Abstract
Visual categorisation judgments often involve integrating multiple samples of information that occur in series. Some behavioural and neural studies suggest that evidence that is expected wield greater influence over decisions than evidence that is unexpected, whereas other studies have observed the converse. Here, we set out to test whether expected or unexpected samples are processed with higher gain using a hybrid detection/integration task. 10 healthy human participants viewed a stream of between 1 and 8 Gabor patches in rapid succession, with a view to later estimating the average tilt of the distribution from which they were drawn. Distribution means fell close to +45 degrees or -45 degrees from vertical. To-be-categorised streams were succeeded by a stimulus that either contained (signal present trials) or did not contain (signal absent trials) a target Gabor patch embedded in noise, and prior to estimation participants judged whether the signal was present or absent; detection performance was titrated to approximately d’=1 on average. Target Gabors were either tilted consistently or inconsistently with the to-be-estimated category mean. Surprisingly, we found that Gabors inconsistent with category information were detected with heightened sensitivity. These findings are inconsistent with Bayesian models of perceptual inference, and suggest a heightened sensitivity to erroneous or unexpected information.
Meeting abstract presented at VSS 2014