Pentagon ramps up artificial intelligence efforts

by

This article originally appeared in the April 23, 2018 issue of SpaceNews magazine.

The tech industry has software tools that can pore over a thousand images per minute and identify objects with 99-percent accuracy. The Pentagon badly wants this technology, and sees it as essential to military combat capability as bombs and missiles.

“How can I leverage the innovation happening in AI?” asked Gen. Stephen Wilson, vice chief of staff of the U.S. Air Force.

The military and intelligence agencies have sought to bring artificial intelligence into areas like data mining and autonomous vehicles. But the demand is outpacing the supply. Human analysts are overwhelmed by the information fire hose of drone footage and satellite imagery, and machines have shown they can perform better than people in many tasks.

“An intelligence analyst gets it right 75 percent of the time,” Wilson said during a recent event hosted by New America, a Washington think tank funded by Google.

Staff Sgt. Ashlie Robledo and Senior Airman Thao Bui, 11th Special Operations Intelligence Squadron analysts, participate in a data-tagging training event last August. Credit: U.S. Air Force photo by Senior Arirman Lynette M. Rolen/DOD
Staff Sgt. Ashlie Robledo and Senior Airman Thao Bui, 11th Special Operations Intelligence Squadron analysts, participate in a data-tagging training event last August. Credit: U.S. Air Force photo by Senior Arirman Lynette M. Rolen/DOD

The Pentagon has reached out to companies like Google for help connecting the military to the AI world. Under an initiative called Project Maven, the Defense Department and the Air Force are funding the development of AI algorithms to analyze drones’ live video stream. Eric Schmidt, the executive chairman of Google’s parent company Alphabet, chairs the Defense Innovation Board. The panel of private sector and government adviser has been hugely influential in raising awareness within the Pentagon of the capabilities of AI.

The partnership has gotten off to a rocky start, however. Project Maven has come under fire from the tech community. As The New York Times reported this month, more than 3,100 Google employees signed a letter urging CEO Sundar Pichai to consider ending the relationship with the Pentagon, as “Google should not be in the business of war.”

Diane Greene, head of Google’s cloud operation who sits on Alphabet’s board of directors, told the tech website The Verge that the AI project is not to “operate or fly drones” and “will not be used to launch weapons.”

Wilson defended Project Maven as an important research effort. “We use it to ID objects in full-motion video,” he said. “This is not weaponizing. We are trying to take repetitive task that are routine, like processing of pictures, and automating them.”

The military worries that it is falling behind in the application of AI while China continues to invest billions of dollars, said Wilson. “We are going to have to compete to win.”

Defense Secretary Jim Mattis told lawmakers this month that the Pentagon plans to consolidate many disjointed AI projects from across the military into a central program office. The reorganization will be led by Undersecretary of Defense for Research and Engineering Michael Griffin.

“We’re looking at a joint office where we would concentrate all of DoD efforts, since we have a number of AI efforts underway right now,” said Mattis. “We’re going to move things into production, prototyping,” he said. “We’re not going to have more papers, we’re going to move.”

“We’re not going to have more papers, we’re going to move.” James Mattis, U.S. Secretary of Defense Credit: U.S. Air Force photo by Wayne A. Clark
“We’re not going to have more papers, we’re going to move.”
James Mattis, U.S. Secretary of Defense Credit: U.S. Air Force photo by Wayne A. Clark

Defense procurement chief Ellen Lord said AI is key technology priority that will require DoD to “leverage across programs” that already exist but do not necessarily share information or resources. “We have talked about taking over 50 programs and loosely associating those,” Lord told reporters. “That’s our challenge at DoD. We have many example of silos of excellence.”

Griffin announced April 13 at a Hudson Institute event that he will oversee a “Joint Artificial Intelligence Office” and possibly stand up a similar organization for hypersonics. These offices, he said, will not replicate traditional joint program offices like the one that oversees the F-35 Joint Strike Fighter.

The new AI office will bring in “elements of the intelligence community,” Griffin said. But many details remain to be worked out, such as “how we would create it, where we would locate it, who would head it, who would participate in it.”

Griffin has been insistent that projects need to move fast. “The U.S. has been on holiday for 25 years,” he said.

So far much of the military-focused AI activity has been led by the Pentagon’s Silicon Valley technology outreach office, the Defense Innovation Unit Experimental.

DIUX and the National Geospatial-Intelligence Agency (NGA) in February launched a competition that could help advance AI big-data capabilities. Participating companies were given government satellite imagery to develop machine-learning algorithms. The so-called xView Detection Challenge was promoted to “advance key frontiers in computer vision and develop new solutions for national security and disaster response.”

DIUX officials said this was one of the largest publicly available datasets of overhead imagery.

Companies in the AI and intelligence sectors called this a smart move by DIUX and NGA. To develop algorithms that are credible, companies need quality data, and the government offering it at no cost is huge, said Brady Cline, vice president of defense and intelligence at SpaceKnow.

“Training data is extremely valuable,” he told SpaceNews. “If we have to buy it from DigitalGlobe or Airbus, or create data, it is really expensive,” he said. “DIUX experiments democratize the data.”

Offering training data is “crucial to make this kind of thing successful,” said Derek Edinger, vice president of strategic partnerships and co-founder of Ursa Space. “Machine learning applied to satellite imagery is getting a lot of hype these days. The tricky part is getting training data to calibrate the algorithms.”