Abstract
The perceived direction of a random-dot kinematogram in a fine discrimination task is systematically biased away from the decision boundary. Stimulus uncertainty combined with a neural read-out that is optimized for that specific discrimination task was thought to cause these biases (Jazayeri and Movshon, 2007). Based on this explanation, we would expect the biases to disappear if subjects did not have to perform the discrimination task. We conducted two experiments to test this prediction. Experiment 1 (control) was aimed to reproduce the effects of the original study, using stimuli that consisted of an array of line segments (length of 0.5 degs). The orientation of each line segment was drawn from a Gaussian around the array's mean orientation (stimulus orientation). Stimulus orientation was uniformly sampled from a range +/−21 degrees relative to a reference orientation. Black marks indicated the randomly chosen reference orientation. Gray wedges around the reference mark indicated the total range of stimulus orientations (prior). Subjects first had to indicate whether the stimulus orientation was clockwise (CW) or counter-clockwise (CCW) relative to reference orientation. Subsequently they had to estimate the stimulus orientation. Experiment 2 was identical to Experiment 1 except that subjects were given the stimulus' category (CW/CCW) before stimulus presentation. The results show very similar repulsive biases in both experiments. Hence, we conclude that the biases are not the result of a specific, task-dependent neural read-out as previously suggested, but instead are consistent with a “self-consistent“ Bayesian observer model (Stocker and Simoncelli, 2008).