Deep space changes the role of AI. Onboard Orion, AI is doing what was physically impossible in the Apollo era. ● Orion's computers run 20,000 times faster than the Apollo Guidance Computer and carry 128,000 times more memory. ● NEC's SIAT system analysed data from nearly 150,000 sensors during ground testing and established over 22 billion logical relationships between systems, a model that now underlies how future Orion vehicles are monitored. ● In flight, Reid Wiseman sits in the left seat not to fly manually, but to oversee automated systems and step in when human judgment is required. Then there is radiation. Beyond Earth's orbit, the planet's magnetic field no longer protects the crew. NASA is testing two models from the University of Michigan: a machine-learning model that forecasts radiation threats 24 hours ahead, and a physics-based model that simulates solar particle behaviour in the corona, where acceleration begins. NASA reserved 3,000 supercomputer nodes to run them, because a delay here is literally life-threatening. Also onboard is AVATAR. The experiment uses organ-on-a-chip devices containing bone marrow cells from the astronauts themselves, flying alongside the crew and exposed to the same deep-space radiation. After the mission, researchers will analyse how spaceflight affected those cells at the molecular level, data that will be critical for future Mars expeditions, where radiation doses will be significantly higher. This points to a central issue. A signal takes 1.3 seconds one way to reach the Moon. To Mars, between 3 and 22 minutes. The further from Earth, the more essential it becomes for onboard systems to decide autonomously. Every such decision on this mission is, in effect, training AI for a future where real-time control from Earth will simply be impossible. The directions ahead are clear: – AI monitoring of crew psychological state during long missions; – autonomous robots preparing lunar bases before crews arrive; – real-time analysis of geological data to optimise resource extraction at the lunar south pole; – orbital traffic management systems, developed by companies like LeoLabs and Neuraspace They are becoming critical infrastructure as orbit grows ever more crowded. Artemis II is a test not only of a rocket and a spacecraft. It is a test of a new principle: AI not as a support tool, but as part of the crew. And the further humanity flies from Earth, the more important that principle becomes. How much autonomy should AI really have onboard?
Jabe Bean
1mo
It’s hard to comprehend how much more efficient these systems will be now. Very cool!
Paige D.
1mo
Nice synopsis
Dara, your framing of Wiseman as supervisor rather than pilot is the most important line in this post, and it echoes a shift that happened in industrial controls decades ago. After 27 years commissioning autonomous systems in petrochemical and biotech plants, the pattern is identical: when communication latency makes human intervention impossible, you pre-authorize the system to act within defined parameters. In my line of work, we call these setpoints, alarm limits, and SIL ratings. NEC SIAT finding invariant relationships across 150,000 sensors is essentially alarm rationalization at cosmic scale. Same logic, different stakes. The question that sharpens yours: not how much autonomy, but who writes the envelope, and how does that envelope learn as the mission evolves?