Abstract
Performance across a wide range of tasks, such as memorizing spatial positions, tracking multiple objects, performing a visual search, memorizing the visual appearance of objects, or rapid counting, often seem to suggest that the visual system can localize, track, search, memorize, or count 3–4 objects at once. These findings place a strong constraint on the architecture of visual attention, suggesting that visual selection relies on a fixed number of mechanisms that each deal independently with a single object.
We argue that at least three sets of data are inconsistent with this conclusion. First, there is substantial between-subject variance in these capacity limits, and training can change these limits even within a single subject. Second, the number of items that can be selected is not fixed: we will present data showing that in each case, there is an alternative limiting factor that actually limits performance. For example, in the case of concurrent selection or object tracking, changing the limiting factor (object crowding) can shift the magic number from 1 to 8. Third, the limiting factor can vary depending on the task, and ranges from crowding to limited representational space, temporal decay, and measurement noise. Unlike a fixed number of objects, these signature constraints can distinguish between cognitive systems (e.g., attentional selection is influenced by the position of attended items in the visual field, whereas visual working memory is not).
The existing data appear to be inconsistent with the notion that a single, unitary system limits processing in all of these visual tasks to a fixed number of items. In place of magical number theory, we propose an information/resolution based model of visual attention, in which attentional capacity is more accurately characterized both in terms of the number of items attended and the accuracy or precision with which they are selected.