In the long human effort to understand what lies beyond our world, NASA has turned to artificial intelligence to illuminate the moon's most hidden places — the permanently shadowed craters near its poles where sunlight has never reached and where water ice may have rested, undisturbed, for billions of years. Drawing on seventeen years of orbital data from the Lunar Reconnaissance Orbiter, AI foundation models are now mapping the lunar surface with a depth and speed no previous method could match. This is not merely a technical upgrade; it is a shift in how humanity learns from the cosmos — mac
NASA deploys AI to map moon's darkest regions and unlock lunar resources
Ice locked away for billions of years, waiting to fuel human return
Mark
So NASA is using AI to look at the moon. What exactly is it looking for?
Mimi
Primarily ice in the permanently shadowed regions near the poles. These areas stay cold enough to preserve water ice for billions of years, and if humans are going back to the moon, that ice is a resource—water for drinking and fuel for rockets.
Luke
But how confident are we that the ice is actually there? The AI is identifying where it *might* be stable, right? That's different from confirming it exists.
Mimi
True. The AI is mapping terrain and thermal properties to estimate likely locations. Confirmation would require actual landing and sampling.
Mark
How much data are we talking about here?
Mimi
Seventeen years of images from the Lunar Reconnaissance Orbiter—more data than all other NASA planetary missions combined. The AI was pre-trained on this massive dataset.
Luke
Pre-trained on unlabeled data, which means the system learned patterns without humans having to hand-label everything. That's efficient, but it also means we're trusting the model to find meaningful patterns. Did they validate it?
Mimi
Yes. They tested it against previous mapping methods and the AI matched or exceeded their performance.
Mark
What else can it do besides find ice?
Mimi
It identifies craters, measures them, spots young volcanic features, and analyzes thermal evolution of the surface. Basically, it's a general-purpose lunar analyst.
Luke
And NASA and IBM are planning to use this approach on other planets?
Mimi
That's the plan. They see this as a template for how to handle the data deluge from future space missions.
Mark
So this is really about scaling up how we do planetary science.
Mimi
Exactly. The bottleneck used to be human analysts. Now it's about training systems that can learn from one mission and adapt to the next.
Le Pouls
- The moon's permanently shadowed regions have long resisted understanding, holding potential water ice that could sustain future human life — but their darkness makes them nearly impossible to study through conventional means.
- NASA's AI models, pre-trained on datasets larger than all previous planetary missions combined, can now detect craters, volcanic features, and ice-bearing terrain that human analysts would take far longer to process.
- When tested head-to-head against older methods, the AI matched or exceeded their performance — signaling not just speed, but a genuine leap in analytical accuracy.
- NASA and IBM are already planning to extend this approach across the solar system, treating the lunar work as a proof of concept for a new era of AI-driven planetary science.
- As the Artemis program prepares to return humans to the moon, this AI infrastructure is becoming the foundation for choosing landing sites, locating resources, and anticipating hazards.
In the long human effort to understand what lies beyond our world, NASA has turned to artificial intelligence to illuminate the moon's most hidden places — the permanently shadowed craters near its poles where sunlight has never reached and where water ice may have rested, undisturbed, for billions of years. Drawing on seventeen years of orbital data from the Lunar Reconnaissance Orbiter, AI foundation models are now mapping the lunar surface with a depth and speed no previous method could match. This is not merely a technical upgrade; it is a shift in how humanity learns from the cosmos — machines absorbing the patterns of an alien world so that human minds can ask better questions of it.
NASA has deployed artificial intelligence to confront one of the moon's most persistent mysteries: what lies in the regions sunlight never touches. Using AI models trained on seventeen years of data from the Lunar Reconnaissance Orbiter — a dataset larger than every other NASA planetary mission combined — researchers are now mapping the lunar surface in unprecedented detail, with particular focus on the permanently shadowed craters near the poles where water ice may have been preserved for billions of years.
The approach relies on what scientists call foundation models: AI systems pre-trained on vast unlabeled datasets that can then be adapted quickly to specific tasks. A single underlying system can identify craters, detect young volcanic features, measure terrain irregularities, and estimate where stable ice deposits are most likely to exist — whether exposed on the surface or buried beneath the regolith.
The stakes are practical as well as scientific. If humans are to live and work on the moon, water ice represents both a life-sustaining resource and a potential source of rocket fuel. The AI helps researchers pinpoint the most promising locations by catching subtle variations in terrain and thermal data that human analysts might overlook or take far longer to assess. When tested against previous methods, the AI matched or exceeded their performance, suggesting the shift is not just faster but potentially more accurate.
NASA and IBM have announced plans to apply this same framework to other parts of the solar system, treating the lunar project as a proof of concept for a broader transformation in planetary science. As the Artemis program moves toward returning humans to the moon, this AI infrastructure will help planners navigate the challenges ahead — not by replacing human judgment, but by expanding the capacity to learn from the enormous streams of data that modern space exploration generates.
NASA is turning to artificial intelligence to solve one of the moon's most stubborn mysteries: what lies in the places sunlight never reaches. The space agency has deployed AI models trained on seventeen years of orbital data to map the lunar surface with unprecedented detail, focusing especially on the permanently shadowed regions where scientists believe water ice may be locked away for billions of years.
The foundation for this work rests on an enormous dataset. The Lunar Reconnaissance Orbiter, which has been collecting high-resolution images since 2009, has generated more information than every other NASA planetary mission combined. Rather than starting from scratch, NASA's AI models come pre-trained on vast unlabeled datasets, allowing them to absorb broad patterns and then adapt quickly to specific lunar research tasks. This approach—what researchers call foundation models—lets the same underlying system tackle multiple problems: identifying craters, spotting young volcanic features, measuring surface irregularities, and estimating where ice deposits might exist.
The permanently shadowed regions of the moon present a particular puzzle. These areas, concentrated near the lunar poles, remain cold enough to preserve ice indefinitely. Understanding where these ice patches are likely to be stable, whether on the surface or buried beneath regolith, matters enormously for future exploration. If humans are to establish a sustained presence on the moon, access to water ice could provide both drinking water and fuel. The AI system helps researchers narrow down the most promising locations by analyzing subtle variations in terrain and thermal properties that human analysts might miss or take far longer to process.
When NASA scientists tested the new AI approach against previous methods, the results validated the shift. The AI model matched or exceeded the performance of earlier techniques, suggesting that this is not merely a faster way to do the same work but potentially a more accurate one. The system can accelerate crater identification and measurement, tasks that are fundamental to understanding the moon's geological history and thermal evolution. By automating these analyses, researchers free themselves to focus on interpretation and strategy rather than the grinding work of manual mapping.
The implications extend well beyond the moon. NASA and IBM have announced plans to apply these same AI research methods to other parts of the solar system, treating this lunar work as a proof of concept for a broader transformation in how space agencies conduct planetary science. As human missions return to the moon under the Artemis program, this AI infrastructure will help planners identify safe landing sites, locate resources, and understand the hazards they will face. The technology is not replacing human judgment or scientific intuition; it is amplifying human capacity to process and learn from the vast streams of data that modern space exploration generates.
Citations marquantes
Foundation models acquire broad knowledge through pre-training, allowing them to generalize across multiple scientific domains through quick fine-tuning, making them versatile and efficient in accelerating scientific research— NASA
Permanently shadowed regions remain cold enough to trap and preserve ice for up to billions of years— NASA
Envie de l'histoire complète ?
Lire l'original sur FOX Weather ↗
Science & Space