Abstract
Knowing which of two actions implies a higher risk is a basic prerequisite of rational action. We tested this ability in the visual domain by having observers perform a forced choice between uncertainties. Observers were presented with two stimuli in the context of a left/right orientation discrimination task. The stimuli were two Gabor patches to which Gaussian pixel noise was added, such as to generate high visual uncertainty. Observers were instructed to categorise only the stimulus for which they felt the least uncertain.
Unbeknown to the observers, trials could belong to one of two conditions: True Choice and False Choice. In True Choice trials, the stimuli had equal noise levels but differed in uncertainty. In False Choice trials, the stimuli were rotated so as to look different but to imply equal levels of uncertainty. This allowed us to assess the observers' ability to discriminate effectively between levels of visual uncertainty. Being able to choose the less “risky” of two stimuli implies that performance (in this case, probability of correct discrimination of orientation) will be higher on average than when that ability is not present. When the two options are equally risky, there is no gain to be had, thus providing a performance baseline.
We found that performance across a range of signal/noise ratios was indeed higher when the two stimuli differed effectively in uncertainty, demonstrating that observers are able to compare levels of visual uncertainty usefully.
Chaire d'Excellence to Pascal Mamassian from the French Ministry of Research