Joseph M. Saito¹ (), Philip Li¹, Timothy F. Brady¹; ¹University of California San Diego

Human observers are known to exhibit biases in their reports of visual memories following interference by perceptual distractors, such that memories are reported as being more alike distractors than they actually were. Leading sensory recruitment models posit that these biases can be fully explained by interactions between sensory representations of memory and perception within the local circuitry of visual cortex. Here, we challenge these models by showing that observers’ decision history modulates report biases beyond low-level sensory interactions. Across multiple experiments, we asked observers to explicitly compare a target stimulus held in visual working memory to a perceptual probe stimulus to determine their similarity before submitting a memory report. Target-probe pairs were sampled systematically across consecutive trials to form short runs of highly-similar or dissimilar combinations that were interleaved by critical pairs whose similarity was more ambiguous. Despite their fixed visual features, observers perceived the same ambiguous pairs to be similar more often when presented in runs of similar pairs than dissimilar pairs. More importantly, these history-related changes in perceived similarity were associated with corresponding changes in the magnitude of report biases, suggesting bona fide modulations in target-probe integration rather than mere priming of the similarity judgment. In a series of follow-up experiments, we extended this trial history effect by testing its generalizability across abrupt changes in stimulus type and by showing that changes in report bias are not explained by changes in memory strength. Finally, using simulations, we grounded out the predictions of sensory recruitment models to show that sensory interactions alone are unable to reproduce these behavioral patterns without the inclusion of separate decision-making processes that implement additional top-down weighting of mnemonic and perceptual information. Together, these findings highlight the necessary collaboration between early- and post-perceptual processes to determine the readout of memory contents following perceptual interference.

Acknowledgements: This work was supported by NSF BCS-2146988 awarded to TFB