WASHINGTON — A data collection and dissemination program the Army has used over the past five years in Africa could be a model for how military intelligence analysts make better sense of the firehose of information they face each day.
The Africa Data Science Center started as a pilot project in 2018 using data science tools to cull thousands of incident or intelligence reports to find the relevant information an intel soldier needs to analyze for their command.
Col. Mark Denton, commander of the 207th Military Intelligence Brigade, told an audience Tuesday at the annual Association of the U.S. Army gathering that the days of analysts sifting through stacks of paper reports or creating their own spreadsheets to track data are gone.
The current system can “easily consolidate 3,000 reports in a matter of seconds,” Denton said.
“What this does is transform chaos into clarity,” said Maj. Gen. Todd Wasmund, commander of Southern European Task Force-Africa.
Denton put up a map of the African continent for the audience. Large swaths of it were saturated in overlapping purple bubbles. Each bubble was a report of a “conflict event,” such as a bombing, riots or other significant developments in a city, state, nation or region.
“The purple stuff is the chaos coming in that the analyst needs to make sense of,” Denton said.
The system the data center uses takes machine learning tools and natural language programming to both find the data an analyst will need and combine it into a readable report that doesn’t require a computer programmer to decipher.
Most data analysis done by hand is complicated, takes too long and makes it difficult to spot trends quickly, Denton said.
Using natural language processing, which makes computers deliver information in language a human can understand rather than computer programming language, analysts can categorize text reports and use machine learning tools to automate the process.
During the data science center’s pilot, users saw a 70-90% reduction in irrelevant reports returned from searches. The tools also allow analysts to create visualized reports, with charts, graphics and maps that can give commanders better context and a quicker understanding of the intelligence picture.
Programmers and data scientists sit hip-to-virtual-hip with analysts to determine what kinds of questions the analyst can ask of the available data, and how the answer will be delivered. The analyst tells the data scientists what they need to know and the scientist uses programming and algorithms to query the data for the answer.
After the system delivers those data summaries, users can provide feedback to make the machine-learning model more accurate. A user gets to tell the system what was useful and what wasn’t. That step should help improve the models over time, making them more efficient and accurate, Denton said.
Currently, the data science center runs only for U.S. Africa Command initiatives. But, Denton added, intelligence branch commanders and other combatant commands are working to develop their own data collecting and analysis tools.
The idea is to create a kind of “hub and spoke” system so that all Army intelligence would be linked across regions, Denton said. This will allow analysts to find patterns or pull data to inform their own area’s operations, he added.
Todd South has written about crime, courts, government and the military for multiple publications since 2004 and was named a 2014 Pulitzer finalist for a co-written project on witness intimidation. Todd is a Marine veteran of the Iraq War.