AI’s next target could be NASA’s mission control

“There is a realization that the time to action is shrinking. Now it’s microseconds.”

In the iconic movie 2001: A Space Odyssey, astronaut Dave Bowman must deal with HAL 9000, a sentient artificial intelligence computer that operates his spaceship. The computer is all-knowing and all-controlling, saying at one point, “Dave, this conversation can serve no purpose anymore. Goodbye.” It portends a dark future for automated AI and space travel.

No one wants that outcome for real-world spacecraft and computers, but Hewlett-Packard Enterprise is starting to think about how to automate many spacecraft systems and outsource critical decisions to an on-board computer. Presently, with the International Space Station, flight controllers on Earth monitor the spacecraft’s overall health continually, and flight directors relay information to astronauts on board when problems occur.

But what happens when humans fly into deep space? Although astronauts are unlikely to venture beyond the Earth-Moon system before the late 2030s—if then—when they do, they will confront significant time delays. At some points of the orbits between Earth and Mars, the one-way time delay is as long as 20 minutes, making real-time conversations and problem-solving impossible.

For Kirk Bresniker, the son of an aerospace engineer who is now Hewlett-Packard Labs' chief architect, the answer to this problem may come in the form of more sophisticated artificial intelligence on board. He envisions an “introspective system” that constantly measures itself while looking for anomalies and coming up with fixes. “We’re talking about a much smarter spacecraft,” he said.

During the long journey to Mars, he said, astronauts will face periods of utter boredom. They will have sleep cycles. Accordingly, they will need an intelligent spacecraft that is capable of identifying problems and reacting to them by making course corrections or taking other measures within a few seconds. This is a computationally intensive solution to the problem that would require systems to continually recalculate and then act on those computations. But it's feasible, the computer scientist says. “It is a hardware and a software problem,” Bresniker said. “We need to employ all of the disciplines to realize the vision.”

Replace mission control?

Mission Control is a cherished institution within NASA and the broader US culture. From directing the Moon landings to saving Apollo 13 and through the space shuttle program, the men and women behind the consoles at Johnson Space Center are always watching, patiently to ensure the safest possible spaceflight. How would the space agency respond to automating many of the functions of mission control?

The truth is that NASA has been grappling with the correct amount of autonomy to grant crew members on missions that extend beyond the capability of real-time communications. And if this sounds otherworldly, it probably shouldn't. When Blue Origin begins to fly paying customers into space, perhaps next year, there will be no astronauts or pilots on board, just passengers. The computers will be flying. (Of course suborbital trips are simple compared to Martian flights, but automation in spaceflight is clearly the future.)

Bresniker said that Hewlett-Packard has had ongoing discussions with many national agencies, including NASA, about how machine learning algorithms can manage large volumes of information and make for smarter cities, more efficient power grids, and enable autonomous vehicles. “There is a realization that the time to action is shrinking,” he said. “Now it’s microseconds.”

Hewlett-Packard believes its hardware, and that of other providers, is going to advance far enough to enable all of this. On Tuesday, as part of its “The Machine research project,” Hewlett-Packard Enterprise introduced what it says is the world’s largest single-memory computer. The prototype contains 160 terabytes of memory, which Hewlett-Packard claims is capable of simultaneously working with the data held in every book in the Library of Congress five times over.

Bresniker said this represents the leading edge of exascale high-performance computing, and that within about five years, supercomputers used for science and engineering will operate at greater speeds than the sum of the Top 500 supercomputers today—and at a small fraction of the power. For spacecraft, he acknowledged, the computers will have to be significantly lighter than today’s supercomputers and consume vastly less power.

Ultimately, he said, Hewlett Packard doesn’t want to completely replace flight directors and flight controllers—or the astronauts inside the spacecraft. “We’re not designing humans out of the loop,” he said. “Humans have full forensic capability, and we are designing systems that are fully informing us of their rationale. This is still a human-led mission.”

Until the HAL 9000 comes online, that is.