Adaptive but non-optimal visual search behavior with highlighted displays

Previous research has suggested that making certain items visually salient, or highlighting, can speed performance in a visual search task. But designers of interfaces cannot always easily anticipate a user’s target, and highlighting items other than the target can be associated with performance decrements. An experiment performed suggested that people attend to highlighting less than what an optimal model of visual search in highlighted displays predicts. Users’ sensitivity to highlighting’s predictiveness depends upon the proportion of trials in which highlighting actually predicts target location. We constructed an ACT-R model that reproduces the major effects of the experiment and which suggests that learning occurs at a very low level of this task.