Federal Agencies Using Machine Learning and AI to Tackle Problems
WASHINGTON — Machine learning and artificial intelligence technology are fast-moving, complex technologies businesses and government agencies are increasingly putting to use to develop efficiencies of people and processes. Federal IT practitioners, in particular, are seeing the benefits of this technology today.
The Federal News Network convened such experts from the Defense Threat Reduction Agency, Defense Intelligence Agency, and Naval Information Warfare Systems Command as well as from Booz Allen Hamilton, Snowflake, and DataRobot to share how strategies and initiatives around machine learning and artificial intelligence are being implemented and primed for future use.
Machine learning is the idea that machines should be able to adapt and learn through experience. Artificial intelligence, on the other hand, applies machine learning to execute tasks and solve actual problems.
“As a company, early on we recognized this as an area of focus for our business,” said Melissa Sutherland, vice president, Defense Military Intelligence, Booz Allen Hamilton, adding that the government contracting company started investing in ML and AI a couple of years ago as it acknowledged a change in the information environment.
AI projects have also been running in DIA for several years and used to craft a defense intelligence strategy according to the agency’s Director of Artificial Intelligence, Brian Drake. And NAVSEA has also invested in hundreds of projects and deliberately focused on building its ML and AI workforce to apply learning to real-world problems.
But most efforts up to this point have been “really focused on building capabilities based on a holistic vision of integrating problem domains and hardware/software programs collectively,” said Chris Brazier, division chief at DTRA. “Problem domains have more tech space data than humans can find.”
“We discovered that by 2025, we’ll be encountering zettabytes of data in the public sphere,” said Drake. Putting that into perspective, he offered, “A zettabyte is like 36,000 years of high definition video. If we were to commit the entire human capacity of the U.S., we couldn’t get through a zettabyte.”
“Most of the challenges we’re solving are in terabytes of data,” said Sutherland. But as agencies focus on their real-time efforts to stay ahead of crises, like COVID or exploiting materials from enemy combatants, capacities need to have the ability to scale.
“As we look at our knowledge of how to apply ML and AI… [we] needed software environments… that ensure quick development of ML and AI while addressing strict security needs,” said Brazier. And that’s where companies like Snowflake and DataRobot come into play.
“Snowflake is the first data platform built for cold object storage and elastic compute,” said Nicholas Speece, Snowflake’s chief federal technologist. “It looks like a standard database, but it also has many more capabilities, and we’ve been applying a lot of those capabilities to the AI, ML space.”
DataRobot, for its part, is an enterprise AI platform that is involved in every step “once you have a data set until you create mission impact,” according to Chad Cisco, DataRobot’s general manager for its Government Business Unit. “DataRobot can automate or streamline the steps… to include a platform for data preparation, an AI catalog in automated feature engineering, and building a number of ML models…with automated capabilities to evaluate those models for a number of features from trust and bias to how they will perform under different operations. We find those places the government needs solved and leverage commercial tech to streamline the process and leapfrog the development of new tools.”
ML and AI have been used largely to automate manual and repetitive workflow, to quickly access and train models, and to “inform the design of algorithms and build trust in human-machine teaming,” according to Carly Jackson, chief technology officer at NAVWAR. In the future, agencies see an increased demand to automate open source and PAI data for supply chain analytics and automating other long and intensely manual processes.
While ML and AI have already provided insights into soldier health and readiness, financial intelligence, and other partnerships with efficient integration, Federal IT practitioners have identified that there is a need for greater awareness to transform the culture of how we use AI and “push the concept further into the hands of the warfighter.” [Jackson]
“[AI is] beneficial, not harmful to human welfare” insisted Sutherland, “the goal is to assist humans in daily activities, not replace them… to make them more efficient through time-saving. We want to achieve real human-machine teaming.”