Developmental Change in Explore-Exploit Decision-Making

• In Event: Poster Session 01
  In Poster Session: PS 01 Topic - Cognitive Processes

Abstract

The development of explore-exploit decision-making has garnered growing interest in recent years. This basic mechanism of human learning reflects the balance children must achieve between focusing on the consistency of exploiting what they know versus open exploration of what has yet to be experienced. Various tasks are used to measure explore-exploit behavior, but each differs in overall structure and demands. To date, most studies examining developmental changes in these processes have relied upon single-task designs, leaving open questions about the extent to which various tasks tap similar phenomena.
Method: We compared early adolescent (ages 10-13, n = 62) and young adult (ages 18-32, n = 59) behavior across three different types of explore-exploit decision-making tasks: a foraging task, two sequential search tasks, and a bandit task (the Horizon task). While they differed in structure, each of these tasks asked participants to choose between gathering additional information (explore) or taking advantage of what they already knew (exploit) on a given trial. Exploration on the bandit task was further classified as either directed (strategic, goal-oriented) or random (by chance, based on decision noise) using computational modeling.
Results: Age-group differences in explore-exploit choices varied depending on the task. In the bandit task, youth engaged in less directed exploration than adults (F(1,118), = 4.9, p < .03; see Figure 1), while youth explored marginally more than adults in a sequential search task F(1,118), = 2.66, p < .1). We then used factor analysis to assess the overall structure of explore-exploit decision-making in each group. Factor analysis showed that patterns of decision-making were represented by factors that differed slightly between youth and adults (see Table 1). Among early adolescents, factor analysis revealed that performance across tasks converged on two factors (chi-square = 0.34, p = .56). The first factor (alpha = .48) included exploration in the foraging task and directed exploration in the bandit task. The second factor (alpha = .56) included exploration in the two sequential search tasks, and random exploration in the bandit task. Factor analysis in the adult sample also resulted in two factors (chi-square = .077, p = .78), but their composition was somewhat different: as with youth, the two sequential search tasks and random exploration clustered together, but directed exploration did not cluster with foraging (see Table 1).
Conclusion: This study suggests that there may be multiple forms of explore-exploit decision-making depending on the structure of one’s environment, and that different processes involved in this phenomenon may operate at varying developmental stages. Exploration appears to become more goal-directed from early adolescence to adulthood, which may reflect cognitive maturation processes.

Authors

• Madeline B Harms, University of Minnesota - Duluth
  Presenting Author

• Yuyan Xu, University of Wisconsin - Madison
  Non-Presenting Author

• C. Shawn Green, University of Wisconsin-Madison
  Non-Presenting Author

• Kristina Woodard, University of Wisconsin - Madison
  Non-Presenting Author

• Robert Wilson
  Non-Presenting Author

• Seth David Pollak, University of Wisconsin - Madison
  Non-Presenting Author