Performance, Trust, and Workload in an Automation-aided Visual Search Task

Samuel Monfort

Advisor: Patrick E McKnight, PhD, Department of Psychology

Committee Members: Tyler Shaw, Ewart De Visser

David J. King Hall, #2064
October 27, 2017, 12:00 PM to 02:00 PM

Abstract:

Identifying vehicles on the battlefield quickly and accurately is an important part of soldier performance. Currently, automation is being developed to aid in target identification efforts, but some ambiguity remains regarding the accuracy required for these systems to be helpful. Past meta-analytic efforts have identified a 70% reliability threshold between when automated systems help performance and when they interfere with performance. However, this threshold was calculated as an aggregate estimate from a great variety of studies, and warrants further exploration before being applied to a target search and identification context. Therefore, this dissertation was designed to identify moderators specific to target search and identification that might shift the 70% reliability threshold. I found that the type of error issued by the automation (misses versus false alarms), the range of the targets (close versus far), as well as the type of judgment required of the soldier (search: vehicle/no-vehicle versus identification: type of vehicle), all affect the reliability required for automation to have a net-positive effect on performance. Although these moderators have strong relationships with target search/identification performance, they have less of an impact on subjective trust and workload, suggesting that observers might not be consciously aware of how their own performance is changing as a function of automation properties. These results are discussed in light of the way that soldiers are trained and how automation is designed to aid performance on the battlefield.