Work

Learning What Matters: Goal certainty shapes working memory prioritization

Year: 2026
Type: preprint
Abstract: Cognitive resources are limited, yet people face an abundance of information. Often, which information matters is not stated and can change over time, so people must learn it through experience. This creates a tradeoff: uncertainty requires broad maintenance of multiple information sources, whereas certainty allows resources to be allocated more selectively. Does this reorganization of resources reflect enhancement of relevant information, degradation of irrelevant information, or both? Working memory provides a window into this process because differences in attentional priority should be reflected in what is remembered. To address these questions, we developed the Continuously Assessed Rule Inference Task, in which participants (N = 104) judged whether multidimensional stimuli followed a hidden rule that changed without warning, requiring them to continually update their beliefs about which information was relevant. On intermittent probe trials, participants reproduced the color, shape, and motion direction of the preceding stimulus. A reinforcement-learning model estimated trial-by-trial rule certainty from participants’ choices. As certainty increased, rule judgments relied more on evidence from the rule-defining (active) dimension, and memory for inactive features declined. A mixture model showed that active features became more likely to be retained, and inactive features less likely, with little evidence that the precision of retained features changed. These findings suggest that learning what matters reorganizes working memory by changing which information is retained, rather than how precisely it is represented. This shift from broad to selective maintenance may reflect a tradeoff between flexibility under uncertainty and efficiency once relevant information has been identified.
Language: English
Open Access status: green
OpenAlex ID: W7226028287
Cites:
Cited by:
Related to: 10