Deptartment of Optmometry and Visual Science, City University, London, UK
According to Bayesian theory, the influence of “a priori” biases on perception should be greatest when certainty about stimulus likelihood is least. These biases are thought to have evolved because certain types of stimulus are encountered more frequently than others. For example, our environment is particularly rich in perfectly orizontal and vertical things. Degrading information about stimulus orientation may shift perception toward a priori biases. We tested this hypothesis by asking observers to align a pointer with the average orientation of a briefly displayed array of Gabor patches. Unbeknownst to the observers, each patch’s orientation was drawn from a Gaussian distribution with near-horizontal (i.e. 0∘± 2∘ or 0∘± 14∘) or near-vertical mean (i.e. 90∘± 2∘ or 90∘± 14∘), and one of two possible standard deviations: σ = 2∘ or σ = 14∘. On average, responses elicited by the larger standard deviation were closer to the cardinal orientations. In particular, our results indicate a predisposition for seeing things as being perfectly horizontal or vertical. More generally, our results demonstrate that visual biases can be revealed by statistically degrading stimulus identity.