ABSTRACT

Every animal or intelligent machine requires an attentional mechanism in order to make sense of events in its environment and predict their consequences on the basis of past contingencies. If a neural network has no criterion for selecting which stimuli to attend to and which to ignore, it becomes so overwhelmed with stimuli as to make functioning impossible. Solutions to the problem of selective attention in biological organisms are likely also to have profound implications for the control of adaptive machines, through the incorporation into such machines of goal direction or planning (see, e.g., Athale, Friedlander, & Kushner, 1986; Barto, Sutton, & Anderson, 1983; Cruz, 1992).