Toward Safe Human–AI Interaction on the Flight Deck: Preliminary Requirements for Single-Pilot Operations in Commercial Aviation
The integration of Artificial Intelligence (AI) into flight deck systems is a potential opportunity for aircraft systems development needs, but also a big challenge due to the need for development of new layers of protection and regulations assurance. Although single pilot operations are already available for general aviation on small aircraft, the commercial air transport category requires more rigid rules and the replacement of second pilot for this category is still challenging. No certified AI-based system currently exists to support a single pilot operation, even more in commercial air transport operations. Also, no regulatory framework has been established for the development and certification of such systems. This thesis proposes a preliminary set of requirements for an AI decision-support system (called AI-SPO system), considering the operation as an assistant for the single pilot in commercial air transport category. The research adopts a Design Science Research (DSR) methodology. First, a systematic literature review following PRISMA guidelines was conducted as exploratory research to identify the core themes of research regarding human systems integration and AI on the flight deck, where research on single pilot operations was identified as one with more recent studies and feasible for defining a set of preliminary requirements. The requirements were derived mainly from EASA and FAA regulatory roadmaps and literature review, being organized into nine main categories: system level, pilot authority, operational design domain, functional behavior, explainability, interface, workload management, training and organizational requirements. These groups were categorized based on EASA AI development levels 1B and 2A. The main results demonstrate that the AI-SPO system should function as a structurally advisor, permanently subordinate to the pilot authority, whose integration into the flight deck must be conditioned on system’s ability to reduce cognitive workload and increase situational awareness. The SysML development was performed in Astah SysML, covering the requirement diagrams, Block Definition Diagram, Use Case and sequence diagrams, and by the end a verification plan of seven test cases. The resulting requirements were justified and demonstrated traceability between architecture, behavior scenarios and verification. The contribution is positioned as a preliminary design framework that can serve as a basis for engineering design and certification once official regulatory requirements for AI systems on the flight deck are established.