It is estimated that humans can spend nearly 50% of their waking hours in MW, where one's attention shifts from external tasks to internal thoughts (Killingsworth & Gilbert, 2010; Seli et al., 2018). The costs of MW are extensively documented, ranging from reduced performance on lab-based cognitive tests and deleterious effects on everyday activities (McVay, Kane, & Kwapil, 2009; Mooneyham & Schooler, 2013; Szpunar, Moulton, & Schacter, 2013; Yanko & Spalek, 2014) to its profound emotional cost (Killingsworth & Gilbert, 2010). Given these detrimental effects, a key question arises: Why do we engage in MW so much? What are its evolutionary benefits? Although previous research has focused on the drawbacks of MW, its potential advantages remain unclear. In this study, we sought to explore the potential benefits of MW by examining its relationship to statistical learning. Our findings suggest that while MW can impair visuomotor task performance by altering the SAT, it can also facilitate the efficient extraction of predictable patterns, resulting in improved statistical learning.
It has been suggested that adaptive aspects of MW encompass divergent thinking/creativity and autobiographical planning/prospective simulations (for review see Mooneyham & Schooler, 2013) which aligns well with the association between MW and activity in the brain’s default mode network (Christoff, Irving, Fox, Spreng, & Andrews-Hanna, 2016; Mittner et al., 2014) and the proposed functional relevance thereof (Buckner & Carroll, 2007; Shi et al., 2018; Takeuchi et al., 2011). However, both the contribution of MW to creative incubation (Murray et al., 2021; Smeekens & Kane, 2016; Steindorf et al., 2020), and its link to the default mode network (Groot et al., 2021; Groot, Csifcsák, Wientjes, Forstmann, & Mittner, 2022; Kucyi, Hove, Esterman, Matthew Hutchison, & Valera, 2017) have been questioned recently. In light of such findings, the beneficial influence of MW on cognitive performance awaits further empirical support. In this study, we report that MW facilitates the extraction of predictable patterns in the environment, resulting in improved statistical learning while also affecting visuomotor performance by shifting the SAT. Given the ubiquitous role of statistical learning and predictive processes both in terms of shaping behavior and underlying neural computations (Fiser, Berkes, Orbán, & Lengyel, 2010; Fiser & Lengyel, 2019; Santolin & Saffran, 2018), the current findings open new avenues for delineating the functional relevance of task-unrelated thought in human cognition and everyday functioning.
Our results indicative of increased statistical learning during MW could be interpreted in the so-called competition framework (Daw, Niv, & Dayan, 2005; Poldrack & Packard, 2003), that proposes an antagonistic relationship between cognitive control and statistical learning, where greater involvement of the former processes may impede the extraction of environmental statistical properties. For example, previous research has indicated that statistical learning performance is negatively associated with control functions mediated by the prefrontal cortex (Nemeth et al., 2013; Smalle, Daikoku, Szmalec, Duyck, & Onen, 2022; Virag et al., 2015). Moreover, a functional connectivity study using EEG phase synchronization (Tóth et al., 2017) has found a similar pattern, showing that statistical learning negatively correlates with the activity of a large-scale fronto-parietal network. Moving from correlational towards causal evidence, inhibitory non-invasive brain stimulation targeting the dorsolateral prefrontal cortex has resulted in improved predictive processing, as measured by statistical learning (Ambrus et al., 2020). On the other hand, MW has also been linked to the shift in allocation of executive resources and impaired task-associated cognitive control. According to the executive failure view, MW episodes emerge as a result of the inability to maintain current goals via sustained task-focus and shielding against task-unrelated interference (McVay & Kane, 2010). The negative association between MW and executive performance has been consistently shown in the finger-tapping version of the classical random number generation task (Baddeley, Emslie, Kolodny, & Duncan, 1998), in which participants are asked to provide random sequences of finger taps to the rhythm of an ongoing metronome, while intermittently being probed about their mental states (Alexandersen et al., 2022; Boayue et al., 2021; Groot et al., 2022). In line with our results, these studies (along with many others, e.g., Cheyne et al., 2009; Bastian & Sackur, 2013; Stawarczyk, Majerus, Maj, Van der Linden, & D’Argembeau, 2011) also showed increased behavioral variability during MW, providing support for the validity of our assessment of task-focus in the ASRT task. Thus, given that MW is coupled with impaired cognitive control, while statistical learning is typically more efficient in states of depleted executive resources, it is reasonable to assume that the facilitated extraction of the statistical contingencies in the ASRT task during MW periods was mediated via (failure of) the executive system. Nevertheless, since we did not directly measure executive control (neither behaviorally, nor its neural correlates), future studies should address if MW is beneficial to statistical learning through reduced executive control.
Studies on the neural correlates of MW suggest that impaired executive control is not necessarily the sine qua non of MW. Reduced amplitude of canonical event-related potentials (P100, N100, and P300) in electroencephalographic (EEG) signals appeared to be robust markers of dampened cortical processing when participants mind wandered, compared to periods when participants focused on the task, that in turn, elicited larger evoked potentials (Kam et al., 2022). Interestingly, reduced cortical processing linked to task-unrelated thoughts was observed in response to both target and distractor stimuli, indicating that MW reflects a general decoupling from the external environment instead of failures in task-relevant processing and problems of distraction due to impaired executive control (Barron, Riby, Greer, & Smallwood, 2011). This argument is strengthened by the fact that the aforementioned P300 component has been found in previous research - both with stimulus-locked and response-locked event-related potentials - on predictive processes measured with the same statistical learning task used in the current study (Kóbor et al., 2019, 2018). Future studies directly testing the neural correlates of MW during statistical learning and predictive processes seem highly warranted.
