AI/ML for safety-critical software: the case of the space domain

Spacecrafts increasingly rely on software to fly and talk to Earth. Using Artificial Intelligence and Machine Learning for safety-critical software in space brings advantages and new challenges. We analyze the current way of producing safety-critical software and the reference safety and assurance standards in the domain, namely, ECSS-Q-ST-80C and ECSS-E-ST-40. Then, we explore the readiness of correct practices to ensure that ML- or AI-enabled systems are safe and reliable, and we discuss new practices and methods that should be introduced. The analysis refers to the different criticality classes safety-critical software can have.