Explainable AI (xAI) and Autonomous Systems for Persistent Human-Machine Operations in Space, on the Moon, and on to Mars
B. Danette Allen · 2025
With NASA’s Artemis missions, humans will again set foot the cratered surface of Earth’s nearest neighbor, exploring more of the Moon than before as we establish the first long-term persistent presence on the Moon and learn to live and work on its surface. We will use what we learn on and around the Moon to take the next giant leap: sending the first astronauts to Mars. During missions in orbit around and on the Moon, explorers will conduct science investigations and test technologies needed for a persistent presence on the lunar surface and future human missions to Mars. These explorers will be human, or they may be robotic agents exploring on behalf of humankind. To achieve the persistence, pace, and performance we expect and need, these agents require intelligence and agency in decision-making so that, instead of relying on humans for direction, they team with humans to achieve common goals. These autonomous capabilities must be reliable enough to perform mission tasks, run spacecraft systems, and make changes to operations without input from humans or external information when necessary. Perhaps more critically, these autonomous systems will be responsible for maintaining living environments when crew are not present and guaranteeing health and safety when they are. These systems of systems that comprise the lunar infrastructure ecosystem present complexity never-before instantiated off the Earth’s surface. There is no canonical definition of “complexity” but an informal one that is especially useful for our purpose here is a “state in which the [system] is difficult to understand and verify-- and an increase in risk”. This difficulty can be due to any number of factors including but not limited to under-determinism, non-determinism, incorrect information, stale information, etc. all of which are rooted in uncertainty about the system state, inputs, outputs, or environment. Confidently accepting that these systems are making good decisions means understanding risks and quantifying the uncertainties that produce those risks.