Model declarative memories. The retrieval process keeps continuing
Our model of distraction (Figure 1) consists primarily of a until at some point a memory that says “remember to attend”
competition between a sub-model for paying attention to the is retrieved. At that point, the model returns to paying atten-
task and a sub-model for mind-wandering. The model was tion and the whole cycle can start again. There is spreading
implemented in the Adaptive Control of Thought-Rational activation between memories, which ensures that–as in real
(ACT-R) cognitive architecture (Anderson, 2007). Tasks are life–memories that are of the same valence (positive, nega-
implemented in this cognitive architecture by specifying a set tive, or neutral) tend to be recalled in sequence (van Vugt,
of if-then statements (production rules) that describe how dif- Hitchcock, Shahar, & Britton, 2012).
ferent cognitive resources interact. Two ACT-R mechanisms Our main goal in this paper is to find out whether the hy-
are of crucial importance for our model. First, ACT-R has a pothesized mind-wandering model can in fact describe em-
memory store, where the activation of each memory chunk pirical mind-wandering data. Studies have experimentally
determines its use and its retrieval time. The activation in studied mind-wandering by giving participants a very boring
turn is determined by how often a chunk is retrieved, its ac- task, in which participants are likely to drift off. Here, we
tivation at baseline, and how much activation spreads from will model data from two experiments: Mrazek, Smallwood,
other, related memory chunks. The second mechanism that and Schooler (2012) (Experiment 1) and Bastian and Sackur
determines what happens in the model at a particular moment (2013) (Experiment 2). Both experiments are variants of the
is the utility associated with each production rule. When pro- sustained attention to response task (SART), in which partic-
duction rules help to generate rewards, their utility goes up, ipants are requested to press a button as quickly as possible
leading them to be used more frequently. However, given that every time a target is presented, but to withhold a button press
in mind-wandering there are no external reward processes to a more rarely presented non-target (Cheyne et al., 2009;
that guide the process, we will not make use of this second Smallwood et al., 2004).
mechanism in our model. When the distraction model is inserted in a model of
In this application, the model starts out by focusing its at- the SART task, we assume performance is determined by
tention on the stimulus on the screen. When there is a stim- the following mechanisms, building on Peebles and Bothell
ulus, it will process the stimulus and perform the appropriate (2010)’s model. When task stimuli are presented while the
action. When there is no stimulus, it will continually run a model is in paying attention mode, the model will look at
production which checks what the most active goal (“paying the stimuli and retrieve the relevant stimulus-response map-
attention” or “distraction”) is in declarative memory (“check ping from episodic memory. Conversely, when the model is
whether attending” in Figure 1). The activations of the goals distracted, it will not retrieve the stimulus-response mapping
in declarative memory are governed by rules from episodic from episodic memory but instead respond with the habitual
memory decay (Altmann & Gray, 2008). This means that response. However, responding may take a little while, be-
items that are retrieved in activation, but over time the acti- cause the model will only be able to respond when it is not
vation decays. At the start of the task, the “paying attention” busy retrieving a memory in its mind-wandering train. This
253
Figure 1: Model time line. Each box cor-
responds to a production (some less im-
portant productions have been left out).
The model starts on the left top with
retrieving its current goal, correspond-
ing to goal checking. Initially, the “at-
tending” goal has the highest activation
(see dashed blue box), but over time the
ATTENDING DISTRACTED
attending goal declines in activation to
become similar to the distracted goal.
retrieve-current-goal
(from declarative)
"goal: attending" retrieve memory When this “distracted” goal is retrieved,
(from declarative)
the model switches to retrieving memo-
any memory
"goal: distracted" "remember to attend" ries from declarative memory, represent-
stimulus appearance ing mind-wandering. Mind-wandering
identify-stimulus (cyan) continues until “remember to at-
give standard response tend” (purple) is retrieved. At that time,
retrieve S-R*-mapping the model goes back to monitoring goals.
When a stimulus is presented (pink line),
activation
goal
respond or withhold
time
then the model identifies it and retrieves
key the stimulus-response mapping in case it
time
return to task is attending. When it is distracted, it fin-
*S-R = stimulus-response
after
distraction ishes retrieving the current distraction and
then presses the default response.
254
1.00
●
0.75
0.4
RT CV
0.50 ●
4.0
0.2 ●
●
0.25
3.5
0.00 0.0
data simulation data simulation
3.0
●
Figure 4: Model simulation of performance in the SART task
0−2 2−4 4−6 6−8 8−10
described by Bastian & Sackur (2013). The model (blue) cap- time into task (min)
tures both accuracy and variability in response time observed
empirically (red) reasonably well.