Our findings suggest that MW facilitates the processing of probabilistic patterns of sensory inputs. Apparently, this finding is somewhat paradoxical: MW is linked to sensory decoupling, but still, it facilitates the processing of sensory inputs? Here we argue that MW as the mental reflection of a transient and spatially local off-line state (Andrillon et al., 2019; Jubera-Garcia, Gevers, & Van Opstal, 2021; Jubera-García, Vermeylen, Peigneux, Gevers, & Opstal, 2021; Wienke et al., 2021) facilitates information processing (in our case, statistical learning) during task-acquisition. More specifically, we posit that improved statistical learning during MW can also be attributed to rapid memory consolidation processes during periods of sensory decoupling. The stabilization of memory traces is known to be either time-dependent or sleep-dependent (Wamsley, 2022), with the latter being linked to low-frequency neural activity, which has been extensively documented (Diekelmann & Born, 2010; Rasch & Born, 2013). Although MW has also been associated with slow waves (Andrillon et al., 2019; Wienke et al., 2021), these slow waves are generated in resting wakefulness and expressed in more localized networks, a phenomenon also known as local sleep (Andrillon et al., 2019; Krueger, Nguyen, Dykstra-Aiello, & Taishi, 2019; Vyazovskiy et al., 2011). Our findings suggest that enhanced statistical learning observed during MW may be driven by memory consolidation associated with local sleep in the waking brain (Wamsley, 2022; Wamsley & Summer, 2020). This implies the existence of a third category of memory consolidation - in addition to sleep-and time-dependent consolidation - referred to as local sleep-dependent consolidation. This type of consolidation may provide a compelling explanation for the inconsistent findings of sleep-dependent memory consolidation in procedural or statistical learning tasks (Pan & Rickard, 2015). According to this model, the brain consolidates learned material during task performance by utilizing local slow-waves that manifest as task-unrelated MW. This idea is consistent with the opportunistic theory of memory consolidation (Mednick et al., 2011), which posits that the brain consolidates memory traces when it can, whether awake (time-dependent consolidation), asleep (sleep-dependent consolidation), or during local-sleep states. In addition, our finding fits into the emerging number of studies indicating that even ultra-short periods of post-learning waking rest are beneficial for the stabilization of memory traces (Wamsley, 2022; Wamsley & Summer, 2020). Nevertheless, validation of this speculation requires sophisticated magnetoencephalography or EEG studies to provide direct evidence of the relationship between MW, local sleep, and improved learning performance.
Our speed and accuracy results demonstrate how MW is related to attentional lapses and affects choice behavior and decision-making. Performance errors in sustained attention tasks (Robertson, Manly, Andrade, Baddeley, & Yiend, 1997) are widely regarded as behavioral markers of attentional lapses and have been reported during episodes of MW (McVay & Kane, 2012a; Mooneyham & Schooler, 2013; Stawarczyk et al., 2011). Despite this, commission errors (inadvertent responses to no-go stimuli) in the sustained attention tasks are preceded by shorter RTs (McVay & Kane, 2012), raising the possibility that performance errors during MW are not necessarily sensitive to (failures of) sustained attention per se, but rather, they reflect changes in response style, i.e., the SAT (Dang, Figueroa, & Helton, 2018; Seli, Jonker, Solman, Cheyne, & Smilek, 2013). While the MW- and SAT-accounts of impaired sustained attention task performance are not mutually exclusive (Seli, 2016), evidence on MW-associated shifts in favor of faster (albeit more erroneous) responses in other cognitive tasks is scarce. In our analysis of visuomotor performance, we have found a clear indication that MW was associated with a faster response style at the expense of accuracy. This result corroborates previous conclusions (Seli, 2016) and is also in line with a more recent report on the relationship between experimentally-induced worrying (a special manifestation of task-unrelated thoughts) and SAT (Hallion, Kusmierski, & Caulfield, 2020). Given that MW frequency increased while faster responses became more dominant towards the end of the ASRT task (time-on-task effects), it is also possible that the shift in SAT in the current study was only an epiphenomenon to MW. However, we argue against this assumption since task progression did not influence the MW-SAT association, pointing at a potential causal interplay between these phenomena. Overall, we provide support to the notion that lapses of attention during cognitive tasks might be characterized by a faster response style rather than be merely viewed as periods of generally impaired performance. Due to the strong impact of speed vs. accuracy arbitrations on decision-making (Gold & Shadlen, 2007), this finding can refine our understanding of the effect of MW on choice behavior.
Taken together, MW may hinder precise reactions to external stimuli, but it can also enhance an individual's ability to recognize consistent patterns in their surroundings. This, in turn, can lead to more effective predictions of upcoming events. Future studies may examine if our findings could be generalized to other memory domains. The automatic and unintentional acquisition of predictable patterns in the environment might specifically benefit from MW, as such learning occurs without effortful, controlled processes. Moreover, statistical learning underlies the prediction of stimulus-outcome dependencies. Such elementary processes may align with the dominantly prospective nature of MW that is assumed to play a central role in planning and future-oriented behavior. The extraction of probabilistic regularities of the surroundings might serve as one of the building blocks of this endeavor. Furthermore, considering that statistical learning is a crucial aspect of skill and habit development (Lieberman, 2000; Nemeth et al., 2011; Romano Bergstrom, Howard, & Howard, 2012; Spencer, Kaschak, Jones, & Lonigan, 2015), the results of our study may possess generalizability to these fundamental learning functions. As a result, our findings could have promising implications for future research that investigate the advantages of MW in various learning domains, including language acquisition, motor skills, music learning, and social skill development.