Probasco: AI Risks Hollowing Out U.S. Military Judgment

The Defense Department's guidance on autonomous weapons was last updated in 2023, before Anthropic had even launched its first commercial model, Claude. Claude has since gone through more than 20 new versions.

That gap frames Emelia Probasco's essay in Foreign Affairs: the risk from mixing AI and armed forces is less a runaway robot than a hollowed-out human. Probasco, a senior fellow at Georgetown's Center for Security and Emerging Technology, writes that "avoiding Terminator-like dystopias requires maintaining control not just of AI, but of human behavior."

She points to a growing body of research showing that leaning on AI can degrade the people using it. In studies she cites, "computer scientists and oncologists performed worse at their jobs after using AI assistance and having it removed." AI use, she argues, is affecting people's cognitive capacity to evaluate evidence, make independent judgments, and even do their basic jobs.

The concrete case in the Pentagon's own inventory is the Maven Smart System, Palantir's AI-enabled targeting and decision-support platform, which was designated an official program of record earlier this year. Probasco has previously argued that the claims about Maven's abilities might be overstated, and that much of the American advantage came from the scale of data flowing in and the skills of the people using it. Those are the skills that atrophy when the tool takes over the work.

Staffing is the compounding problem. Reductions in military staff sizes, she writes, are often justified by imagined AI efficiencies, and even if doctrine caught up, the Pentagon would struggle to keep soldiers current with the systems they are handed. Her prescription is that Pentagon officials focus on "reworking how the military trains, deploys, and enables AI systems" and not just on engineering or buying them.

The essay attaches no numbers to the deskilling claim or to the staff cuts it points at.