A probe could wake itself when something interesting happens
Across the vast distances of space, humanity's robotic emissaries still think with the minds of the 1990s — not out of neglect, but out of hard-won wisdom about what survives the void. Scientists like Victoria Da Poian are working at the edge of this paradox, training artificial intelligence to one day grant probes the autonomy to make decisions faster than any signal from Earth could carry. The gap between what AI can do and what space hardware currently allows is not a failure of imagination, but a reminder that in exploration, reliability is its own form of courage. New processors promising a hundredfold leap in computing power may soon close that distance, opening an era where machines in deep space learn to think — and act — for themselves.
- NASA's deep-space probes still run on 30-year-old PowerPC 750 chips, chosen because radiation-hardened reliability matters more than raw speed when a mission costs hundreds of millions and cannot be rebooted.
- AI has not yet operated in space — autonomy remains a promise rather than a reality, with Earth-based scientists still issuing every meaningful command across communication delays that can stretch to hours.
- When a probe loses contact with Earth and enters sleep mode, it can miss fleeting events like volcanic eruptions on distant moons — entire scientific opportunities erased by hardware and protocol constraints.
- NASA's new partnership with Microchip Technology targets a processor 100 times more powerful than current space-grade chips, with lab tests already showing speeds 500 times faster than existing hardware.
- Researcher Victoria Da Poian demonstrated that AI can distinguish biotic from abiotic chemical signatures that appear identical to human eyes — a capability that could redefine how life is searched for across the solar system.
- Even as the technology matures, Da Poian insists that discovery will remain a collective act — no single measurement, no single algorithm, but teams of scientists converging on truth together.
There is a peculiar contradiction at the heart of modern space exploration: while AI transforms life on Earth, the probes NASA sends to distant worlds still think with processors from the 1990s. Victoria Da Poian, a data scientist collaborating with NASA's Goddard Space Flight Center, has spent years working inside this tension — applying AI to missions like ExoMars and Dragonfly, which is bound for Saturn's moon Titan in 2034.
For Da Poian, the promise of AI in space reduces to one word: autonomy. The current process is unchanged — scientists on Earth make a decision, transmit it to the probe, and repeat. AI could break that cycle. But the transformation has not yet arrived. Some automation exists, like the Perseverance rover selecting its own landing site to avoid rocky hazards, but fundamental decisions still originate from Earth. And even as autonomy grows, Da Poian warns, rigorous verification will be essential to ensure independence does not endanger the mission.
The hardware explains much of the caution. NASA's probes rely on chips like the PowerPC 750, chosen for radiation resistance and proven reliability rather than speed. Launches are extraordinarily expensive, and no one risks the unknown without cause. Yet change is approaching: NASA recently announced a partnership with Microchip Technology on a next-generation processor offering up to 100 times more computing capacity, with lab tests clocking it at 500 times the speed of current space-grade hardware.
The stakes are concrete. Probes receive only megabytes of data from Earth, and when communication drops, they sleep — missing events like volcanic eruptions on distant moons entirely. A probe with genuine autonomy could wake itself at the critical moment, choose what to observe, and decide what to transmit home. That capability, Da Poian explains, is what would allow humanity to truly maximize what a probe can learn.
Her own research centers on mass spectrometry and the detection of life's chemical signatures. At a recent conference, she showed two spectra — one from a living process, one from a non-living one — that appeared identical to human observers. AI told them apart instantly. Yet she remains measured about what that means. False positives happen, algorithms err, and any genuine discovery would demand years of collaborative validation. 'It's never a single measurement,' she says. 'We work as a team.'
Dragonfly itself is not a life-detection mission but an exploration of Titan's chemistry and physics — a world with a dense atmosphere, methane lakes, and a rain cycle that mirrors Earth's own, substituting methane for water. It launches in 2028 and arrives in 2034. Until then, Da Poian and her colleagues continue building and testing algorithms on Earth, preparing for the moment when machines in deep space will finally be trusted to think for themselves.
There is a peculiar contradiction at the heart of modern space exploration. While artificial intelligence reshapes everything from smartphones to automobiles, the robotic probes NASA sends to distant planets still run on processors built in the 1990s. Victoria Da Poian, a data scientist and software engineer at Tyto Athene LLC who collaborates with NASA's Goddard Space Flight Center, has spent years grappling with this tension. She works to give space probes greater autonomy, applying AI to missions like ExoMars, which searches for signs of life on Mars, and Dragonfly, bound for Titan, Saturn's largest moon, in 2034. When asked why this gap exists, the answer is both practical and humbling.
For Da Poian, the real promise of AI in space comes down to one word: autonomy. "It's necessary if we want to explore farther, beyond Earth, the Moon, or Mars," she explains. "Right now the process is always the same. Scientists make a decision on Earth, send it to the probe, and repeat. AI could change that." But there is a surprise waiting for anyone who assumes this transformation has already happened. AI is not yet operating in space. It remains a project, not a reality. Some automation exists for specific tasks—like the Perseverance rover's real-time selection of its landing site on Mars, where it chose terrain that would avoid rocky hazards. But the fundamental decisions still originate from Earth. And even when probes gain more autonomy, Da Poian cautions, extensive verification and validation work will be required to ensure that greater independence does not jeopardize the mission itself.
