Displays for effective human-agent teaming: Evaluating attention management with computational models

Abstract

In information-dense work domains, the effectiveness of display formats in drawing attention to task-relevant information is critical. In this paper, we demonstrate a method to evaluate this capability for on-screen indicators used to proactively monitor multiple automated agents. To estimate the effectiveness of indicator formats in drawing attention to emerging problems, we compared the visual salience of indicators, as measured by computational models, to task-relevant attributes needed during proactive monitoring. The results revealed that standard formats generally do not draw attention to the information needed to identify emerging problems in multi-indicator displays, and validated the success of formats designed to more closely map task-relevant information to visual salience. We additionally report an extended saliencybased monitoring model to predict task performance from saliency and discuss implications for broader design and application.