Personalized and transparent AI support for ATC conflict detection and resolution: an empirical study

Artificial Intelligence provides both opportunities and considerable challenges to the continued growth of Air Traffic Control (ATC) services. This paper presents a study where a personalized and transparent machine learning decision aid for ATC conflict resolution was built and empirically evaluated with air traffic controllers. Multi-site simulations were conducted with 34 controllers working together with an AI agent to solve conflicts between aircraft in enroute traffic scenarios. Resolution advisories varied in conformance (degree of personalization) and transparency. Main effects of conformance were found on controllers’ resolution performance and response to advisories in terms of acceptance and ratings of agreement and similarity to own solution. The separation distance aimed for by the advised solution was found to be particularly important for the response to optimal advisories. More positive responses were measured for controllers whose separation margin preferences was closer aligned with the advisory. The study provides the aviation community with knowledge on how conformal and transparent AI support systems affect operators’ responses to system-generated resolution advisories.