The reason for this caution lies partly in the hardware. NASA's space probes still rely on processors developed in the 1990s, such as the PowerPC 750, chosen not for speed but for their proven reliability and resistance to radiation. "You don't risk something new without knowing it works, and launches are extremely expensive," Da Poian notes. The old saying holds: a winning team doesn't change its equipment. Yet change is coming. This month, NASA announced a partnership with Microchip Technology on the High Performance Spaceflight Computing project, which is developing a new processor promising up to 100 times more computing capacity than current space-grade chips. Testing in Los Angeles showed the processor operating 500 times faster than existing hardware. The goal is to allow probes to use AI to respond in real time to complex situations without waiting for human intervention from Earth.
The hardware limitation creates a cascading problem. Probes can receive only a few megabytes of data from Earth—not gigabytes, not terabytes, but megabytes. When a probe loses communication with Earth, it enters a sleep mode to conserve power. If a volcanic eruption occurs on a distant moon during that dormant period, the probe misses the opportunity entirely. "We want to maximize the time a probe is active on another planet to collect as much data as possible," Da Poian explains, "but we also have to ensure we can send that data back." This is where AI becomes essential. A probe equipped with autonomous decision-making could choose what to observe, when to activate its instruments, and what data to transmit home. It could wake itself during a volcanic event and take measurements at the critical moment—something impossible under current Earth-dependent protocols.
Da Poian's own work focuses on detecting signs of life using mass spectrometry, a technique that analyzes the chemical composition of samples. AI enters the picture to identify patterns invisible to human eyes. During a presentation at the Data Makers Fest in early May, she displayed two spectra—one from an abiotic process, one from a biotic one. To the human eye, they appeared identical. AI distinguished them immediately. "That's where it makes the difference," she says, "detecting things we cannot." Yet she remains cautious about the implications. False positives occur. Algorithms still make mistakes. Any genuine discovery would require years of collaborative validation among teams. "It's never a single measurement," she emphasizes. "We work as a team. It's not 'I found this, that's the answer.' It's 'I found this, you found that, let's talk.'"
The Dragonfly mission itself is not framed as a search for life. It is an exploration mission designed to understand the chemistry and physics of Titan, the only moon in the solar system with a dense atmosphere, methane lakes, and a rain cycle—all mirroring Earth's water cycle but with methane instead. "It's more like detective work," Da Poian says. "We want to understand the environment and perhaps say: these processes could be biotic or abiotic." Dragonfly launches in 2028 and arrives at Titan in 2034. Until then, Da Poian and her team continue building and testing algorithms on Earth, preparing for a future where space probes think for themselves.
Notable Quotes
You don't risk something new without knowing it works, and launches are extremely expensive— Victoria Da Poian, data scientist at Tyto Athene LLC
We want to maximize the time a probe is active on another planet to collect as much data as possible, but we also have to ensure we can send that data back— Victoria Da Poian
The Hearth Conversation Another angle on the story
Why does NASA still use chips from thirty years ago? Aren't there better options available now?
The short answer is cost and certainty. A space mission costs hundreds of millions of dollars. You don't gamble on new hardware when you have processors that have been proven to work in the harsh radiation environment of space. The PowerPC 750 has a track record. A newer chip might be faster, but if it fails halfway to Mars, you've lost everything.
But surely modern chips are more reliable than they were in the 1990s?
They are more reliable in normal conditions. The problem is space isn't normal. Radiation damages electronics in ways we can predict for older chips but not always for newer ones. The testing and certification process for new hardware in space takes years. By the time you've validated it, you're already committed to using it for the next decade of missions.
So AI in space is still mostly science fiction?
Not fiction, but not deployed yet either. We're using AI for specific tasks—like helping the Perseverance rover choose where to land. But the big decisions, the real autonomy, that's still coming. It requires both better hardware and years of validation to make sure we're not putting missions at risk.
What changes when you have a faster processor?
Everything becomes possible that isn't now. A probe could wake itself up when something interesting happens. It could analyze data in real time instead of waiting to send it back to Earth. It could make decisions about what to observe and what to ignore. Right now, if a volcano erupts on a moon while your probe is sleeping, you miss it. With autonomy, you don't.
Is there a risk that AI makes the wrong call?
Absolutely. That's why validation matters so much. We're not going to send a probe to another planet and let it make life-or-death decisions without extensive testing. But the potential gain—being able to respond to unexpected events in real time—is worth the effort to get it right.
When do you think we'll actually see AI making decisions in space?
The new processors are being tested now. If they work as promised, we could see them on missions in the next five to ten years. But even then, the autonomy will be limited and carefully controlled. It's not about replacing human judgment. It's about extending human reach to places where the speed of light makes real-time communication impossible.