Attention and visual inference
Vision, for a long time, was regarded as a process in which information from a particular point in space is analysed in a purely feed-forward manner. More recent models describe vision as an inferential process that relies on activation history of the neuronal network in conjunction with extensive 'on-board' knowledge of the prior probabilities of particular feature constellations in the external world. Perception will be biased towards an expected feature constellation if the system has 'reason to believe' that this feature constellation is present, possibly disregarding 'true' constellations. However, the balance between prior information, and that coming in from the eye must be struck carefully, and in a dynamic fashion: it might prove disastrous if prior assumptions over-rode immediate information about features that are unexpected but crucial for survival. Spatial attention may mediate this balancing, typically so as to decrease the effect of spatio-temporal priors.
We use single and dual task human psychophysical approaches to determine the influence of attention and prior spatio-temporal information on visual perception.
Our data demonstrate that spatial attention reduces the impact of spatio temporal prior information on visual processing, while withdrawal of attention increases the impact of spatio-temporal priors, i.e. inferential processes occur mainly in the absence of attention.