Adaptive Decision Support Systems using Cognitive Models of Trust

Abstract

Decision support systems are becoming more important and more frequently used in every day life. These systems are also getting more complex and are providing more information at the same time. It is therefore more difficult for users to trust these systems appropriately, when working with the decision support systems. An improvement of these so-called human-computer teams could be to make the system more aware of how it is being trusted and whether this is appropriate or not. It could then intervene when this is considered to be necessary. In this thesis a way of estimating trust is described and possible interventions based on this estimation are discussed. The interventions are an adaptation of the provided decision support: 1) advice argumentation and 2) advice censorship. The trust models and two types of adaptive support strategies were implemented for a classification task. A pilot experiment showed that the trust models predicted trust well, but no significant differences in team performance have been found of the adaptive support types compared to static support. The different possible reasons for this are discussed.