Abstract
A central purpose of exploration is to reduce goal-relevant uncertainty. Consequently, individuals often explore by focusing on areas of uncertainty in the environment. However, people sometimes adopt the opposite strategy, one of avoiding uncertainty. How are the conflicting tendencies to approach and avoid uncertainty reconciled in human exploration? We hypothesized that the balance between avoiding and approaching uncertainty can be understood by considering capacity constraints. Accordingly, people are expected to approach uncertainty in most cases, but to avoid it when overall uncertainty is highest. To test this, we developed a new task and used modeling to compare human choices to a range of plausible policies. The task required participants to learn the statistics of a simulated environment by active exploration. On each trial, participants chose to explore a better-known or lesser-known option. Participants generally chose to approach uncertainty; however, when overall uncertainty about the choice options was highest, they instead avoided uncertainty and chose to sample better-known objects. This strategy was associated with faster decisions and, despite reducing the rate of observed information, it did not impair learning. We suggest that balancing approaching and avoiding uncertainty reduces the cognitive costs of exploration in a resource-rational manner.