Abstract
Given attentional limitations and the dynamic nature of many scenes, strategic gaze direction may often be necessary to acquire needed information. Neural recordings have indicated that saccadic eye movement circuitry is sensitive to both stimulus probability and reward magnitude. We investigated the influences of these variables on strategic gaze allocation in visual search. Subjects searched an array of abstract objects for a change in orientation across successive frames while statistics of object changes and reward magnitude were manipulated. In one condition, we manipulated the probability that each object would change across all trials while keeping reward magnitude constant across all objects. As the experiment progressed, fixations became distributed among the objects in proportion to their relative probabilities of change. Total fixation durations to the highest-posterior objects (1063ms) were longer than those to the lowest-posterior objects (607ms), even on trials when these objects were not changing. The utility of this strategy is reflected in the faster reaction times on trials in which objects with higher change probabilities were identified as changing (642ms vs. 817ms). In a second condition, the probability of each object changing was held constant and the reward magnitude for detecting changes to each object was varied. As the experiment progressed, subjects learned to select objects that gave them greater rewards more often than objects with the same probability of changing but lower point values. Such sophisticated exploitation of environmental probabilities and rewards suggests that gaze allocation is governed by learnt models of the probabilistic structure of the environment.
Supported by NIH grants EY05729 and RR09283