Repeated exposure to task-relevant and task-irrelevant information – and their interaction – affect visual search performance

The human visual system adapts to statistical regularities in the environment to facilitate visual processing. While laboratory-based tasks make clear distinctions between how task-relevant and task-irrelevant visual information can guide this adaptation, such discretization is rarely available in the real world. As such, it remains unclear exactly what information the visual system tracks to flexibly adapt to a given task. The current study used a massive visual search dataset from the mobile game Airport Scanner. Effects of exposure over a range of more task-relevant (e.g., target presence) to less task-relevant (e.g., background context) features were analyzed in an omnibus model to predict response times in both target-present and target-absent trials. As in previous work (Kramer et al., Journal of Experimental Psychology: General, 151 (8), 1854, 2022), increased exposure to target-present trials significantly sped up the detection of targets and slowed the rejection of target-absent trials. Exposure to salient distractors reduced response times for target-present trials, potentially as a result of learned distractor suppression (Gaspelin & Luck, Trends in cognitive sciences, 22 (1), 79-92, 2018) or increased familiarity (Mruczek & Sheinberg, Perception & psychophysics, 67 (6), 1016-1031, 2005), but had no effect on target-absent trials. Exposure to background information decreased response times in both target-present and target-absent trials, with notable interactions between target and background exposure. Specifically, the effect of background information was more pronounced when target exposure was low, suggesting that less task-relevant context information is more likely to be tracked in the absence of more task-relevant information, namely, the presentation of targets. The findings highlight the importance of considering multiple sources of exposure in visual search tasks and demonstrate the value of large datasets in quantifying their complex interactions.