Figure 6: Predicted frequency of distractions in Experiment
1.
predicted
observed
SART task. However, the results are fairly weak since we
0.010 only fit two average numbers: the number of errors and re-
sponse time variability. More data are needed to adequately
constrain our cognitive model. We therefore use the com-
plete dataset collected by Bastian and Sackur (2013) to fur-
0.005
ther test the model, which allows us to examine more behav-
ioral measures. An additional advantage of that dataset is that
the task was interspersed with thought probes that asked the
0.000 participant to report on the content of their thoughts. The re-
0 250 500 750 1000 sponses to thought probes are another constraining factor for
RT (ms) our model. Moreover, it highlights an important advantage of
modeling mind-wandering explicitly, as we did here. When a
Figure 5: Response time distribution of SART performance model has no explicit process description of mind-wandering,
of Experiment 2, overlaying actual data (blue) with model it cannot predict responses to thought probes.
predictions (red).
Experiment 2
In Experiment 2, participants performed a very similar task
terminated much too quickly. A caveat in this assertion is that as in Experiment 1, although the timing was a little bit dif-
there may potentially be ways to change ACT-R parameters ferent. Importantly, we did not change the model parameters
to increase the duration of mind-wandering episodes. at all to predict performance in this task. In this experiment,
The duration of mind-wandering is determined by the the non-target consisted of the digit 3, and the target con-
episodic memory retrievals that make up the mind-wandering sisted of all other digits. The digits were presented for 500
process. When the pool of to-be-retrieved memories is ms with an interstimulus interval of 1500 ms. There were
larger, then distractions will tend to persist longer, because in total 888 stimuli; 811 targets and 77 non-targets. In addi-
the chance that the distraction-ending memory is retrieved tion, 24 thought probes that were randomly interspersed in the
is smaller. A larger number of retrievable memories corre- task. These thought probes asked a series of four questions
sponds to something akin to the number of retrieval cues. In about task performance. First, participants were asked “How
some contexts, people may be able to think of many different focused were you on the task? 0: on-task, 1: task-related
things, while in other context they can only retrieve a limited thought, 2: distraction, 3: mind wandering.” Secondly, “Did
number of items. A further determinant of distraction dura- you know that you were in the just-reported mental state or
tion is the association structure of the distracting memories. did you only notice it when asked? 0=aware, 1=unaware.”
When memories spread activation to the memory that ends The third question concerned the phenomenology/type of the
the distraction, this will decrease distraction duration; when thoughts, while the fourth question assessed the temporal ori-
they spread activation to other memories, this increases dis- entation of the thoughts (past, present, future, or no particular
traction duration. These factors could potentially be manipu- time). In this paper, we will only model the question about
lated to account for individual differences in distractability. whether the participant is on-task.
Together, these results show that it is possible to use Figure 4 shows that task performance could be modelled
our model of mind-wandering to simulate performance on a accurately with the model for Experiment 1, although in this
255
case, the model is performing slightly too well for the partici-
fraction on−task
parameters. 0.75
256
References mind-wandering in an executive-control task. Journal of
Experimental Psychology: Learning, Memory, and Cogni-
Altmann, E. M., & Gray, W. D. (2008). An integrated model
tion, 35, 196-204. doi: 10.1037/a0014104
of cognitive control in task switching. Psychological re-
McVay, J. C., & Kane, M. J. (2012). Why does working
view, 115(3), 602.
memory capacity predict variation in reading comprehen-
Anderson, J. R. (2007). How can the human mind occur in
sion? on the influence of mind wandering and executive
the physical universe? Oxford University Press.
attention. Journal of experimental psychology: general,
Andrews-Hanna, J. R., Smallwood, J., & Spreng, R. N.
141(2), 302.
(2014). The default network and self-generated thought:
Mrazek, M. D., Smallwood, J., & Schooler, J. W. (2012).
component processes, dynamic control, and clinical rele-
Mindfulness and mind-wandering: finding convergence
vance. Annals of the New York Academy of Sciences, early
through opposing constructs. Emotion, 12(3), 442-448.
view. doi: 10.1111/nyas.12360
doi: 10.1037/a0026678
Baird, B., Smallwood, J., Mrazek, M. D., Kam, J. W. Y.,
Peebles, D., & Bothell, D. (2010). Modelling performance in
Frank, M. J., & Schooler, J. W. (2012). Inspired by
the sustained attention to response task. In Proceedings of
distraction. mind wandering facilitates creative incubation.
the international conference on cognitive modelling.
