Exclusive Source for Defense Industry News and Military Technology

BREAKING NEWS !

Updated! - ARMY 2019 EXPO in Pictures

INTERNATIONAL MILITARY-TECHNICAL FORUM «ARMY-2019» exhibit between June 25-30 at PATRIOT EXPOCENTER _____________________________...

Could artificial intelligence save the Pentagon $15 billion a year?

The first time Chinese government officials came to visit Tom Siebel in hopes of convincing him to work on artificial intelligence projects it was 2011. Then delegations of mayors and senior leaders came back in 2012, and in 2013, and in 2014.



U.S. Air Force aircraft maintainers perform post-flight maintenance on an E-3 Sentry AWACS. (Senior Master Sgt. Roger Parsons/U.S. Air National Guard)




Siebel is a businessman based in Silicon Valley. He founded business application software company Siebel Systems in 1993, sold it for $5.9 billion in 2006, and in 2011 started another company: C3IOT, which is known as C3.ai.


An artificial intelligence company specializing in predictive analytics, C3.ai has helped Baltimore Gas Electric save $20 million annually by checking the health of the utility’s meters and reducing the amount of unbilled energy usage. The company’s web site also boasts testimonials from Fortune 500 giants, such as 3M and Shell. Since its founding, C3 has invested more than $500 million in a software stack that performs artificial intelligence tasks.


Because of all that, the Chinese teams wanted C3.ai to help with the country’s Smart Grid project, a $250 billion program to transform how China uses and distributes electricity. Condoleezza Rice, the former secretary of state currently on C3.ai’s board of directors, told Siebel to “run, don’t walk,” he recalled.


“I actually had very candid conversations with them when they were here,” he said. “I simply said, ‘I don’t understand how this could possibly work. I’m in the business of developing intellectual property, and you’re in the business of massive, state-sanctioned intellectual property theft. I mean, how are we ever going to find a business model that works for both of us?’”


This could have been the end of the story. C3.ai’s founders hadn’t expected to work in the national security sector in their first years. In 2015, however, shortly after former Secretary of Defense Ash Carter established the Department of Defense’s Silicon Valley outpost known (at the time) as the Defense Innovation Unit-Experimental, Pentagon representatives contacted Siebel. Like their Chinese counterparts, they too wanted help with artificial intelligence. Ninety days later, they had him on contract.

Former Secretary of Defense Ash Carter (shown in 2016) formed Defense Innovation Unit-Experimental as a way to embrace Silicon Valley innovation. (Senior Master Sgt. Adrian Cadiz/Office of the Secretary of Defense)


How people view this story is often colored by their view of how the Defense Department works, a kind of Rorschach test on Pentagon acquisition.


Maybe it’s evidence, perhaps unfair, that the Chinese government is at least four years ahead of the United States in how it thinks about artificial intelligence. Maybe the Pentagon is taking steps to improve how it uses artificial intelligence and is ready to fully take advantage of the expertise on the subject that now resides in Silicon Valley. Maybe this is work the nation’s largest defense contractors could already perform and haven’t been asked.


Siebel has his own interpretation.


To comprehend how Siebel, 66, an expert salesman and an adventurer who was once attacked by an elephant, first attracted the attention of the Pentagon requires understanding the Defense Innovation Unit.


Carter founded DIUx (now, initial experiments proven successful, just DIU) in 2015 with the thinking that the technology and the innovation the Pentagon needs for future battles won’t exclusively come from the defense industrial base around Washington, D.C. Instead, the Defense Department needed to attract new businesses, specifically those tech and IT companies in California, to bid on its work.


Today, DIU’s headquarters are housed in a two-story brick Army building that sits in the shadows of NASA’s Ames Research Center and its giant wind tunnels. On the inside, the meeting rooms carry Star Wars references.


Since its founding, occasionally lawmakers in Washington have questioned the success of DIU as a host of other organizations focused on future technologies have popped up. DIU received $30 million in 2016 and $10 million in 2017. The White House’s budget request for fiscal 2020 asks for the organization to receive about $30 million for each of the next five budget cycles.


But sources in the Silicon Valley tech community said shortly after DIU’s arrival that the Pentagon staff was a walking example of what’s described as “tech tourism.”


That pejorative refers to the regular occurrence of officials, usually from the East Coast, wandering the offices of upstart IT companies with open office plans and casual dress codes to talk about what it could mean for their agencies or businesses, but without investing or signing on to become a customer.


More recently, the organization has won over skeptics with a series of low-cost, high-profile contracts. In the cyber realm, the team has set up contracts with security firms, including one with Recorded Future, to learn more about cyber intrusions.


Previously, federal agencies only had access to a small percentage of threat intelligence, but by purchasing information DIU was able to better understand open-source data associated with threat actors and shrink research times. The contract was only a few million of the Pentagon’s $700 billion budget but was also used to incentivize other organizations to buy the information as well.


In the software arena, leaders at DIU helped open Kessel Run, an Air Force laboratory that has reimagined software development.


