Ying Zhou¹, Kartik Sreenivasan^1,2, Daryl Fougnie¹; ¹Division of Science and Mathematics, New York University Abu Dhabi, ²Center for Brain and Health, New York University Abu Dhabi

Our memory of the world is influenced by what we expect. For example, working memory (WM)-guided decisions are biased towards highly probable stimuli (Jabar & Fougnie, 2021). However, it is unclear precisely how expectations bias WM-guided decisions: expectations may directly modulate WM representations themselves; alternatively, expectations may alter the process by which WM representations are translated to behavior. Perceptual studies provide evidence for both possibilities — some find that expectations bias perceptual representations (Kok et al., 2013), whereas others demonstrate that expectations influence decision-making instead of perception (i.e., cognitive impenetrability; Firestone & Scholl, 2016; Rungratsameetaweemana et al. 2018). To adjudicate between these alternatives, we had 15 participants perform a WM delay task for orientation. In the expectation condition, orientations were not random but instead drawn from a Gaussian distribution that was explicitly communicated to participants. Participants’ reports indicated that WM-guided decisions were biased towards likely stimuli (bias = 6.62 ± 2.91°; p < .001). Moreover, modeling confirmed that this bias arose from the integration of memory and expectation, rather than a shift in response strategy. To investigate whether expectations were integrated with WM representations, we used a Bayesian decoder to identify the content of WM from fMRI activity measured during the delay period of each trial. We focused on WM representations in early visual areas, as these regions provide the most reliable decoding of orientation information. Despite seeing clear behavioral effects of expectation, there was no consistent evidence for bias in single-trial WM representations in early visual areas, suggesting that WM representations maintained in early visual areas may be impenetrable to expectation. How WM representations are converted into decisions is a critical, yet poorly understood topic in WM. The present work highlights that even simple WM-guided decisions may require integrating knowledge across multiple stages of processing.

Acknowledgements: This work was supported by the NYUAD Center for Brain and Health, funded by Tamkeen under NYUAD Research Institute grant CG012 to KKS