Abstract
​Observers were shown varying numbers of red and green dots randomly distributed in space. Observers indicated whether there were more green than red dots. Surprisingly, we found that only a small fraction of dots was taken into account for decision making. Similar results were found for oriented lines. Classically, such errors are explained by noise corrupting the neural representation of the dots. The idea that internal noise causes errors is at the very heart of Signal Detection Theory, and Thurstonian models in general. Here, we propose instead that errors occur because of undersampling rather than noise, i.e., observers choose only a small fraction of dots for decision-making, and the undersampling rate determines the error rate. This scenario can be best described by classical Bernoulli urn models. Our results are similar to attentional blindness, where salient objects often go unnoticed, such as a gorilla in a basketball game. Here we showed that this can also occur in processing of elementary features, such as dots or lines. Our results open a fundamentally new view on visual perception and decision making.
Meeting abstract presented at VSS 2014