Imperfect Diagnostic Automation: An Experimental Examination of Priorities and Threshold Setting
DOI: 10.1177/154193120605000301
archive: indexed pipeline: cataloged
Abstract
Diagnostic automation, such as alarms, alerts, or automatic target recognition systems can vary in their reliability and threshold setting, the latter influencing the balance between misses and false alarms. This experiment examined the implications of both of these, when the system detected military targets in parallel with the human operator in an unmanned air vehicle simulator. Unlike previous investigations of this paradigm in a dual task setting, the automation diagnosis task was very difficult, and priority between this task and a concurrent task was explicitly varied, along with the threshold setting. The results revealed that: (a) reliability as low as 0.6 aided human performance. (b) the effects of the automation threshold shift could be partially modeled in terms of the classic reliance-compliance dimensions of automation dependence, at both task priority settings. (c) The priority effects were more pronounced with miss-prone automation. (d) directing attention to false-alarm prone automation actually degrades human-system performance.
Access
Route: Publisher paywall (check institutional access or ResearcherGate)