Is this project an undergraduate, graduate, or faculty project?
Graduate
Project Type
group
Campus
Daytona Beach
Authors' Class Standing
Aarohi Srivastava, Graduate student
Lead Presenter's Name
Aarohi Srivastava
Lead Presenter's College
DB College of Arts and Sciences
Faculty Mentor Name
Briana Sobel
Abstract
The integration of artificial intelligence (AI) into aviation decision support systems (DSS) introduces new opportunities for pilots. While these systems offer considerable benefits, ensuring safe and optimal human-AI interaction requires further research. In the case of student pilots, trust calibration, system transparency, and appropriate reliance are all factors that influence the operational success of an integrated AI system. The systems are designed with the goal of decreasing cognitive workload, increasing situational awareness (SA), and providing data-based recommendations to help the pilot make quick and accurate decisions. This study will evaluate the effects of disclosed vs undisclosed AI accuracy on student pilots’ decision-making during simulated diversion tasks. Participants (student pilots) will be given a sample flight map, an emergency scenario, and three potential landing sites, of which they will be required to select the optimal choice as determined by a subject matter expert. One of these options will be designated as the AI recommendation, with accuracy (high, 90%; low, 60%) as the within-subjects variable. There will be two groups: disclosed and undisclosed accuracy in which participants may or may not be made aware of the system’accuracy rate. Scales will be used to measure workload, trust, SA, and decision accuracy. This study is in progress, but it is anticipated there will be a significant effect of automation level (high accuracy vs low accuracy) and AI disclosure on pilot decision accuracy, trust, SA, and workload. The higher accuracy AI recommendations should result in better performance and reduced cognitive workload. The disclosed reliability should improve calibrated trust with moderate acceptance, not overtrust. Additionally, this condition should show higher accuracy, SA, and moderate workload levels. Overall, the findings of this experiment will provide insight into pilots’ formation of mental models and trust calibration when working with AI systems, further suggesting methods of improving human-AI collaboration.
Did this research project receive funding support (Spark, SURF, Research Abroad, Student Internal Grants, Collaborative, Climbing, or Ignite Grants) from the Office of Undergraduate Research?
No
Artificial Intelligence Decision Support and Pilot Performance
The integration of artificial intelligence (AI) into aviation decision support systems (DSS) introduces new opportunities for pilots. While these systems offer considerable benefits, ensuring safe and optimal human-AI interaction requires further research. In the case of student pilots, trust calibration, system transparency, and appropriate reliance are all factors that influence the operational success of an integrated AI system. The systems are designed with the goal of decreasing cognitive workload, increasing situational awareness (SA), and providing data-based recommendations to help the pilot make quick and accurate decisions. This study will evaluate the effects of disclosed vs undisclosed AI accuracy on student pilots’ decision-making during simulated diversion tasks. Participants (student pilots) will be given a sample flight map, an emergency scenario, and three potential landing sites, of which they will be required to select the optimal choice as determined by a subject matter expert. One of these options will be designated as the AI recommendation, with accuracy (high, 90%; low, 60%) as the within-subjects variable. There will be two groups: disclosed and undisclosed accuracy in which participants may or may not be made aware of the system’accuracy rate. Scales will be used to measure workload, trust, SA, and decision accuracy. This study is in progress, but it is anticipated there will be a significant effect of automation level (high accuracy vs low accuracy) and AI disclosure on pilot decision accuracy, trust, SA, and workload. The higher accuracy AI recommendations should result in better performance and reduced cognitive workload. The disclosed reliability should improve calibrated trust with moderate acceptance, not overtrust. Additionally, this condition should show higher accuracy, SA, and moderate workload levels. Overall, the findings of this experiment will provide insight into pilots’ formation of mental models and trust calibration when working with AI systems, further suggesting methods of improving human-AI collaboration.