Training the human in the loop.

The gap

DoD Directive 3000.09 requires appropriate levels of human judgment. The Department’s own explanation concedes that appropriate admits no single standard, and the AI ethical principles stop at five adjectives. No published DoD or Army guidance says what anyone on the staff must check before signing machine-drafted product, or how much checking is enough.

Meanwhile GenAI.mil passed a million users within two months of launch. Fielded tools, unwritten standard.

What the course does

• Day one. Decision science taught on MDMP steps — mission analysis, COA development, the wargame, the running estimates.
• Days two and three. A corps staff runs a crisis simulation with an AI advisor on every desk. It is deliberately wrong some of the time. Every query and response is recorded.
• The last half-day. The room argues with its own record, while it is still fresh.

What the command gets

• A record per warfighter. Who consulted the advisor and when, whether the base order was opened before or after, which checks ran, and confidence set against correctness.
• The staff pattern. Which step the room let go of first, and which output nobody challenged.
• A draft AI-use standard written by the cohort, since no published one exists.

Why this vendor

• Seen from both chairs. Ten years as a participant in U.S. military wargames as a Navy Reserve officer, and several years designing and running crisis simulations for graduate students at Georgetown University. Sixteen written participant reflections are published on the program site.
• AI agent agnostic. No vendor, model or version named in anything taught, so the habit survives the tool being replaced.