Artificial Intelligence (AI) System Interface Attributes: Survey and Analyses

Abstract

This study represents a first phase in the design of a human factors tool for artificial intelligence (AI) system assessment. Desirable attributes of AI interfaces were identified as a result of a review of the literature. A questionnaire was developed where explicit definitions were presented for 17 selected attributes. Nineteen AI system developers rated the attributes under four different context conditions: (1) no context (i.e., general application); (2) a bomber crew system; (3) a command and control station; and (4) an intelligence analyst position. Examination of the ratings showed that attributes associated with communication and education aspects of AI were given the lowest ratings, whereas attributes pertaining to tasks which impose a high level of time stress received the highest ratings of importance. The ratings data were subjected to Multidimensional Scaling (MDS) analyses where the following dimensions were determined: (1) tasks performed principally by the system versus tasks requiring system-human communication; and (2) system attributes that principally require algorithmic interpretation versus those that require a high level of AI capabilities.