Abstract
Familiar spatial contexts are typical in everyday tasks, and we use knowledge of such regularities to promote optimal behavior. However, in the real world, we face a dilemma between exploiting learned context information to maximize performance and exploring novel information to discover new, potentially informative regularities. Here, we investigate this tradeoff. Participants searched for a T among Ls in displays containing a vertical borderline, dividing displays into two sides. Each side contained distinct spatial arrangements of items. On each trial, we presented one of 16 invariant arrangements (presented once per block) on one side and a random arrangement on the other side. Additionally, we manipulated the task-relevance of the repeated arrangements during a Training Phase, across three groups, to see whether people exploit regularities when it is beneficial or not: in the Target-Inside group, a target always appeared at one location inside the repeated arrangement and never appeared in the random arrangement. In contrast, a target was always presented outside the repeated arrangement in the Target-Outside group. Finally, a target appeared equally often inside or outside the repeated arrangements in the Target-Random group. During the Test Phase, to determine whether attention was biased to repeated vs. random display sides, a target was presented equally often inside or outside of the repeated arrangement. Training results revealed an early but short-lived acquisition and expression of context learning, demonstrating biasing attention towards regularities is transient when it cannot benefit behavior. Test results showed faster RTs for targets appearing outside compared to inside repeated arrangements for the target-outside group but no inside/outside RT difference in the target-inside/random groups. The failure of the target-inside group to bias toward the regularity – though it would benefit behavior – confirms a drive to explore novel/random stimuli. These results support an exploration/exploitation tradeoff in spatial context learning.
Meeting abstract presented at VSS 2018