The Defense Advanced Research Projects Agency sees artificial intelligence as a powerful, but still fundamentally limited military tool, Director Arati Prabhakar said Monday at an Atlantic Council event in Washington, D.C. Prabhakar suggested movies depicting humanoids and the increased use AI technologies, such as facial recognition, search engine optimization, and self-driving cars have created a false perception that AI technology is exploding. But, she said, the available machine-learning technology has its limits. She noted the best systems can analyze images better than humans, but aren’t right every time. “The problem is that when they’re wrong, they’re wrong in ways that no human would ever be wrong,” Prabhakar said. Rather than just trying to fix those limitations, she said, DARPA is investing in a new wave of AI technology that will enable machines to explain their limitations to humans, understand causation rather just correlation, and apply gained knowledge across multiple domains. In the meantime, DARPA is using available AI technology in new ways. Prabhakar said DARPA is building radar-jamming systems that will allow aircraft to scan the radio spectrum, identify an adversary’s radar signal, and instantly create a jamming profile. Today, US aircraft have a library of radar-jamming profiles to transmit when pinged by an adversary’s radar, but they still regularly encounter new signals, she said, and building a countermeasure can take years. “We need a way today to deal with the changes that are happening on the ground,” Prabhakar said.
More than 100 B-21s will be needed if the nation is to avoid creating a high demand/low capacity capability, panelists said on a Hudson Institute webinar. The B-21's flexibility, stealth, range and payload will be in high demand for a wide range of missions, both traditional and new.