Abstract
Working memory (WM) allows for the online maintenance and manipulation of information. Although many cognitive functions depend on WM, its capacity is strictly limited. The strategic allocation of the resources that support WM can mitigate these limitations and improve the quality of WM. For example, incentives can improve visual WM accuracy, suggesting that people can flexibly allocate WM resources to maximize their expected gains. Here, using fMRI and eye-tracking we tested the hypothesis that cognitive effort impacts the quality of WM representations stored in the human brain. Participants performed a visual-spatial WM task with two trial types which required more (hard) or less (easy) memory precision. The amount of precision on each trial type was determined within-subject through two separate staircase procedures. On hard compared to easy trials, participants were slower to respond and had larger pupil sizes during the memory delay, suggesting that hard trials required more cognitive effort. Neurally, the amplitudes of BOLD activation during the memory delay were greater on hard relative to easy trials in retinotopically defined visual field maps in visual, parietal, and frontal cortex, suggesting that the gain of neural responses is mediated by cognitive effort in visual WM. Moreover, using a Bayesian decoder, we found that in some visual field maps the decoded WM representations were more accurate and more precise during hard trials. These results first demonstrate that subjective and difficult to control constructs like cognitive effort can be studied effectively and can have effects on WM quality. They also demonstrate that effort, like priority and incentive, belongs in the family of factors that influence the flexible allocation of limited WM resources. Finally, they have implications for translational research within psychiatric populations, where motivation deficits could underlie apparent WM impairments, but also serve as potential targets of clinical remediation.