Pentagon’s AI scrutiny

The Pentagon is moving artificial intelligence from back-office analysis into military planning, surveillance and targeting workflows under rules written to keep humans in control. (defense.gov) In plain terms, these systems sort huge piles of images, signals and text faster than people can, then flag patterns for analysts and commanders. The Department of Defense says its 2023 AI adoption strategy is meant to push those tools “from the boardroom to the battlefield” for faster decisions. (defense.gov) The Pentagon’s main policy for autonomous weapons was reissued on January 25, 2023. Directive 3000.09 says autonomous and semi-autonomous weapon systems must be designed to reduce the risk of unintended engagements and ties AI use to the department’s ethical and responsible AI rules. (esd.whs.mil) The procurement push is already concrete. Replicator, announced on August 28, 2023, aims to field thousands of uncrewed systems by August 2025, with the first line focused on low-cost “attritable” autonomous systems that the military can afford to lose in combat. (congress.gov) Project Maven shows how that shift works in practice. The effort began with an April 26, 2017 memo to apply machine learning to military intelligence, and by November 7, 2023, Maven had become a formal program of record at the National Geospatial-Intelligence Agency. (dodcio.defense.gov; breakingdefense.com) Maven’s job is not to “pull the trigger” on its own. It uses computer vision — software that scans imagery the way a search engine scans webpages — to detect, identify and track objects of interest for human users. (meritalk.com) The chain of command around these tools is also changing. A November 18, 2024 directive formalized the Chief Digital and Artificial Intelligence Officer, or CDAO, as the Pentagon official responsible for department-wide AI functions across the military departments, combatant commands and defense agencies. (esd.whs.mil) The Pentagon says speed and safeguards have to advance together. Its responsible AI pathway says AI must be designed, tested, procured, deployed and used in “lawful, ethical, responsible, and accountable ways,” and its five AI ethics principles are responsible, equitable, traceable, reliable and governable. (defense.gov; defense.gov) Congress and watchdogs are still pressing on oversight. A March 22, 2024 Congressional Research Service brief said lawmakers need enough information to judge Replicator’s merits, and a Government Accountability Office report said the department still had gaps in defining and planning for its artificial intelligence workforce. (congress.gov; gao.gov) That leaves the central question less about whether the Pentagon will use AI than where it draws the line. The answer is being written through contracts, doctrine and review rules that decide how much judgment stays with software and how much stays with people. (defense.gov; esd.whs.mil)