Trust But Verify Design for Course of Action Displays

摘要

Automation, particularly of complex cognitive tasks, is bound to be incomplete, simplistic, or otherwise less than completely reliable. Recently, we have begun developing Trust but Verify techniques for increasing the effectiveness of even unreliable automation. The user's trust should be conditioned on known situational factors that affect the reliability of the automation, and users should be able to verify the automation results and operation to various qualitative degrees as the level of trust dictates. Here, we describe our preliminary work on these concepts in the domain of Course of Action (COA) selection for an Intruder Interception Task. This task involves deciding which of several available aircraft should be chosen to perform an interception of an unknown aircraft intruding into the air space. Based on repeated interviews with four subject matter experts, we identified and then distilled a set of factors essential to evaluating the optimal COA. We then designed a set of alternative displays to illustrate the factors based on the Trust but Verify concept and general human factors display guidance. Here we analyze the benefits and costs of two major design decisions: whether to display the COA factors using a tabular or graphic organization, and whether or how to integrate the COAs with the map or with each other in a common table.