Informavores: Active information foraging and human cognition

Unlike a passive sponge floating in a sea of information, humans are active information foragers – informavores – who gather and consume new knowledge. From controlling the movement of our eyes to determining which sources of news to consult, judging the quality of alternative sources of information is a critical part of our behavior. The goal of this symposium is to bring together researchers who are working to understand the cognitive processes underlying active information foraging and how they interact with more general aspects of cognition. The study of active information search is in the midst of a renaissance. Psychological research from diverse areas ranging from developmental psychology (Schulz & Bonawitz, 2007), to higher level cognition (Nelson, 2005) to visual perception (Najemnik & Geisler, 2005) have begun to understand information gathering strategies in terms of a common set of computational principles. Simultaneous developments in machine learning on “active vision” and “active learning” (Settles, 2009) have resulted in new algorithms that optimize their own learning by focusing on useful training data. Similarly, models from optimal foraging theory from biology are being brought to bear on cognitive search processes both within and outside the mind (Pirolli, 2007; Todd, Hills, & Robbins, 2012). This symposium aims to bring together leading experts in this area to discuss how active information foraging can be understood from a diverse set of perspectives within cognitive science. Key themes include how prior knowledge influences search (Markant & Gureckis), how information and reward interact to determine choice (Meder & Nelson), developmental patterns in information seeking behavior (Nelson et al.), information foraging in complex sensemaking tasks (Pirolli), and the allocation of attention during statistical word learning (Yu). While each represents a distinct area of research, all discussants in the symposium share a core approach of applying computational models to understand information search in humans. The symposium should appeal to a broad set of attendees including educators, developmental psychologists, cognitive modelers, and computer scientists. The influence of priors on sequential search decisions Doug Markant and Todd Gureckis Normative models of information acquisition predict that people’s search decisions should be strongly influenced by their prior beliefs, which capture the set of alternative hypotheses they are considering. In the present experiments we tested whether people adjusted their information search behavior in response to sequential changes in the prior. Participants played a search game in which they had to identify the shape and location of multiple hidden targets in a display (similar to the board game Battleship). During the task they were told that the set of possible shapes had changed, and the key question was whether they would adjust their search decisions according to the predictions of a normative model. Manipulations of the prior included changes in the frequency of certain classes of targets as well as the introduction of higherorder constraints (e.g., that all targets would have the same shape). The results showed that an individual’s prior could be recovered from their sequences of search decisions, but that there were notable differences in their ability to adjust to certain changes in the hypothesis space, an effect that is not predicted by the normative model. We discuss the implications of these findings for how people generate and represent hypotheses during the course of information foraging. Is people’s information search behavior sensitive to different reward structures? Björn Meder and Jonathan Nelson In situations where humans actively acquire information for classification, information search preferentially maximizes accuracy (Nelson et al., 2010). However, the goal of obtaining information to improve classification accuracy can strongly conflict with the goal of obtaining information for improving utility when there are asymmetries in costs and benefits for classification decisions (e.g., in many medical diagnosis situations). Is people’s information search behavior sensitive to such asymmetries? We addressed this experimentally via multiple-cue probabilistic category-learning and information-search experiments, where the payoffs corresponded either to accuracy, with equal rewards associated with the two categories, or to an asymmetric payoff function with different rewards associated with each cate-