Dissociating 'what' and 'how' in visual form agnosia: a computational investigation

Patients with visual form agnosia exhibit a profound impairment in shape perception (what an object is) coupled with intact visuomotor functions (how to act on an object), demonstrating a dissociation between visual perception and action. How can these patients act on objects that they cannot perceive? Although two explanations of this 'what-how' dissociation have been offered, each explanation has shortcomings. A 'pathway information' account of the 'what-how' dissociation is presented in this paper. This account hypothesizes that 'where' and 'how' tasks require less information than 'what' tasks, thereby allowing 'where/how' to remain relatively spared in the face of neurological damage. Simulations with a neural network model test the predictions of the pathway information account. Following damage to an input layer common to the 'what' and 'where/how' pathways, the model performs object identification more poorly than spatial localization. Thus, the model offers a parsimonious explanation of differential 'what-how' performance in visual form agnosia. The simulation results are discussed in terms of their implications for visual form agnosia and other neuropsychological syndromes.