Overcoming the costs of selective attention: Resolving rule-uncertainty supports acquisition of generalizable memory traces

Selectively attending to goal-relevant information supports performance. However, it can also limit knowledge about the statistical properties of the broader environment, as well as memory for features that are not currently relevant but may be useful in the future. This study investigated how a naturally selective and goal-oriented attentional system can be leveraged to support both immediate performance outcomes and generalizable knowledge through strategic broadening of attention. Undergraduate participants (N=120) completed a category learning task where optimal performance could be achieved by mapping the features of one spatially-defined dimension to the available category labels. Tests of feature memory and category generalization followed. Participants were randomly assigned to one of three instruction conditions: (1) a baseline condition in which the goal-relevant dimension was cued and the later memory test was undisclosed, (2) a rule-uncertainty condition in which the goal-relevant dimension was not cued, and (3) a memory-goal condition in which participants were informed of the upcoming memory tests. Attentional breadth during categorization, as measured by eye-tracking, predicted both subsequent memory for goal-irrelevant features and successful category generalization. Compared to baseline, early uncertainty about which dimensions were relevant to categorization significantly increased attentional breadth and memory test performance. No meaningful differences were observed between baseline and the memory-goal condition. The results suggest that manageable uncertainty can promote broader initial attention and more flexible, generalizable learning, thus overcoming the costs of selective attention.