Abstract
What you are seeing now is likely to resemble what you saw moments earlier. It is thought that our brains exploit this stability to maintain our perception of the world. This is evident in a perceptual effect termed “serial dependence” where current perception can be attracted towards previous perception, causing misperception of current stimuli. This may well be a way of compensating for noise in visual input; when current visual stimuli become ambiguous, the recent past may serve as a useful guide for what we should be seeing. For this approach to make sense, the level of uncertainty associated with past and present stimuli should be considered; it does not make sense to rely on previous observations if our perception of them was uncertain. This consideration of uncertainty could make serial dependence a form of Bayesian perceptual inference. We tested this idea using an orientation judgement task featuring different levels of stimulus uncertainty. Participants were asked to reproduce the orientations of briefly observed (500ms) orientation stimuli. These stimuli were produced by applying a bandpass orientation filter to the Fourier amplitude spectrum of a 1/f noise pattern. The width of this filter was altered to produce low and high noise stimuli. Participants repeated the orientation reproduction task over hundreds of trials, allowing us to quantify the effect of past and present levels of uncertainty on serial dependence. The level of uncertainty in current stimuli was found to influence serial dependence; higher levels of uncertainty were associated with enhanced serial dependence. This suggests that observers rely more on the recent past when current observations are ambiguous. However, we could not identify an effect of previous stimulus uncertainty. These results constitute partial evidence for serial dependence as a Bayesian process. Other, non-Bayesian explanations may be able to account for this phenomenon.