Big Iron goes to Mars
This article originally appeared in the July 3, 2017 issue of SpaceNews magazine.
The idea of artificial intelligence (AI) in space is not new, but also has a bad reputation. The first thing that comes to mind to most people when they think about intelligent computers on spaceships is HAL, the malevolent, murderous computer from 2001: A Space Odyssey nearly half a century ago.
Grady Booch is well aware of that. “Let me assure you — and our PR people have me make this point — we are not building HAL,” the chief scientist of IBM’s Watson/M project said during a presentation at the Humans to Mars Summit in Washington in May.
Then he adds, in a stage whisper, “But we’re building something awfully close to it.”
What he and his team are building is a version of the company’s AI system that, in previous incarnations, defeated the world’s best in competitions as varied as chess and the game show Jeopardy!. IBM wants to move beyond answering trivia questions and similar projects to more complex analyses and interactions with people.
The concept has, of course, a buzzword: embodied cognition. “Imagine taking Watson and putting it into the physical world, and giving it, quite literally, eyes and ears and hands and feet, and act in the world,” Booch said. “It’s not something out there that you talk to, but participates in the world itself.”
Such a system, he said, would be more than just a system like Amazon’s Alexa or Apple’s Siri, able to respond to questions or requests. It would personalize its interactions based on the person with whom it’s interacting, and address issues with varying degrees of autonomy. “It must reason and learn,” he said.
A mission to Mars, he argued, is an ideal use for embodied cognition. “I can’t rely on mission control on Earth, but rather, I have to take mission control with me,” he said, because of the long communications delay between Earth and Mars. “What happens if, in the Orion spacecraft or in facilities on the ground, I had a cognitive assistant who was always there with me?”
That ends up with a system that, like he acknowledged, is something like HAL. “Consider what HAL did on that mission. HAL was in the walls; Watson will be in the walls. It was always embodied in the spacecraft itself. In a sense, HAL was the embodiment of that spacecraft, always listening,” he said.
Advances in AI make that science fiction from a half-century ago feasible today, and more. Booch envisions extending that AI from a spacecraft or base to what he calls cooperative robots, or “cobots,” that would work with and assist astronauts, such as when they venture across the Martian surface.
Kirk Bresniker has a similar vision of computers assisting human Mars missions, but a different technical approach. The chief architect and fellow at Hewlett Packard (HP) Labs is focused on the making the most of the exponential growth in data today, which he expects to continue with initiatives like the Internet of Things.
HP, like IBM, was once known for building “big iron” mainframe computers with state-of-the-art processors to handle complex tasks. But the challenge today, Bresniker said, is that the processing ability of modern computer systems can no longer keep up with this growth of data, as Moore’s Law reaches its physical limits.
“The incremental improvements we are seeing with our computing power will not meet the needs, the exponential demands, of our future challenges,” he said during The Humans to Mars Summit in Washington in May.
Bresniker and his group at HP have been working on an alternative approach, which they call “memory-driven computing.” This concept is intended to get around the processing bottlenecks created by dealing with large data sets by putting that data into memory that is far more readily accessible to a processor.
“For 60 years we’ve focused on a tiny bit of data running through a faster and faster calculator,” he said. “With memory-driven computing, we end the workarounds by inverting that problem.”
Such a system, Bresniker said, would be ideal for a Mars mission, for the same reason that Booch and his IBM group > are pursuing embodied cognition. “At 20 light-minutes away, Mars is too far to rely on communications from Earth for real-time support,” he said. “While ground control once helped guide Armstrong and Aldrin to the moon, the Mars missions and crew will be guided by a computer, capable of performing extraordinary tasks.”
Those extraordinary tasks, he said, would include monitoring spacecraft systems and identifying potential problems before they become serious. Similarly, the computer could monitor astronaut health, coordinate the large volumes of data coming from sensors on the Martian surface and in orbit, “combining and integrating these data sets to find those hidden correlations that can keep a mission and a crew alive.”
Such a system, Bresniker said, isn’t feasible with current computer technologies. “With today’s technology, we’d need a massive data center, attached to a nuclear power plant, to achieve this computing power that a Mars mission will demand,” he said. “What we’ve got today is just too big, too heavy, too slow, too inflexible and too power-hungry.”
Bresniker said HP has built what he believed was the most power memory-driven computing system to date, with 160 terabytes of data in memory. The concept could work with up to 4,096 yottabytes — 4,096 trillion terabytes — of data, which he said is about 250,000 times the amount of data in all computer systems worldwide today.
Both HP and IBM are funding these efforts on their own, as research and development projects with implications far beyond supporting human Mars missions. If an advanced computer system can support humans on Mars, they argue, it can do many more, and more lucrative, things on Earth.
“We began to realize that the characteristics of our mission to Mars were exactly the kind of challenge that would drive us,” Booch said, recalling discussions within IBM on embodied cognition projects. “It was that vision that grounded what we were trying to do.”
Booch said IBM was already working on those other applications for AI projects beyond how it could support a Mars mission. One example he showed was a similar computing system in a meeting, interacting with people working on a business deal. Instead of Watson in the walls of Orion, saving a Mars mission, it was in the walls of a conference room, saving a project.
Bresniker offered a similar explanation, likening a Mars spacecraft to a “smart city” embedded with sensors that could be processed by advanced computing technology.
“In short, the Mars spacecraft will need to be a smart city, an intelligent power grid, and an autonomous vehicle fleet all at once, and it will need to be controlled by a computer vastly more powerful than anything we’ve put into space,” he said.
Thus, a computer that could run a Mars spacecraft could also be sold to companies and governments to handle more terrestrial challenges.
Those systems will take advantage of the latest computing technologies, but also incorporate the lessons of science fiction, like HAL. “Obviously,” Booch said, “you want to avoid the homicidal features.”