Situated Visual Alarm Displays Support Machine Fitness Assessment for Nonexplainable Automation

Determine if situated visual alarm displays can support machine fitness assessment (MFA), facilitating improved hazard recognition and alarm accuracy assessment in the presence of inaccurate alarms. Poor performance of opaque automation is more difficult to detect, which increases the likelihood of...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on human-machine systems 2022-10, Vol.52 (5), p.984-993
Main Authors: Rayo, Michael F., Horwood, Chelsea R., Fitzgerald, Morgan C., Grayson, Marisa R., Abdel-Rasoul, Mahmoud, Moffatt-Bruce, Susan D.
Format: Article
Language:English
Subjects:
Alarm design
Alarms
Artificial intelligence
Automation
computer interface
Data visualization
design display principles
Displays
Emergency response
explainable artificial intelligence (AI)
Fitness
human systems integration
Man-machine systems
Nurses
Physiology
Qualitative analysis
Task analysis
trust in automation
Visual signals
Visualization
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Determine if situated visual alarm displays can support machine fitness assessment (MFA), facilitating improved hazard recognition and alarm accuracy assessment in the presence of inaccurate alarms. Poor performance of opaque automation is more difficult to detect, which increases the likelihood of cascades resulting in overall system failure. MFA reduces the negative impact of poor automation performance. Integrated alarm visualizations were shown to 32 nurses for 10 cases focused on patient outcome and 17 focused on alarm quality, all using real patient data. Five of the ten outcome cases would ultimately result in an emergency (unbeknownst to the nurse). Alarm cases ended with a true, false, or unnecessary alarm. Responses for nurses' concern, confidence, alarm quality, and intended response were recorded. Qualitative analysis of interviews was performed. Using the situated visual alarm displays, nurses reported less confidence (6.5 vs. 9.1, p < 0.001), more concern (5.4 vs. 1.6, p < 0.001), and more urgent responses for emergency cases. Their alarm event detection was better than the alarms' detection (0.608 vs. 0.438, p < 0.001), as was their interpretation accuracy (0.453 vs. 0.243, p < 0.001). Nurses showed differentiated concern for emergency cases, nonemergency cases with alarms, and those without alarms (5.4 vs. 3.8 vs. 1.6, p < 0.001). Situated visual alarm displays combining visual trends with alarm signals improves detection of hazardous events and mitigates the negative effects of poor opaque automation performance.
ISSN:2168-2291
2168-2305
DOI:10.1109/THMS.2022.3155714