PANEL DISCUSSION: Challenges of AI Certification

Abstract

The utilization of artificial intelligence (AI) is growing throughout all domains. There is strong demand for the use of AI in the aviation domain. With AI, there is a huge potential for automating complex tasks, e.g., vision-based tasks to support the landing operation. The overall goal is to improve the performance of operations, and increase the number of operations, while at the same time reducing the cost of operation, and also maintain and increase safety [1]. This implies following rigorous certification standards. This is a challenge, because traditional standards and processes for certification do not fit the new paradigm of data-driven designs of AI components [2]. In traditional software, the behavior is determined by software design and software algorithms. This enables a coherent and bidirectional traceability from requirements, over design, to software source code, executable code and tests [3, 4]. However, with AI systems the behavior is determined to a large degree by the data that was used for training. Using different data results in different behavior. As a result, the same rigor that is used for software development is also required for the data handling. The requirements for data quality, the characterization of the data, as well as the overall development lifecycle needs to be considered [5, 6]. This is just one of the challenges that emerge with the AI software. This panel is targeted towards the certification challenges of AI systems for aviation. The panel will invite speakers on the topic of AI certification and the challenges regarding that topic. Targeted topic areas include: gaps of existing standards, standardization groups and efforts towards new standards, research perspectives on trustworthy AI, industry perspectives on AI utilization, safety monitoring and assurance of AI systems, environmental conditions and operational design domain of AI systems, and more. The authors of this abstract will be the moderators of the panel. As invited panelists, we are proud to present our highly valued guests: Guillaume Soudain (EASA), Burak Ata (Helsing GmbH), Konstantin Dmitriev (TUM), and Umut Durak (DLR). We kindly refer to the biography for further information.