How NASA uses AI on Mars
By Nurfilzah Rohaidi
Interview with Tom Soderstrom, IT Chief Technology Officer, Jet Propulsion Laboratory, NASA.
Image: NASA/JPL-Caltech/MSSS
But it’s not alone out there. Artificial Intelligence on this plucky little robot helps the humans back at mission control to analyse and take pictures of interesting rocks that may contain water. Then, the humans can make an informed decision to send Curiosity back to those rocks to analyse them, says Tom Soderstrom, IT Chief Technology Officer at the Jet Propulsion Laboratory (JPL).
There is applied AI - or “built-in intelligence” - in practically everything that JPL does, Soderstrom says. “If you think about it as an intelligent assistant or digital assistant, that’s the entry point into going much deeper,” he tells GovInsider on the sidelines of the recent AWS re:Invent conference in Las Vegas, Nevada.
Digital assistants in space
These digital assistants help the humans to identify anomalies on orbiting spacecraft, for example, through miniscule fluctuations in the data that are sent back to Earth. “We already have our data and we know the problems that people found; now we want to find the problems that people didn’t find,” Soderstrom says.
One particular benefit is that digital assistants can identify these problems with equipment on the spacecraft in real-time. “Now, all of a sudden, we can be proactive and see problems occurring, especially if there is a problem going on with the radar or something similar,” says Soderstrom.
He hopes that one day, “every scientist and engineer at JPL” will have Siri-esque digital assistants chugging away in the background, shifting through petabytes of data so the humans won’t have to.
These assistants will also become much more sophisticated over time: “In the past you were typing. Today, you’ll be speaking, you’ll be using a touch screen, you’ll be blinking through your augmented reality glasses, and soon, thinking,” Soderstrom proclaims.
Solving Mars problems on Earth
Constant innovation is the name of the game for JPL, which has set aside a space, its innovation experience centre, precisely for this. Scientists, engineers or businesspeople work with JPL on problems that they want to solve, and create prototypes together. “We show it to the people who are important, and if they get excited about it or have ideas, we continue. If they don't, we stop it,” Soderstrom explains.
And some of these ideas have actually made it onto the surface of Mars. The Curiosity rover withstands icy cold temperatures on the red planet, but its instruments may only operate within a certain temperature range. Previously, experts would take weeks to pore over pictures of screenshots of temperature gauges to determine the best time to use these instruments. As a result, Curiosity “gets very little drive time on Mars per day”.
Soderstrom’s team worked with end users and IT specialists at the innovation centre on a solution that could display constantly-updating temperatures for each of Curiosity’s instruments. “Could we use the same idea as you do on the stock market, where you plot stock A against stock B and you just change it?” he explains.
JPL took data from various rover systems and put in on the cloud, where the humans could quickly visualise and interact with billions of data points and determine the exact times of the day that they could carry out experiments on the rover. This new capability, called telemetry analytics, has directly led to more drive time for Curiosity, Soderstrom says. “We have used that on every mission since,” he remarks.
He adds that the planned Mars 2020 rover will get even more drive time, as telemetry analytics will be carried out in the cloud everyday. “By doing that, we don’t miss any days on Mars.”
Iterating towards perfection
As the CTO - or Chief Toy Officer, as he likes to call himself - Soderstrom has the responsibility of testing, experimenting and prototyping new technologies and capabilities for JPL. “Today’s toy is tomorrow’s tool,” Soderstrom says, but “you don't know which toy is going to become a tool until you try them, and play with them.”
This rapid prototyping approach allows JPL to quickly “iterate its way towards perfection”, making mistakes along the way but at very little cost. “Once we land on Mars, it has to be perfect,” he notes, “but in the beginning, we can’t think of everything - so we experiment.”
“Once we land on Mars, it has to be perfect, but in the beginning, we can’t think of everything - so we experiment.”When it comes to AI, Soderstrom advises governments to explore the use of digital assistants in the form of chatbots “right away, because it has immediate impact and is not that hard to do”. They are “sometimes not very good”, he points out, but indeed one easy way to get immediate feedback to quickly improve digital services.
The trick is to make them “very narrow in scope”, Soderstrom says, and once natural language processing and text analytics improve, widen it. “You have a digital assistant, for instance, just to answer questions about travel, and another one to answer question about security. Don't mix them, because it’s not ready yet,” he explains.
Mobile services are another area for governments to focus on, he says, and so is a “focus on small, rapid experiments of new things, new capabilities”.
On Mars, Curiosity trundles on, determined as ever - with a little help from her AI assistants. Here on Earth, however, a ‘fail fast’ and ‘can do’ attitude towards innovation can be the key to governments leaping light years ahead.