Psychological Science, 23(10), 1117-1122. doi: 10.1177/
Ratcliff, R. (2002). A diffusion model account of response
0956797612446024
time and accuracy in a brightness discrimination task: Fit-
Bastian, M., & Sackur, J. (2013). Mind wandering at the
ting real data and failing to fit fake but plausible data. Psy-
fingertips: automatic parsing of subjective states based on
chonomic Bulletin & Review, 9(2), 278–291.
response time variability. Frontiers in Psychology, 4, 573.
Smallwood, J., Beach, E., Schooler, J. W., & Handy, T. C.
doi: 10.3389/fpsyg.2013.00573
(2008). Going AWOL in the brain: mind wandering re-
Bengson, J. J., Mangun, G. R., & Mazaheri, A. (2012). The duces cortical analysis of external events. Journal of Cog-
neural markers of an imminent failure of response inhi- nitive Neuroscience, 20(3), 458–469.
bition. NeuroImage, 59(12), 1534-1539. doi: 10.1016/ Smallwood, J., Brown, K. S., Tipper, C., Giesbrecht, B.,
j.neuroimage.2011.08.034 Franklin, M. S., Mrazek, M. D., . . . Schooler, J. W. (2011).
Cheyne, D., Carriere, J. S. A., & Smilek, D. (2009). Absent Pupillometric evidence for the decoupling of attention from
minds and absent agents: Attention-lapse induced alien- perceptual input during offline thought. PLoS ONE, 6(3),
ation of agency. Consciousness and Cognition, 18, 481- e18298. doi: 10.1371/journal.pone.0018298
493. doi: 10.1016/j.concog.2009.01.005 Smallwood, J., Davies, J. B., Heim, D., Finnigan, F., Sud-
Fox, K. C., & Christoff, K. (2014). Metacognitive facilita- berry, M., O’Connor, R., & Obonsawin, M. (2004). Sub-
tion of spontaneous thought processes: When metacogni- jective experience and the attentional lapse: Task engage-
tion helps the wandering mind find its way. In The cognitive ment and disengagement during sustained attention. Con-
neuroscience of metacognition (pp. 293–319). Springer. sciousness and Cognition, 13, 657-690. doi: 10.1016/
doi: 10.1007/978-3-642-45190-4 13 j.concog.2004.06.003
Gonzalez, C., Best, B., Healy, A. F., Kole, J. A., & Bourne Jr., Smallwood, J., & Schooler, J. W. (2014). The science of
L. E. (2011). A cognitive modeling account of simulta- mind wandering: Empirically navigating the stream of con-
neous learning and fatigue effects. Cognitive Systems Re- sciousness. Annual review of psychology.
search, 12, 19-32. doi: 10.1016/j.cogsys.2010.06.004 Teasdale, J. D., & Chaskalson, M. (2011). How does mindful-
Gunzelmann, G., Gross, J. B., Gluck, K. A., & Dinges, D. F. ness transform suffering? ii: the transformation of dukkha.
(2009). Sleep deprivation and sustained attention perfor- Contemporary Buddhism, 12(01), 103–124.
mance: Integrating mathematical and cognitive modeling. Tulving, E. (2002). Episodic memory: from mind to brain.
Cognitive Science, 33, 880-991. Annual Review of Psychology, 53, 1-25.
Howard, M. W., & Kahana, M. J. (2002). A distributed rep- van Vugt, M. K., Hitchcock, P., Shahar, B., & Britton, W.
resentation of temporal context. Journal of Mathematical (2012). The effects of MBCT on affective memory as-
Psychology, 46, 269-299. sociations in depression: measuring recall dynamics in a
Kane, M. J., & McVay, J. C. (2012). What mind wandering randomized controlled trial. Frontiers in Human Neuro-
reveals about executive-control abilities and failures. Cur- science, 6, 257.
rent Directions in Psychological Science, 21, 348354. VandeKerckhove, J. A., & Tuerlinckx, F. (2007). Fitting
Killingsworth, M. A., & Gilbert, D. T. (2010). A wandering the Ratcliff diffusion model to experimental data. Psycho-
mind is an unhappy mind. Science, 330(6006), 932. nomic Bulletin & Review, 14(6), 1011–1026.
Larson, M. J., Perlstein, W. M., Stigge-Kaufman, D., Kelly,
K. G., & Dotson, V. M. (2006). Affective context-induced
modulation of the error-related negativity. NeuroReport,
17(3), 329–333.
McVay, J. C., & Kane, M. J. (2009). Conducting the train
of thought: Working memory capacity, goal neglect, and
257