Intelligent Pilot Advisory System: The journey from ideation to an early system design of an AI-based decision support system for airline flight decks
Authors: Jakob Würfel, Boris Djartov, Anne Papenfuß, Matthias Wies
Abstract: The availability of more and more data from commercial aircraft , opens up new development potential for AI-based assistance systems on the flight deck . In this context, a number of research areas need to be addressed, including the interaction between AI and the cockpit crew. Designing systems which can facilitate this new interaction between the crew and the AI merits exploration.This concept paper discusses the need for an AI-based decision support system in the cockpit of commercial aircraft. An initial proposal is made on how such a system can be developed to address human-machine interaction aspects of implementing AI-based systems in the cockpit.To discuss the need for such a system, interviews with pilots, as well as videos of emergency situations from simulator-sessions were analyzed. Based on the findings, potential use cases and research topics were identified and a system design for a selected use case was developed.The interviews have shown that crews would like to be assisted by a system during high workload events, such as emergencies and abnormal situations. According to the interviews, one of the most important demand is assistance in the decision-making process. Situations where the crew need to abruptly select an alternate airport due to a technical failure or unforeseen changes at their destination, are of particular interest. In such situations, pilots must collect and interpret information and discuss options for action. The system requested should help pilots improve their available information, situation analysis and option generation. Based on the interviews, a use case and a basic system design will be outlined, and finally two approaches for implementing the system will be proposed, focusing on separate research topics . One approach develops the system from the end-user's perspective while the other examines it from the AI development perspective.For the first approach, the challenge of designing an interface between the crew and the artificial intelligence is discussed. One requirement is that the system present the information and generated options in a comprehensible way so that the crew can understand them quickly. One possible approach would be the research area of explainable AI, which attempts to explain to the user how the system computed the proposed solution. However, there is also the approach that domain-specific, more interpretable Human-AI-Interfaces is the better solution for end-user friendly systems. In this paper, we propose a human centered approach to develop an AI-based decision support system so that the AI-based information is presented in an understandable and interpretable way.Second ly, we present some additional details regarding the AI core. The focus here would be on how this system might be envisioned and which contemporary methods from the field of AI may be employed. Given the nature of the system the Learning Classifier System (LCS) will be discussed as one of the most suitable models for the AI core, as it is characterized by a more human understandable approach. Additional benefits of LCS’s will also be presented and compared with the most currently relevant models from the field of AI and machine learning.
Keywords: Human AI Interaction, Human Systems Integration, Systems Engineering, Artificial Intelligence, Interpretable AI, Aviation
Cite this paper: