Abstract
The effects of prior knowledge and expectations on recognition and decision-making are well-established. Detecting and recognizing objects get easier when they are presented in their usual context. Yet, related low-level sensory processing and the computational mechanisms underlying those effects remain controversial. Here we investigated behaviorally the effect of expectations on category-specific detection thresholds (Urgen & Boyaci, Vis. Res. 2021). At the beginning of each trial a task-irrelevant cue (face or house) provided information about the category of an upcoming target image (face or house) with a certain validity (75%, 50%, 100%, neutral). Next, the target image and its scrambled version were presented and backward masked on either side of a central fixation mark for a variable duration determined adaptively by a 1-up 2-down staircase procedure. Participants (N = 8) were asked to report the spatial location of the target image (2AFC). Duration thresholds were estimated in expected, unexpected, and neutral trials (in terms of cue-target category associations). Our behavioral results showed that compared to the neutral baseline, thresholds do not change in the expected trials, but they increase in the unexpected trials. Next, we show that a recursive Bayesian model can successfully predict the behavioral results. Modeling results suggest that internal parameters of the system are not altered with expectation, instead simply additional processing is required under the unexpected condition. Overall, our findings show that expectations do not speed up sensory processes, rather unmet expectations delay them. We argue that this happens because when expectations are violated further processing is required by the system. We also discuss our findings within the framework of predictive processing models and suggest that a simple neuronal model (Heeger, PNAS 2017) can parsimoniously explain the observed behavioral findings.