AI Still Far Away from Mission-Critical Role, DoD’s Porter Says

DoD Pentagon Military

Lisa Porter, Deputy Under Secretary of Defense for Research and Engineering, had a lot of good things to say about the promise of artificial intelligence (AI) technologies at the GEOINT Symposium on Tuesday, with one important caveat: AI isn’t ready for prime time in Department of Defense (DoD) critical applications, and likely won’t be for some time.

Speaking on June 4, Porter spoke about AI and DoD apps, and made it clear that the best way to take advantage of AI is to put significant effort into finding a problem that the technology can really help solve. What is most useful, she said, “is a well-structured problem that is suitable to AI … otherwise AI is just a shiny tool.”

“Not every problem is ideal for AI,” she said, and advised attendees to “understand the problem better” as a first step. “Take more time to understand what problem you are trying to solve,” she said. “Then see if it’s really possible to generate the right kind of AI data.”

Porter further urged technologists to spend a lot of time at the beginning of the process with end-users evaluating whether a potential AI project features the right data, reasonable outcomes, and proper metrics to evaluate results.

As for mission-critical DoD applications, Porter ticked off a list of problems with the current state of AI development that she said collectively constitute a “very big problem” for using the technology in vital situations. Those include:

  • The “brittle” nature of many AI algorithms;
  • Problems with reproducing consistent results;
  • Vulnerability to spoofing; and
  • A lack of built-in security features.

She also said agencies still seeking to move past legacy IT systems to AI-ready systems are facing a “very hard, heavy lift” in that process.

To companies looking to pitch the Federal government on AI applications, she strongly urged them to “explain why your product is effective,” and to fully discuss data sources, algorithms, and how applications produce consistently repeatable results. Companies that can’t show enough evidence on that front might win a pilot project from DoD, but “people who try to take short cuts get caught in pilot purgatory,” and aren’t likely to win more lucrative contracts, she said.

Reporting on AI efforts already underway within DoD, Porter said the agency’s Joint Artificial Intelligence Center (JAIC) “is just starting to get going” after being created a year ago. She said the effort “really has the right focus” on “the impact of AI at scale.”

“They realize this is very hard,” Porter said, adding, “It’s all about how we do AI at enterprise level.”

“There’s nothing very smart about today’s AI tools … That’s what we need to improve,” she said. The achievement of “common sense” in human-machine teaming would be “nirvana,” Porter added. “That team could be very powerful … All of these things require some degree of cognition.”

On the DARPA front, Porter said about one-third of the organization’s current projects involve AI “to some degree.”

While advanced technology development remains a daunting task, “we will always be ahead if we play to our strengths,” she said. “Those who cheat and steal from us will never win if we play to our strengths,” including adhering to the rule of law, Porter said.

There are probably few Federal officials better positioned to judge the capabilities timeline for AI than Porter. In her current position, she oversees research, development and prototyping activities across the DoD enterprise, along with the activities of the Defense Advanced Research Projects Agency (DARPA), the Missile Defense Agency, the Strategic Capabilities Office, the Defense Innovation Unit, and the DoD Laboratory and Engineering Center enterprise. Before her current post, she was executive vice president at In-Q-Tel, and was the first director of the Intelligence Advanced Research Projects Activity (IARPA).