And one of its biggest successes, the example often cited by leaders in case studies and congressional hearings, has been with predictive maintenance and artificial intelligence.

Air Force maintenance teams check a tire as part of a routine inspection. (Senior Airman Mya M. Crosby/U.S. Air Force)


Predictive maintenance is far from the science fiction future of robot overlords or Terminator vs. Iron Man debates popular in national security circles. The idea calls for essentially identifying system failure before it happens in order to repair before things breaks.


Analysts at Global Market Insights guess the predictive maintenance and machine inspection segment in AI in the manufacturing market (not the defense market) is projected to grow by 44 percent annually between today and 2025.


Now consider that the average aircraft in the Air Force is about 28 years old and on any given day about 30 percent of the aircraft cannot deploy. The service wants to shrink those numbers.


C3.ai’s first project with the Air Force looked at the E-3 Sentry, an aircraft that provides the service surveillance, command and control. The Air Force has roughly 40 of those aircraft, which were first designed in 1958. But C3.ai had no telemetry data to help predict when something would go wrong.


“All we have is basically unstructured data in textual form,” Siebel said. “These consist of maintenance logs, which tend to be pretty messy and undecipherable. They also had flight logs, pilot logs and records from the maintenance system that might include oil inspections.”


C3.ai ran all that information through a natural language processing program, took the text and extracted bits of information and stored it on Amazon Web Services’ GovCloud. The challenge was that not every piece of information is valuable.


“Temperature doesn’t mean much, but high temperature might be meaningful. Pressure doesn’t mean much, but low pressure does. So, we were able to find those signals that correlated with system failure,” he said. “We built the algorithms, designed the user interface, built the application, loaded all the data and trained the users. And brought the application live in six months.”


Siebel is proud of the timeline.


“No chance [the Air Force gets] its [request for proposals] out in six months. Zero chance … This is a testament to DIU; I mean it wasn’t a big battle, it was a couple million bucks, right? [For] a fraction of the cost it takes to draft and evaluate RFPs, we delivered a production system,” he said.


“The last time the United States Air Force had a system in production out in six months would be World War II, right?”


Also, it worked. Within six months, C3.ai had demonstrated a reduction of as much as 28 percent of unscheduled maintenance events for the E-3, according to a DIU fact sheet.


Then the Pentagon awarded a similar contract for the C-5 Galaxy. Again, it worked. Siebel’s team completed the project in six months. Then awards for the F-15, F-16, F-18 and F-35 joint strike fighter.


The work attracted attention in the Pentagon, including from Gen. Paul Selva, vice chairman of the joint chiefs of staff and the Pentagon’s No. 2 uniformed officer, as well as Heather Wilson, the secretary of the Air Force.


In March, Michael Brown, the managing partner at DIU, told Congress that his organization is working with the Pentagon’s Joint Artificial Intelligence Center to scale this type of predictive maintenance approach across multiple aircraft, as well as ground vehicles beginning with the Army’s Bradley Fighting Vehicle.


“This is one of DIU’s highest priorities for FY19 given its enormous potential for impact on readiness and reducing costs,” he said.


Now Siebel and C3.ai are doubling down.


While a few high-profile companies faced protests over working with the Defense Department, Seibel and C3.ai quietly suggested one of the Pentagon’s most ambitious artificial intelligence programs to date.


In late 2018, C3.ai made an unsolicited proposal to bring the entire Air Force fleet on to the system. Siebel said Wilson’s staff estimates it could lead to a 40 percent increase in aircraft availability. (C4ISRNET could not confirm that figure.) DIU’s estimates say the department could save $15 billion annually if the program was scaled to DoD’s aircraft fleet.


“Forty percent!” Siebel said during an interview in his office in Redwood City, California. “Oh my God, if you’re secretary of defense, and you have 40 percent more stuff than the day before, and what’s a new F-35 cost? $100 million?”


(It costs about $89 million.)


But here’s another way of thinking about it: $15 billion is at least four times the Pentagon’s spending on AI programs alone for fiscal 2020. It’s about 1.5 times what the Pentagon spends on netcentric warfare and the same as the entire Pentagon portfolio in space.


National security experts suggest that the program could have wider implications.


Kara Frederick, an associate fellow for the technology and national security program at the Center for a New American Security, said the real value in C3.ai’s work with the Air Force may be cultural.


“This collaboration, this idea of consistent data standards, the idea that you take the algorithm and upload it to the cloud and then it can be used to cross different components of DoD. It’s that muscle memory that you’re creating with this initiative that could be a bigger deal,” she said.


“This can help create that environment conducive to the behaviors that you’re seeking to encourage for more efficient data processing, efficient use of these functions. … Instead of coming into varied bespoke systems, if you think about plug-and-play architecture, this is starting the ball rolling with some of those behaviors that are going to be used for a long time to come.”


