Deep space missions face significant communication delays that disrupt both operational workflows and psychological support for crew members. Unlike low Earth orbit operations, delays ranging from several minutes to nearly an hour make real-time communication with mission control infeasible, forcing crews to act with greater independence under uncertain conditions. This position paper examines how human-in-the-loop AI, digital twins, and edge AI can be integrated to mitigate these delays while maintaining astronaut autonomy and engagement. We argue that human-in-the-loop AI enables decision-making processes that are responsive to local context while remaining adaptable to changing mission demands. Digital twins offer real-time simulation and predictive modelling capabilities, allowing astronauts to explore options and troubleshoot without waiting for ground input. Edge AI brings computation closer to data sources, enabling low-latency inference onboard spacecraft for time-critical decisions. These ideas are explored through two use cases: using deepfakes to support emotionally resonant communication with loved ones, and applying visual-language models for onboard fault diagnosis and adaptive task replanning. We conclude with reflections on system design challenges under constrained and high-stakes conditions.