The Department of Defense is poised to spend nearly $1 billion on artificial intelligence in the next year.
The Pentagon’s proposed budget for fiscal 2020 includes some $927 million for AI, as well as machine learning, according to Ainikki Riikonen, a research assistant for the Technology and National Security Program at the Center for a New American Security.
This includes $208 million earmarked for the Joint Artificial Intelligence Center, which was created in 2018. The Center’s initial efforts have delivered “a very mature, insightful high-level view” of issues surrounding AI, said Ian McCulloh, chief data scientist at Accenture Federal Services.
AI encompasses hardware, software, people and processes. With nearly a $1 billion bankroll, Defense Department leaders and the intelligence community are now looking for the best ways to leverage this emerging capability most effectively.
Starting point
A deep dive into the numbers shows an early emphasis on basic research. The Defense Advanced Research Projects Agency’s budget request includes $138 million for advanced land systems technology, up from $109 million in fiscal 2019. That program includes research into urban reconnaissance and AI-driven subterranean operations. DARPA’s budget also includes $10 million for the Highly Networked Dissemination of Relevant Data Project, a situational awareness tool, as well as $161 million for the AI Human Machine Symbiosis Project, up from $97 million.
“That’s all about creating systems and people that actually understand each other,” Riikonen said.
These foundational research efforts could yield practical results for the war fighter. But before the Pentagon can make use of AI’s analytic and predictive powers, military leaders will need to ensure they have the underlying infrastructure in place.
“There’s so much data available to the military, but it’s stored all over the place, and rarely in a format that is easily transferrable into an algorithm,” said Todd Probert, vice president for Raytheon Intelligence, Information and Services. “If the military wants to set itself up for success, they should focus on data curation, labeling and cleaning, as well as recruiting and training the data scientists necessary to make use of it.”
Good data requires good technical people, and those aren’t easy to come by. “Talent isn’t cheap and it’s in high demand. The government will be competing directly with industry for a very small pool of people,” Probert said. This indicates a need for early investments on talented professionals.
From there, defense can begin to look at funding specific projects and programs that take advantage of AI’s capabilities.
AI applications
The Pentagon might begin by considering the potential for AI as a weapon of war. “We are only starting to scratch the surface on the impact of AI and how it can be manipulated by adversaries for nefarious purposes,” said Rahul Kashyap, president and chief executive of network traffic analysis company Awake Security.
Machine learning might help military systems be more effective, but the reliance on data could also make those systems vulnerable to new kinds of attack. “With the adversarial use of AI, there are already discussions about ways in which data we have come to rely on may be poisoned to trick the machine inputs and algorithms,” Kashyap said.
Some experts suggest that any early investments should address this potential risk, building in a defensive capability as part of AI’s foundational layer.
Others say that the low-hanging fruit lies in the military’s ability to leverage AI in support of mundane, but nonetheless critical, tasks.
In the near term, for example, AI spending could help provide transparency around inventory and supply chain management.
“AI could help manage the complexity behind the connectivity and flow between transportation, people, facilities and supplies including equipment, spare parts and fuel in a predictive manner,” said Brigham Bechtel, chief strategy officer for intelligence and defense at big data applications firm MarkLogic.
In this scenario, AI would leverage existing data on materiel availability and equipment performance to drive preventative maintenance, as well as parts procurement — “keeping records of millions of screws, wire couplers, and even tank gun barrels to support scaling to operational demand,” Bechtel said. That’s a task for which machine-scale intelligence is ideally suited.
In the realm of ISR, some industry representatives point to “open-source intelligence” (such as social media) as a prime target for AI investments.
Sources such as Facebook and Twitter contain “significant intelligence that is beyond the scale of humans or classic computation analysis,” said Chad Steelberg, chief executive and chairman of AI-based analysis company Veritone.
As in logistics, open-source intelligence offers ample data in a space where machine-scale analytics could have a deep impact. “The war of ideas, ranging from ISIS recruiting to state-sponsored propaganda, is the most dangerous battlefronts today,” Steelberg said. “With the source of ideas now being influenced by AI, the countries that harness this new weapon most effectively will have a distinct advantage.”
The intelligence community also could benefit from AI’s analytic powers to manage the sheer volume of sensor data in the field.
“Is the analyst overwhelmed with data? If so, AI has the potential to help,” said Graham Gilmer, a principal in Booz Allen Hamilton’s analytics business. “Generating a more robust search capability, fusing data from multiple sources, and generally doing the heavy lifting to cue the analyst are the most immediate applications.”
In addition to addressing external data, the intelligence community could score an early win by building AI models that scrutinize conversations amongst analysts themselves.
“In an ISR suite there can be as many as 15 chat rooms going at any time, with info coming in from various units and intelligence agencies,” Probert said. “That’s too much data and crosstalk for a person to manage, so information is inevitably going to be missed. We need machine learning tools that can flag critical data and alert analysts to what’s important.”
All these represent valid points of inquiry and the Pentagon likely will pursue diverse variations on these themes. In the short term, though, analysts predict AI will mostly be about robots.
“Advanced automation is the fastest growing category in AI, with the rise of unmanned systems,” Riikonen said, noting it would be a natural evolution for the military to leverage private sector learning to utilize AI in support of autonomous systems. “That fits very well with the overall U.S. defense strategy, which is all about having more of these autonomous systems that support war fighters in denied and contested environments.”
In order to achieve those goals, the Defense Department may have to adopt a new way of investing in technology.
Rather than a single development effort that leads to a completed product, however, AI requires an iterative process in which the computers learn over time.
“You do small chunks, you do small bites,” said Paul Johnson, Grant Thornton public sector senior strategic adviser for the defense and intelligence community.
In this light, AI investment will require not just algorithmic development, but investment in organizational change, to spur deep interactions between stakeholders. “We need to get the coders in the same room with the end users and start having the conversation about the art of the possible,” Johnson said. “You have to have that conversation early, often and repeatedly, for the coders to understand what they need to do.”
