Abstract
Why don't we normally see a plaid when presented with two orthogonal gratings one in each eye? A common answer is that the system implements a physical constraint of one image being projected to the two eyes at one place and time. This constraint could be pre-wired and fixed or alternatively, acquired and modifiable. Here we provide evidence for the latter by reporting three types of observers who see fused plaids where normals see binocular rivalry (BR). One type (two observers) with a history of corrected strabismus and currently normal vision, had a high voluntary control over BR and most strikingly could perceive a fused plaid when instructed to do so for most (50–80%) of the time (e.g. 4 cpd, 40% or uneven contrast). Another type (one observer) had a 10 deg. opposite torsion in each eye following an accident, which caused him diplopya and constant perception of dichoptic plaids. The third type (four observers) consisted of individuals with strabismic amblyopia who could not perceive BR, but perceived dichoptic plaids when the contrast of the dominant eye was appropriately reduced. These cases suggest that rivalry reflects the natural outcome of a system that adapts to the statistics of the external input. Rare stimuli (such as local dichoptic plaids) are poorly represented while extensive exposure to such stimuli as in the cases reported here makes them interpretable and visible. To account for our observations we propose a computational model of BR, based on the ideas of statistical inference and representational learning.