In addition, if C3.ai secures a contract for the Air Force’s entire fleet, it would likely mark one of the Pentagon’s largest AI contracts to date. If such a contract reached tens of millions of dollars it could help entice other tech companies, including those in Silicon Valley, to reconsider doing business with the Pentagon, Frederick said.


“A couple million dollars? That’s nothing. … I think a big number will raise some eyebrows in a good way and cause companies to say now the economic risk maybe worth it,” she said. “Frankly, the smaller, single-digit million dollars contracts are not going to make it worth it to some of these companies just for the privilege of working with the Department of Defense.”


Pat Kumashiro, a retired Air Force colonel who led the maintenance division for the service’s logistics, engineering and force protection directorate and is currently director of the Air Force market at not-for-profit consulting firm LMI, said companies must demonstrate deep functional expertise in the area. For example, are there maintenance chiefs on staff who understand the inspection logs?


“Clearly the pressures will be to go faster and improve scale. It’s hard to slow down this fast-moving train,” he said. “What all of us want as taxpayers want is that they are not going to take unacceptable risks.”


The day I spoke with Siebel in February was the day after the Trump White House announced the nation’s AI strategy and on the day the Department of Defense released its own long-awaited strategy on the topic. “Implementing predictive maintenance and supply” is one of four examples the Pentagon uses to discuss AI.


“We will use AI to predict the failure of critical parts, automate diagnostics, and plan maintenance based on data and equipment condition. Similar technology will be used to guide provisioning of spare parts and optimize inventory levels. These advances will ensure appropriate inventory levels, assist in troubleshooting, and enable more rapidly deployable and adaptable forces at reduced cost,” it reads.


The Pentagon strategy also formally notes the work with C3.ai.


Yet, Siebel sees predictive maintenance as a bridge to broader artificial intelligence applications. Most media attention in artificial intelligence has focused on science fiction’s idea of killer robots. And perhaps rightfully so. During the tenure of former Secretary of Defense Jim Mattis, the Pentagon chief put a premium on “increasing lethality” for every budget item. Siebel suggests AI systems can lock onto a target 1,000 times faster than a human.


“You get into cyberwarfare? That’s entirely an AI problem,” he said. “We get into space command, these are AI problems. Now, will robotics play a factor? Absolutely. What do you want to send across a minefield? Do you want to send an infantryman? I mean, I don’t think so. Who do you want to send into some cave in Afghanistan? Why would you want to send a human being when you could send a machine? I would say that these UAVs, these are robots, basically, right? Some are remotely controlled, but why put people, human beings, in jeopardy when you can just have a machine blow up? And so, you lose a little bit of metal and maybe it costs you, who cares how much it costs. It’s just metal.”


Finding targets is a touchy topic.


When they talk about AI, national security leaders have spent much of their time talking about Project Maven, an Air Force program that worked with Google to use artificial intelligence and machine learning to identify objects in drone imagery. The project inadvertently became the posterchild for the Pentagon’s efforts on AI in Silicon Valley.


While the program received mainstream attention, in large part because Google contractors backed out of the project, folks in the national security community have stressed it was a pathfinder.


(Siebel says Maven was “a silly little trivial prototype. There was nothing to it. It was a silly project.”)


As for his own systems, he says he will always build a system where humans are in the loop to make decisions, not robots or computers.


“I’ve never encountered anybody in the Pentagon or in any places where we play, whether it’s [the Defense Security System, the Defense Information Systems Agency] that wants to have any discussion about an application where there isn’t a human in the loop. All we do is use AI to inform a human being. With the same data that they would have otherwise. It’s a MiG and we have it in our crosshairs.”


And in the next breath, he says, he recognizes not everyone will do it that way.


So now back to that Rorschach test. Is the Air Force’s predictive maintenance work playing catchup or is it an act of innovation in which service leaders are planting seeds for future bloom? Are Pentagon leaders forgetting the industrial base of defense companies in the Washington area?


Siebel’s interpretation is more ominous: It’s a war to win artificial intelligence.


“The way this war is being played out is that you have this highly controlled five-year-plan totalitarian state in China that is controlling it top-down and making massive, massive investments. And these guys are smart as hell. They are highly motivated, and they throw a hundred teams of people at a time at this,” he said.


“And then you have this kind of very American idea … Rather than fix this through [requests for proposal] that we run through Lockheed Martin, and let these guys line up at the trough for these multibillion-dollar multiyear projects, can we get a thousand points of light of innovation going on everywhere and make little experiments and find the ones that work and latch onto them? Which is kind of a very, Western, American idea versus this totalitarian idea. It’ll be interesting to see who wins.”


Air Force officials declined to speak on C3.ai’s unsolicited proposal, but Siebel said the service is taking it seriously. This winter, he sat down with Wilson and Will Roper, the service’s top acquisition official, at the Pentagon. He’s also met with Ryan McCarthy, the Army’s undersecretary. It is unclear how or even if the Air Force is pursuing the idea further.


Still, Siebel looks at the aircraft’s handwritten data logs and sees not just unstructured data, but a smarter way.


“I know who I’m betting on.”