State-dependent serial dependence in perceptual decisions

Abstract

Traditional views suggest that human perception handles uncertainty using optimal strategies. For instance, when prior stimuli are more reliable than current ones, perceptual decisions rely more on the past, leading to stronger serial dependence. Here, we report findings that challenge this view. We asked human observers to reproduce the average orientation of an ensemble of stimuli under varying stimulus uncertainty. Contrary to optimal strategies, we found that serial dependence is stronger when prior stimuli are more uncertain. We hypothesize that fluctuations in stimulus uncertainty may influence internal states of observers, such as participants expectations about uncertainty and beliefs about their own performance. A striking finding is that manipulating these internal states through rigged feedback can yield drastic effects on serial dependence, even when external input (i.e., stimulus uncertainty) remained constant. Our findings suggest that phenomena like serial dependence can be better understood by considering internal states of the observer, beyond fixed computations and optimal strategies.