As NASA sends robotic spacecraft and rovers to the moon and Mars, the space agency has been using artificial intelligence to get the most science out of its missions.
Enterprises could learn a lot from those efforts.
"The more complicated missions get and the farther away spacecraft get, the harder it gets for the normal ways of doing business," said John Bresina, a computer scientist in the Intelligent Systems Division at NASA's Ames Research Center. "A.I. is one of the ways you can deal with that issue."
When NASA launched a robotic probe, the Lunar Atmosphere and Dust Environment Explorer, also known as LADEE, in the fall of 2013, there were three different science teams competing for time to make their own observations during its 100-day mission. The tactical planning was overwhelming.
Sharon Gaudin/Computerworld
John Bresina is a computer scientist in the Intelligent Systems Division at NASA's Ames Research Center.
To plan what science would get done, when each observation would be made and to catch errors, like flight rule violations or to keep one observation from overriding another, Bresina said he knew they were going to need more than human schedulers.
The A.I. system was ground-based, not embedded in the spacecraft's software, and helped NASA scientists send commands to the LADEE.
The flight controllers used the A.I. system – dubbed LASS for LADEE Activity Scheduling System -- to construct and send commands to interact with the lunar spacecraft.
"We needed help planning what science would get done," Bresina told Computerworld. "Around a dozen people used it for the whole mission, which was from October 2013 to April of 2014. We used it from launch to impact, but the most intense use was during that 100-day science phase. We planned all the maneuvers in all the science passes. Everything we did, we used this system.
"It just seemed obvious that we needed something like [A.I.]," he said.
Stephen Smith, a research professor focused on A.I. at Carnegie Mellon University, said artificial intelligence is a great tool for solving particularly complicated scheduling problems, such as NASA's.
"Humans, when manually scheduling, become overwhelmed with the scale of it," Smith said. "We need to rethink A.I. It can be a powerful amplifier of human-decision making without taking over the decision making."
Smith noted that during NASA's Jupiter flyby mission, there was a two-week window when the spacecraft would take photos and make other observations about the planet. Fitting all of that information into such a small window created a scheduling nightmare.
"Something like 300,000 man hours went into building that two-week schedule to squeeze as much science as they could out of it," he said. "That was all done with some kind of software, but a lot of it was manual. If they had used artificial intelligence, they could have been doing other things with that time."
For Dan Olds, an analyst with The Gabriel Consulting Group, there is a big lesson here for enterprise IT and executives.
"These AI systems can take many more factors into account than humans, and find overlaps and inconsistencies much faster and more accurately than an often over-stressed human in a hurry," he added. "What NASA's experience shows is just how far these programs have come and how trusted they are."
A.I. for many missions
Bresina wasn't new to A.I. and he wanted to use it on as many NASA missions as possible.
The computer scientist worked on the missions that sent the Mars rovers Spirit and Opportunity to work on the Martian surface. The mission teams for both rovers, which were launched in 2003, used an A.I. system called MAPGEN or Mixed Initiative Activity Planning Generator.
The ground-based decision support system was designed to act as a reasoning tool to detect flight rule violations and help fix them.
MAPGEN is a mixed initiative planner, which means a human is in the decision-making loop. A human can have input in scheduling and problem solving, and is in control of the process, while the artificial intelligence is doing the heavy lifting.
Some of the guts of that A.I. system have made it onto the LADEE system, as well as about three other space missions, including the Mars rover Curiosity, and a few Earth-based missions, like NEEMO, a research mission that sends astronauts, engineers and scientists to live in an undersea research station.
"We built it to make it general enough to apply to all these different missions," said Bresina. "For LADEE, we wanted people to be able to use it without having an expert sit beside them and hold their hands. In previous missions, that had not been the case. There was often somebody at least on hand in case something went wrong with the system or somebody needed advice. … I wanted everybody to be able to operate the system without having somebody fly out there to help them."
Bresina said he's not sure if the A.I. system he's worked on will be used with the super Mars rover that is expected to launch in 2020.
"It's one of the tools available at NASA to be used," he explained. "So far, the Jet Propulsion Laboratory, Johnson [Space Center] and the Ames Research Center have all used it. There are other tools around, as well, but this is one they can choose."
Lessons for the enterprise
For Oren Etzioni, CEO of the Allen Institute for Artificial Intelligence, there is a big message for the enterprise coming out of NASA's work with A.I.: It's not just for Mars, anymore.
"Because of the constraints of communications in space, NASA has always been a pioneer with A.I. technologies," Etzioni said. "The enterprise is coming to the same realization with the explosion of data. Companies need to yield more automated systems with appropriate oversight … It can be challenging to take corporate knowledge and embed it or build it into A.I., but once you do, it's in there. It's not like one person quits and all their knowledge is gone."
Jeff Kagan, an independent analyst, said artificial intelligence will help companies create order in their massive data stores, as it also helps them schedule and oversee their planning.
"Enterprises should keep a close eye on others who are successfully using A.I., like NASA," said Kagan. "A.I. is still a long-term process and we are still in the very early days of its evolution. What is learned at NASA will next be used at companies. I think we will see enterprises increasingly depend on A.I. as time passes."
Olds agreed, noting that A.I. is perfectly geared at making sense of complex tasks.
"For example, a simple A.I. can load an airplane with the right amount of passengers, luggage and freight to ensure that the plane is fully loaded to the maximum," he explained. "However, a better A.I. may realize that this is the last flight out of that city for the day, that there is a snowstorm predicted overnight, and that with this in mind, it should raise the price on last-minute booking passengers and priority freight."