Listening better and identifying from farther away — US intel agency touts new programs
The U.S. government is teasing a couple of technologies it is pursuing for intelligence gathering teams, at least one of which could deliver intense amounts of biometric data and controversy.
Reporting on an armed forces conference this week, Signal magazine, made some educated guesses about developments hinted at by Director of Intelligence officials.
One apparently involves whole-body biometrics gathering done from buildings and aerial drones.
A second would seem to be amped-up machine translation of any language. In research, the systems reportedly have outperformed Google Translate in information retrieval, according to a government intelligence official quoted in the article.
Signal is published by AFCEA International, once known as the Armed Forces Communications and Electronics Association, a group with roots reaching back to the U.S. Civil War. It was reporting this week on the group’s spring intelligence symposium.
Government researchers are crafting algorithms that can perform whole-body biometric identification, according to the article. Dustin Gard-Weiss, deputy director of national intelligence for policy and capabilities within the Office of the Director Intelligence, told a keynote audience.
According to the magazine, Gard-Weiss could have been discussing the Biometric Recognition and Identification at Altitude and Range project within the Intelligence Advanced Research Projects Activity agency.
Known as Briar, the four-year program was the subject of a solicitation late last year. Its description there goes further, however, outlining how IDs would be made “at long-range and from elevated platforms.”
An initial presentation about Briar is dated October 2020, and an IARPA cash challenge related to what became Briar was held in 2018, according to those documents.
While face biometrics commonly battle with poor lighting, awkward camera angles and subject movement, the researchers in this instance must also overcome air quality and turbulence.
Also vaguely discussed at the event was a machine language program, called Material, that would perform English-in/English-out speech recognition tasks on any language. Two-part queries — English words and domain — would be able to dig out information on, for example “viral load” in the medical domain and the same phrase in the economy domain, according to IARPA.