The proliferation of Artificial Intelligence, particularly Machine Learning and Deep Learning, has reached the space sector, given the method's high computational performance. However, the integration of AI-based systems in spacecraft operations or onboard spacecraft has not moved further than being researched. One of the reasons is that many non-AI-expert users state that the complexity and black-box behavior is leading to a lack of understandability and trust in such AI systems. Hence, for the end-users, in this case, the spacecraft operators, the utilization of ML and DL models comes with the uncertainty of how those models operate and make decisions. To build trust, AI-based concepts must be evaluated based on their explainability and interpretability, their computational performance and robustness, as well as their testing-concept and general verifiability, considering both the needs and knowledge of the developer and the end-user.
The aim of this study is to understand and showcase the rationales of why \ac{AI}-based space systems are not trusted and what tools, techniques, or guidelines and standards need to be established to increase their usability and enable the utilization of AI-based space systems outside of the research field. For this purpose, a questionnaire was conducted among space operations professionals of the German and European space ecosystem to understand the concerns and apprehensions towards the deployment of AI-based space systems. The preliminary results of the evaluation of the questionnaire together with indications on concepts and processes for increased explainability of \ac{AI}-based space systems are presented in this study. The focus is laid on users and applications in the field of spacecraft operations.
«The proliferation of Artificial Intelligence, particularly Machine Learning and Deep Learning, has reached the space sector, given the method's high computational performance. However, the integration of AI-based systems in spacecraft operations or onboard spacecraft has not moved further than being researched. One of the reasons is that many non-AI-expert users state that the complexity and black-box behavior is leading to a lack of understandability and trust in such AI systems. Hence, for the...
»