Explaining drives the discovery of real and illusory patterns

Explaining drives the discovery of real and illusory patterns Joseph Jay Williams (joseph_williams@berkeley.edu) Tania Lombrozo (lombrozo@berkeley.edu) Department of Psychology, University of California, Berkeley Bob Rehder (bob.rehder@nyu.edu) Department of Psychology, New York University Children’s and adults’ attempts to explain the world around them plays a key role in promoting learning and understanding, but little is known about how and why explaining has this effect. An experiment investigated explaining in the social context of learning to predict and explain individuals’ behavior, examining if explaining observations exerts a selective constraint to seek patterns or regularities underlying the observations, regardless of whether such patterns are harmful or helpful for learning. When there were reliable patterns– such as personality types that predict charitable behavior– explaining promoted learning. But when these patterns were misleading, explaining produced an impairment whereby participants exhibited less accurate learning and prediction of individuals’ behavior. This novel approach of contrasting explanation’s positive and negative effects suggests that explanation’s benefits are not merely due to increased motivation, attention or time, and that explaining may undermine learning in domains where regularities are absent, spurious, or unreliable. Keywords: explanation, self-explanation, learning, understanding, generalization, pattern detection, explanation impairment effect Explanation appears to possess a privileged relationship with learning and understanding. To know a fact without knowing why it is true can be deeply unsatisfying, not only to career learners like scientists but also to everyday learners and young children. Engaging in explanation goes beyond rote knowledge to genuine understanding, bringing with it the ability to generalize what is learned to novel situations. Research in education and cognitive development confirms and sheds light on the close connection between explanation and learning. Educational studies on topics ranging from math and physics to biology and computer programming have found that generating explanations has a powerful impact on learning and generalization (Chi et al, 1994; Renkl, 1997; for a review see Fonseca & Chi, 2010). Even young children exhibit an insatiable desire to request and learn from explanations (Chouinard, 2008; Legare et al, 2009), with prompts to explain accelerating major conceptual transitions in number conservation and theory of mind (Amsterlaw & Wellman, 2006; Siegler, 2002). The importance of explanation has been recognized in other disciplines as well. In cognitive psychology, explanations are believed to play a central role in the representation of conceptual knowledge, especially knowledge about causal relationships (Carey, 1985; Murphy & Medin, 1985). Research in artificial intelligence on how machines learn has been inspired by a focus on explanation as a process for learning from individual cases (DeJong, 1986; Mitchell et al, 2006). Finally, philosophers of science have attempted to characterize the nature of scientific explanation (Woodward, 2009). Despite the broad relevance of explanation, little is known about why the process of explaining, in particular, drives effective learning. Previous work has identified explanation’s role in revising beliefs and providing metacognitive insight into what is not known (Chi, 2000). Other investigators have proposed that explaining increases motivation and attention (e.g., Siegler, 2002). However, little experimental work has directly investigated and compared alternative theories of the content and consequences of explanation. This leaves important questions unanswered: What is the nature of the cognitive processing invoked by explaining? And why are the relative benefits of explanation greatest in acquiring knowledge that supports generalization? This paper investigates the hypothesis that the process of explaining drives the explainer to seek general patterns or regularities that can account for or produce whatever observation is the target of explanation. This hypothesis is the central tenet of the subsumptive constraints account of explanation (Williams & Lombrozo, 2010a), which is motivated by work in philosophy on pattern subsumption and unification theories of scientific explanation (Kitcher, 1981). Subsumption and unification theories suggest the defining property of an explanation is that it shows how the observation being explained is an instance of (subsumed by) a general pattern or regularity. For example, in answering “Why did that apple fall?” with “Because gravity accelerated it towards the Earth,” a hypothetical Newton shows how a particular event is subsumed under a general pattern, in this case a law of gravitation. Furthermore, the greater the number and diversity of observations attributable to a single pattern, the better the explanation. If learners are sensitive to a subsumptive constraint on explanations, then asking “Why?” should implicitly constrain their thinking, driving them to seek general patterns that underlie what they are trying to explain. And because patterns typically go beyond the idiosyncratic properties of the individual observations being explained, engaging in explanation should generate knowledge that transfers and generalizes to new contexts and problems, such as knowledge about underlying principles, laws, relationships, and causal regularities. The subsumptive constraints account thus sheds light on why explanation promotes learning, and especially generalization.