Neural evidence for decision-making underlying attractive serial dependence

Abstract

Recall of stimuli is biased by stimulus history, variously manifested as an attractive bias toward or repulsive bias from previous stimuli (i.e., serial dependence). It is unclear when attractive vs repulsive biases arise and if they share neural mechanisms. A recent model of attractive serial dependence proposes a two-stage process in which adaptation causes a repulsive bias during encoding that is later counteracted by an attractive bias at the decision-making stage in a Bayesian-inference-like manner. Neural evidence exists for a repulsive bias at encoding, but evidence for the attractive bias during the response period has been more elusive. We recently [1] showed that while different stimuli in trial history exerted different (attractive or repulsive) serial biases on behavioral reports, during encoding the neural representation of the current item was always repulsively biased. Here we assessed whether this discrepancy between neural and behavioral effects is resolved during subsequent decision-making. Multivariate decoding of magnetoencephalography data during working memory recall showed a neural distinction between attractive and repulsive biases: an attractive neural bias emerged only late in recall. But stimuli that created a repulsive bias on behavior led to a repulsive neural bias early in the recall phase, suggesting that it had already been incorporated earlier. Our results suggest that attractive (but not repulsive) serial dependence arises during decision-making, and that priors that influence post-perceptual decision-making are updated by the previous trials target, but not by other stimuli.