NASA’s Dr Steve Chien is helping shape AI research on easily the biggest open-source database imaginable.
A machine learning AI is only as good as the dataset that is provided to it, and there is no bigger open-source database than the entire universe. That is why Dr Steve Chien – head of all things AI at NASA’s Jet Propulsion Laboratory (JPL) – has arguably one of the most fascinating jobs in computer science.
In the 30 or so years that he has worked with NASA, Chien has been the recipient of a number of the space agency’s Exceptional Achievement Medals for the development of AI that has gone into some of its key systems and spacecraft. Notably, he was one of the key personnel behind the automated science scheduling before the launch of the European Space Agency’s famous Rosetta spacecraft, which travelled to Comet 67p.
Chien was one of the speakers at the recent National Analytics Summit held in Dublin by the Analytics Institute, where he revealed the work underway at the agency to use machine learning and other autonomous technologies to bring warp-like speeds to the rate of cosmic discoveries.
‘I will never say that in the near term we’re going to have anything that would approach having a scientist on board, but we’re making little steps in getting there’
– DR STEVE CHIEN
Speaking with Siliconrepublic.com, Chien was quick to emphasise that much of the work he does in AI isn’t immediately used for deep space missions, but rather a little closer to home. While they may not always be the operations to make headlines, JPL is heavily involved with the scheduling and logistical side of any space mission, particularly those providing crucial data about our changing planet.
“It’s tremendously complicated to bring a space mission together and actually operate a spacecraft,” Chien said. “I’ve often told people that my goal at NASA is to make AI so commonplace that people would ask, ‘Why aren’t you using AI?’, unlike the current situation where people would ask, ‘Why are you using AI?’.”
For JPL, its focus is on robotic space exploration, both in the present and near future, with existing and upcoming robots on the surface of Mars. However, it is also gazing further into the future with robots that could one day traverse beneath the icy depths of Jupiter’s moon Europa.
Finding ‘Swiss cheese’
One of the missions he highlighted was that of Deep Mars, developed by his colleague Dr Kiri Wagstaff. The AI that she and others developed at JPL has given Mars rovers the ability to learn what they are actually looking at, filter out what might be irrelevant and send back data on the interesting stuff. For example, Mars orbiters could be trained to find ‘Swiss cheese’ terrain – the result of icy holes seen on the planet’s surface – that could have frozen water beneath them.
But what happens when AI sees something it has never been trained to see? This, Chien said, is one of the fundamental challenges of AI for deep space exploration.
“I will never say that in the near term we’re going to have anything that would approach having a scientist on board, but we’re making little steps in getting there,” Chien said.
“An example of what we can now do is things like anomaly detection. The principles of this are very simple. You compute a baseline and you measure how far you are from the baseline. This is a very powerful technique.”
Lewis and Clark of the 21st century
In essence, JPL is training its AI to flag spots in imagery where it sees something it hasn’t seen before based on what it has learned already from thousands of hours of observations.
“We’ve actually experimented with this [concept],” Chien said. “We had a poor summer student taking images flying across the US in a commercial airline flight.
“You can then teach the system to cluster those images until you get clouds, mountains and rivers. So instead of sending down thousands of images, you say I went along this path and here are some examples of clouds.”
This technique isn’t perfected just yet, however Chien referred back to the expedition of Meriwether Lewis and William Clark in the Pacific north-west in the early 1800s as an example of how far exploration has come.
“They travelled for more than two years and they brought back a journal of 80 pages with drawings and notes, so we can’t complain at NASA. At least we have digital images and we have computers.”
Robot greeting party
An example of similar AI in action is when scientists previously had to get together and manually analyse hundreds of images taken by a Mars rover in order to find what looked like the best spot for further investigation. Now, thanks to the work of JPL, the Curiosity rover can autonomously choose a site, fire a laser at a rock and record the mineral composition of the sample to send back to Earth.
So with AI’s push into all things JPL and NASA unlikely to slow down, what does the future hold for its human workforce? More specifically, what about its human astronauts?
NASA has said it plans to set up a crewed base on the Moon and has its fingers crossed that astronauts will set foot on Mars. But what about Europa and other locations in deep space that seem beyond the limit of human space travel, something which is incredibly dangerous.
Is the human astronaut doomed? No, Chien said, as he sees them as being complementary rather than being in competition with one another.
“A robot has to be very smart to survive,” he said. “It has to be very smart to know what to look for [in places such as Europa]. Those are places where we can’t imagine sending humans, so we need robots to go there. To me, the ultimate quest is to go to another star. And, who knows, perhaps they’ll meet other robots sent by another species and they’ll have to converse on behalf of us.”
Updated, 3.54pm, 3 December 2019: This article was amended to reflect that Chien was working on the scheduling of Rosetta before its launch, not during the mission. It was also updated to give Dr Kiri Wagstaff her correct name and title.