DARPA developing AI into a mission-critical partner

As artificial intelligence advances, the Defense Advanced Research Projects Agency is moving toward treating computers less as tools and more as partners that can help solve complex military problems, according to Matthew Turek, program manager in DARPA’s Information Innovation Office.

Speaking at FCW’s March 10 Defense Readiness Workshop, Turek said DARPA has approximately 30 programs focused on AI and another 90 that are leveraging AI technologies -- from foundational science and hardware to algorithms, knowledge representations, machine learning and autonomy.  Some of those, he added, are already in the field.

Those programs fall into three waves of AI. The first covers symbolic reasoning, in which engineers create sets of rules to represent knowledge in well-defined in domains, like optimizing the shipping of military equipment. The second wave applies statistical models that have been trained on big data for specific problem domains to deliver nuanced classification and prediction capabilities. This type of AI has been used for face detection algorithms and virtual assistants like Siri – whose foundational technology was developed at DARPA, Turek said.

Service members are using second wave statistical machine learning for intelligence surveillance and reconnaissance and predictive vehicle maintenance. For ISR, the AI systems analyze drone video, detect objects in the scene and rapidly cue people on the battlefield to what's happening around them. Predictive maintenance applications are pulling data from vehicles’ operational systems, their service histories and built-in sensors to schedule maintenance before a part is likely to break.

For the third wave of AI, where computers become real partners, DARPA is building on second-wave learning capabilities by “trying to improve the abstracting and reasoning,” Turek said, so systems can explain the decisions they make. To develop explainable AI, the technology expected to enable third wave systems, DARPA is “investing in things like common sense reasoning, continuing our investments in theoretical foundations of machine learning and applying those to complex DOD problems,” he said.

This article first appeared on GCN, a Defense Systems partner site.