The Defense Intelligence Agency is the latest national security organization to adopt the free-form feel and practices of Silicon Valley companies.
Acknowledging the need to leverage commercial advancements in technology — and even business practices — DIA is trying to help unburden analysts from the deluge of data using state-of-the-art artificial intelligence and machine learning.
DIA has held two previous industry days, which are “designed to find new capabilities and business processes from the private sector and academia, and introduce them to DIA’s collaborative Innovation Hub (iHUB) environment to create, test and evaluate potential products.”
Moreover, the new iHUB, which hosts the industry days and was established September 2016 as a result of meetings in August 2016, seeks to bring DIA employees of all stripes and mission areas together both virtually and in a space emulating a Silicon Valley company. Employees can gather, write on the walls and collaborate in a more informal way.
Bolden said one of the most interesting things is the interaction between the disparate elements of communities at DIA who might be seeking to solve a common problem despite serving in different mission cells.
Several companies presented directly to analysts in the iHUB over two days addressing critical mission needs identified in DIA’s NeedipeDIA, a repository that allows DIA to communicate to industry and provide a gateway into government acquisition through an open broad agency announcement.
Some needs include using machine learning and Natural Language Processing to identify and organize weapon system information in big data; machine learning to support workflow automation activities; tools for predictive analysis, alerting and indications and warning; artificial intelligence and machine learning support to military operations to fuse multi-INT sensor data for real-time battlespace awareness and predictive analysis; and artificial intelligence and machine learning to support open-source information gathering.
During one presentation, Austin, Texas-based SparkCognition offered their solutions in Deep Natural Language Processing, automated model capability and auto encoding to “provide auditable and accountable deep learning solutions with speed and automation beyond traditional AI.”
The SparkCognition team believes their solutions can get at several of DIA’s needs listed on NeedipeDIA.
Another presentation featured technology from Percipient.ai that uses AI that “ingests, analyses and models multi-source imagery intelligence including full motion video, geo-spatial and other forms of SIGINT and MASINT to provide virtually instant analysis.”
Percipient.ai is seeking to strengthen the analyst who is drowning in sensor and data overload.
Percipient.ai’s tools look to move forward activity-based intelligence, CEO Balan Ayyar told reporters following his pitch. Activity-based intelligence enables greater understanding of patterns of life and certain activities as opposed to individual pieces of data. For example, a computer, using state-of-the-art machine learning and AI, can scan full-motion video and understand that a digging motion on a road could represent the activity of an insurgent planting an improvised explosive device and alert the analyst immediately.
One of the most critical challenges for analysts, and especially all source analysts as DIA is an all-source intelligence organization, is a lot of data such as images or full-motion video is not structured, Randy Soper, senior analyst for analytic modernization at DIA, told reporters, meaning a human has to view the data to know what’s in it.
Commercial technology can now make sense of unstructured data much faster than humans can, he noted. This ability allows analysts to get insights faster, creating decision advantage over adversaries, and allows them to go back to the same data sets with new understanding when circumstances change.
A challenge to leveraging machine learning and AI, Soper indicated, is the machine has to be taught. In order for AI to be successful, there needs to be understanding of mission and characterizing of data, Soper said, adding that the AI has to be taught what parts of the data are import to be teased out.
The human element is still critical as they ultimately are the ones that make the judgements based on data that is returned. This was something commercial industry presenters also stressed. The machine can point the human in the right direction, but it is up to the human to infer and make decisions based on the data.
This is part of the draw behind “explainable AI,” which makes the computer provide the human how it arrived at its answer.
When companies are invited to pitch such innovations to the DIA analysts at the iHUB in a “Shark Tank-esque” manner, they demonstrate their capabilities without seed money, Bolden said.
If there is interest in the solutions, mission cells will coordinate with the CFO and innovation office to look to pilot capabilities with the CFO drawing up funds.
DIA has also begun partnering with DIUx, the Pentagon’s rapid prototyping cell, signing a Memorandum of Agreement making it one of the first IC elements to work with DIUx. This partnership involves two projects.
This partnership is key given DIUx’s contract vehicle authorities. DIUx uses Other Transaction Authorities, which are outside the Federal Acquisition Regulation and allow for more agile awards. DIUx still uses the BAA process but is working to draw up an OTA-type agreement.
What’s attractive about these authorities from DIA’s perspective, Bolden said, is for prototyping because they don’t want to buy a system off-the-shelf to prototype it. In the interim, they’re partnering with DIUx.
The first commercial pilot that resulted from the industry day process involved evolving tradecraft, giving DIA a decision advantage, Bolden said while offering limited details given sensitives.