Will intelligent AI agents take on the job of capsule communicator in future missions to the moon, Mars and other space destinations?
It could happen, says James Burk, the executive director of the Mars Society.
“One of our advisers did a really deep dive on how the Apollo astronauts interacted with each other and with the CapCom back on Earth, and he came to the insight that the Apollo 17 astronauts were using CapCom almost like an AI bot — because the CapCom knew everything,” Burk said during a panel discussion focusing on the intersection of artificial intelligence and space ventures.
“You can imagine having an AI edge device which could be like a rover following the crew around, walking around the moon or Mars,” he said. “It’s watching them and taking stock of how everyone’s doing.”
Tuesday’s panel was a crossover session presented at Madrona Venture Labs by the Washington Technology Industry Association for Seattle AI Week, and by Space Northwest for Seattle Space Week. “When you think about the kinds of megatrends of our time, two of the big ones are space and AI,” said Mike Doyle, Space Northwest’s president and co-founder.
Putting AI into space adventures isn’t exactly a new idea: The best-known sci-fi example is HAL, the AI who goes psycho in “2001: A Space Odyssey.” There’s also the no-nonsense computer voice in the Star Trek saga, or Marvin the Paranoid Android in “Hitchhiker’s Guide to the Galaxy.”
But the real world isn’t science fiction. Yet.
Dealing with data
“I don’t think we’re going to see a ‘HAL’ business,” said Keith Rosema, a partner at Madrona Venture Labs who has previously worked at NASA’s Jet Propulsion Laboratory, Jeff Bezos’ Blue Origin space venture and the late Paul Allen’s Vulcan Inc.
Instead, AI is helping humans make sense of the flood of imagery and other data coming down from Earth orbit. Kelsey Doerksen, a Ph.D. student at the University of Oxford who’s affiliated with JPL, said one of her jobs is to “make space scientists’ life easier” — specifically when it comes to doing atmospheric science.
“When we’re trying to run physics-based models, with hundreds if not thousands of different parameters and various tweaks of how you could initialize your parameters and these physics models, it takes hours, days, weeks to run these types of models to get results out,” she said. “Whereas with the AI pipeline that we’re building at JPL, we can do things in a matter of hours.”
Hanna Steplewska, the president of Seattle-based Cognitive Space, said AI-driven software tools are making headway in the commercial space industry. For example, a search engine called Danti is optimized to sift through Earth observation data.
Multiple companies — including BlackSky, which has deep roots in Seattle — employ AI to help government and commercial customers make sense out of a variety of geospatial data, ranging from satellite views to social media. Microsoft and the Allen Institute for AI have also gotten into geospatial data analysis.
Training for space
Steplewska’s company is focused on applying AI tools to the thorny problem of tracking thousands of satellites in low Earth orbit.
“Cognitive Space has a very clear 10-year goal,” she said. “We want to orchestrate a million intelligent machines across a multiplanetary system. So, everything that we’re learning about how to fly constellations effectively applies to constellations of things that are on Earth, on the moon’s surface, in orbit around the moon, in orbit around Mars, on Mars’ surface and beyond.”
What about generative AI, which has quickly revolutionized so many tech sectors? When it comes to space operations, one of the big technical hurdles has to do with the fact that large language models really don’t know that much about the final frontier. Burk recalled a test case that the Mars Society ran, in which ChatGPT was asked to design a valve for a zero-pressure, high-altitude balloon.
“The answer it came back with was factual … but it was totally wrong,” he said.
Doerksen said satellite constellations could provide “the perfect use case” for training better AI models and automating space operations. “If you’ve had the same satellite launched in 2015, and a similar generation in 2022, you can use that historical data to train a model to still be used in the future,” she said.
The AI revolution isn’t just affecting space operations on Earth: The Seattle area’s biggest players in AI and cloud services — Amazon Web Services and Microsoft Azure — are working with partners to expand edge computing to the edge of space. This year, a Seattle-area startup called Lumen Orbit came out of stealth with a plan to send hundreds of data-center satellites into orbit. The idea is to run data through AI models in space, and then downlink only the most valuable bits.
“You know, there was a Super Bowl commercial for Salesforce recently that said something like, ‘If AI is the Wild West, isn’t big data the new gold?'” Burk said. “I think our approach to get ready for AI at the Mars Society, with our scientific research, is to really be thoughtful about how we’re collecting data, to have standards where they don’t exist.”
Power tool? Space pal? Or HAL?
In the years to come, maybe AI will just blend into the woodwork — or more accurately, the silicon and steel — of space infrastructure.
“I will hold back from saying [that] in 10 years we’ll have an AI overlord,” Rosema said. “In all seriousness, I actually think this might be more boring. Right now, AI is very much in our face. And if I look at other historical technical trends — internet, mobile phones — originally, those things were all very much in our face. I hope that AI does the same thing: It just melts into the background and becomes another power tool for us.”
But maybe space-based AI will become more than a power tool. During NASA’s uncrewed Artemis 1 mission, which circled the moon in 2022, Amazon teamed up with Lockheed Martin and Cisco to put an Alexa-type voice assistant inside the Orion capsule. During future missions to deep space, a smarter version of the assistant could keep a spacecraft’s crew up to date on what’s happening around them — and, in the words of an Amazon executive, provide “some form of companionship.”
Hmm … Burk’s AI CapCom might not be such a flight of fancy after all. Let’s just hope it doesn’t end up turning into HAL.
Other highlights from Seattle Space Week:
- Recruiting talent for space projects can be a challenge. “As a VC-backed startup competing with Google for data scientists, talent is really hard,” Steplewska said. But Burk said “it’s been really easy for me to recruit talent, because there’s a lot of tech people who are interested in space.” AI projects are among the top priorities for the Mars Technology Institute that the Mars Society is setting up, potentially in the Seattle area.
- During a Monday session in Redmond, Wash., a panel of entrepreneurs weighed in on other challenges related to sustaining a startup in the aerospace industry. Bill Bruner, co-founder and CEO of New Frontier Aerospace, said he was counting on Congress to pass a bipartisan bill providing additional tax relief for research and development. “This is really an existential issue for startup companies,” he said.
- Seattle Space Week continues tonight with a Space Happy Hour at the Doubletree Southcenter. The invitees include participants in this week’s State of the Space Industrial Base workshop at Seattle’s Museum of Flight. Check the event’s registration page for additional information.
- Space Happy Hour is organizing a night at the ballpark for Thursday’s game between the Seattle Mariners and the Chicago White Sox at T-Mobile Park. Check out the “Spaceball” webpage for ticket information.
