FORT IRWIN, Calif. — Army data scientists designed an experimental AI-powered platform that could transform the military decision-making planning process, to compress planning timelines from days or hours to seconds and to enable rapid, adaptable decision-making on the future battlefield.
As part of the U.S. Army Combat Capabilities Development Command, or DEVCOM, Army Research Laboratory, which is the Army’s sole fundamental laboratory, the Course Of Action Generative Pre-trained Transformer, or COA-GPT, is part of the U.S. Army’s largest experiment exercise at Project Convergence Capstone 6 at Fort Irwin, Calif., from July 17-29. The COA-GPT is a scientific platform designed to accelerate mission planning and provide an adaptable and decisive framework for planning, intelligence and command to solve critical human-AI integration challenges,
“By fusing generative AI with hard-won military expertise and doctrine, COA-GPT provides Army leaders and warfighters, the ability to generate and compare innovative workflow options,” said Dr. Amar Marathe, DEVCOM ARL COA-GPT program manager. “We are not replacing Soldiers with AI—we are arming them with it. COA-GPT integrates seamlessly into the workflow to make military planners faster, more efficient, and more effective.”
COA-GPT represents more than four years of partnership between ARL and the U.S. Army Futures and Concepts Command’s Command and Control Future Capability Directorate. The team uses regular Soldier-driven experiments to characterize planning-related Soldier-AI integration challenges and ensure that COA-GPT reflects the direct feedback of the Soldiers the testbed platform is designed to support.
According to Marathe, a major scientific hurdle is clearly communicating intent to machines across a distributed team of Soldiers and systems. Through an industry partnership with Hyssos Tech, COA-GPT uses an advanced graphic interface within the testbed, enabling leaders across multiple echelons to use intuitive methods, like drawing on a shared map, to better direct agentic AI to reduce miscommunication, cognitive load and friction across personnel.
“We built COA-GPT with a plug-and-play architecture, giving researchers the freedom to easily swap components and test new AI capabilities without having to rebuild the entire system,” Marathe said.
The next steps include refinement of the COA-GPT testbed through rigorous Soldier feedback and industry collaboration. Future efforts will explore how generative AI shifts the traditional responsibilities of Army planning staffs, which new cognitive competencies planners need to prioritize, and how to optimize workflows, ultimately helping the Army maintain a decisive advantage.
By DEVCOM Army Research Laboratory Public Affairs Office
You can skip to the end and leave a response. Pinging is currently not allowed.
Read the full article here