Evidence That Mind-Wandering Enhances Deep Learning
There is a growing body of evidence arising from distinct methodologies that links mind-wandering with behaviors and outcomes that are indicative of improved deep learning. In an early research program that anticipates key elements of contemporary research into mind-wandering, J. L. Singer and his colleagues investigated the psychological correlates of a person's propensity for daydreaming. They found daydreaming was positively correlated with a number of psychological measures of social health as well as higher levels of creativity (Singer, 1974; Singer & Antrobus, 1963; Singer & Schonbar, 1961). These results are well explained on the hypothesis that mindwandering facilitates improved learning of underlying hidden patterns in one's experiences, including one's social experiences.
Another line of research takes advantage of mind-wandering having a reliable neural correlate, activity in the default mode network (Fox et al., 2015), in order to assess the psychological consequences of increased mind-wandering. For example, Wig and colleagues (Wig et al., 2008) examined the extent of default network activation during breaks during a cognitive task, with greater activation plausibly indicating more vigorous mind-wandering. They found higher default network activity was linked to measures of recognitional memory performance, suggesting a link between mind-wandering and the formation of more stable and efficient associative links between memory items (see also Yang, Bossmann, Schiffhauer, Jordan, & Immordino- Yang, 2012, for an analogous finding that coupling within the default network predicted greater depth of social understanding).
Perhaps the strongest kind of evidence for a link between mindwandering and deep learning directly manipulates the quantity of mind-wandering and measures the consequences for deep learning- related constructs. In an elegant study, Baird and colleagues (2012) presented participants with the unusual uses task (UUT), which asks participants to list as many unusual uses as possible for a common object, such as a toothpick, in a set amount of time. They received the same UUT twice, at baseline as well as at a later "time 2." By having participants perform cognitive tasks of different levels of difficulty during the intervening period, the quantity of mindwandering during this period was successfully manipulated. They found that participants in the undemanding task condition, who exhibited the most mind-wandering during the intervening period, had significantly improved scores on their time 2 UUT. A natural explanation of this finding is that, consistent with the deep learning model, mind-wandering during the intervening period served to extract hidden patterns from prior experiences with the relevant UUT object thus enabling better performance when given the UUT a second time.