Abstract
Research on visual working memory (VWM) capacity has traditionally focused on estimating the maximum capacity. Yet, humans rarely load up their VWM maximally during natural behavior, since visual information often remains accessible in the external world. Recent work using more ecologically valid paradigms (where items remain externally accessible) has shown that observers utilize only one or two items from VWM before sampling from the external world again, suggesting that much fewer memory items are loaded in VWM than the typically reported capacity limits. Here, we investigate whether this lower reliance on VWM when information is externally accessible reflects (1) loading less items in VWM, or (2) resampling before VWM is depleted. To distinguish between these possibilities, we devised an online task, in which observers copied a model (six items in a 4x4 grid; always freely accessible) in an adjacent empty 4x4 grid. To manipulate the cost of resampling, we made it either easy or difficult (i.e., more time-consuming) to access the model. Increasing the cost of accessing the model clearly affected copying behavior: observers had fewer, but longer model inspections, longer completion times, and more erroneous item placements. Additionally, we (unpredictably) interrupted observers just before accessing the model and just after inspection of the model, with a 2-alternative-forced-choice (2-AFC) question probing their VWM content. We found that performance on 2-AFC questions was above chance when observers were probed before model inspection, and higher still when probed after model inspection. Surprisingly, 2-AFC performance was not modulated by the cost of accessing the model. Our findings show that with diminished accessibility to the external world, the tendency to resample decreases, while the number of items maintained in VWM remains comparable. We conclude that the external world is frequently resampled, although VWM contains latent information that is available upon forced request.