The Neural Machinery of Working Memory - A Comparative Approach
Is it possible to design several completely different neural architectures to create working memory? If not (or at least not completely), those components which are common to the different architectures might constitute a conditio sine qua non of the relevant neural machinery. This is the central argument of my talk. Working memory was first defined in cognitive studies with pigeons, and thus in animals without a neocortex. However, a detailed analysis of the forebrain structure which enable birds to create working memory shows several astonishing parallels to the prefrontal cortex. These studies make it likely that the essential neuronal component of working memory is the D1-receptor mediated dopaminergic enhancement of the firing rate of multimodal cells. This (phylogenetically probably very old) mechanism enables neurons to ‘hold’ any kind of stimulus-triggered activation and thus to create an internal representation of events which have to be used to guide future actions. This hypothesis is then tested in computational studies and in experiments with patients.