Sequential sampling from memory underlies perceptual decisions unyoked from actions

Perceptual decision-making refers to the class of decisions in which sensory evidence is used to categorize percepts and guide actions. Conventionally, categorical decisions are thought to precede motor actions. However, recent studies in nonhuman primates challenge this assumption -- when perceptual decisions were uncoupled from the actions they bear upon, animals postponed the decisions until relevant response options were revealed. To determine whether this postponement stems from cognitive limitations unique to nonhuman primates, we conducted a similar experiment with human subjects. Naive subjects viewed a random-dot motion (RDM) stimulus that was difficult to categorize. After a delay period following the RDM, two choice targets were presented and subjects decided which target lay closer to the perceived motion direction. Decision accuracy varied across subjects, reflecting individual differences in ability to integrate motion evidence. Notably, subjects with higher decision accuracy showed prolonged deliberation after choice-target presentation. Furthermore, the time they took to report their decisions depended on the strength of the motion evidence. This pattern of accuracy and decision reporting time could be accounted for by a bounded diffusion model in which subjects sequentially sample stored sensory information from memory during the target selection phase. When the RDM was challenging to categorize, the subsequent appearance of the targets provided a framework to interrogate stored evidence and render a decision. Our results reveal a strategic feature of working memory of retaining information based on its future utility. This observation opens new avenues for investigating how memory and decision-making interact.