Robert G. Eggleston, Michael J. Young, and Katherine L. McCreight
Most current computational architectures used as engines for software agents have been derived mainly from classic cognitive science. In this paper we introduce a new computational architecture, called DCOG, that has been motivated from an ecologically oriented study of work. It blends ideas from cognitive systems engineering with ideas from computational neuroscience to produce an architecture that is expected to provide more robust adaptive behavior, even under unanticipated work conditions. We describe key architectural principles for this new type of model and illustrate them through a discussion of a model of an air traffic control software agent responsible for airspace management. This model makes use of the DCOG constructs of work domain knowledge and multi-threaded parallel processing to achieve distributed, self-organized control. It also exploits the coding of information as signs, symbols, and affordances. Performance of the model, in comparison with humans performing the same task, is briefly described. The results suggest that the DCOG architecture yields outcome behavior that is similar in style to and comparable in performance with skilled human workers.