Integrating Artificial Intelligence in Human-Rated Spacecraft Systems for Long Duration Spaceflight Missions

Abstract

As human spaceflight missions extend toward deep-space destinations, artificial intelligence (AI) is increasingly suggested as a means to mitigate communication delays, workload, and operational complexity. However, the role of AI in high-stakes, crewed environments remain insufficiently studied from the perspective of end users (crew members). This study reports results from a voluntary survey of a specialized population (N=123), with 25 respondents representing approximately 20% of the target population. Participants included a majority with astronaut/analog astronaut experience and baseline familiarity with AI systems. Survey results indicate a strong preference for human-AI collaboration over full autonomy, with no respondents favoring fully autonomous decision-making. AI operating as an “Assistant” under human supervision was consistently preferred, though acceptable autonomy levels varied by operational scenario. Respondents placed increasing emphasis on near-perfect reliability as scenario severity increased, strongly valued explainability and transparency, and viewed always-available human override as essential. Trust in AI was moderate rather than absolute, and respondents expressed discomfort with AI making critical decisions without human involvement. Risk tolerance was highly task-dependent, as minimal risk was acceptable for life-critical systems such as life support and navigation, while greater risk was tolerated for scientific experimentation. Notably, all respondents expressed a desire to participate in the development and training of AI systems they would use operationally. When compared with existing literature, these findings reinforce emerging consensus that AI for spaceflight should be designed as a bounded, transparent, and human-centered collaborator rather than as an autonomous replacement. The results support the need for graded assurance approaches to AI verification and validation, functional fitness assessments tailored to mission context, and segmentation of the spaceflight landscape by subsystem criticality. Collectively, this work provides grounded guidance for aligning AI system design, assurance, and deployment strategies aligned with operator expectations for future deep-space missions.