Towards a unified taxonomy for algorithmic transparency

Abstract

With the rapid advancement of drone technology, their applications have become increasingly widespread. However, the integration of drones into the airspace also poses risks to crewed aircraft, particularly around airports. To address this issue, a highly automated Uncrewed Air Traffic Management (UTM) system is being developed. Since fully safe and reliable automation does not exist yet, UTM still requires human supervision to enhance the overall system safety and reliability. Some form of “seeing-into” transparency may be necessary to help operators better understand the limitations and behavior of the automated UTM system. As UTM is a novel concept, research on transparent UTM is limited. Many efforts have been made in other fields, but there still remains a lack of consensus on what transparency entails, particularly for algorithmic systems. Therefore, this article first presents a unified taxonomy for algorithmic transparency, with operational, domain and engineering transparency introduced as its core concepts. From the taxonomy, twenty UTM transparency elements and their corresponding visual prototypes were then designed, which also showcases how the taxonomy can be applied in practice. A survey-based user study was conducted to collect the opinions of air traffic controllers and drone experts regarding the designed elements and prototypes. Results indicate that transparency is deemed imperative for UTM, especially in scenarios featuring automation failure. It also reveals that operational transparency is generally preferable over engineering transparency in nominal operations. Participants were asked to group the designed elements, and their results closely aligned with the structure of the proposed taxonomy.