AI Attitudes: Senior Service College Student Reflections on AI Use

Small Wars Journal | Kelly Ihme

The US Army War College Distance Education Program evaluated 379 officer reflections on artificial intelligence integration in operational planning assignments. The assessed module required students to submit model-generated text for grading before authoring a reflection. Software platforms included GenAI.mil running Gemini 2.5, ChatGPT 5.1 and 5.2, Microsoft Copilot, and NotebookLM.

Between 25 and 30 percent of participating officers employed multiple software tools simultaneously to cross-verify outputs. Student assessments revealed that commercial large language models accelerate research and drafting but lack doctrinal precision. For example, generative systems repeatedly failed to distinguish between landpower and land domain in operational context. Furthermore, these models generated hallucinated citations, broken links, and inaccurate source material. Officers conceptualised artificial intelligence as a junior staff officer. Grounded platforms with restricted data boundaries produced higher user trust and fewer structural errors. Senior leader attitudes and verification frameworks directly dictate force adoption rates.

Comment

Doctrinal vocabulary in joint command structures relies on strict semantic boundaries to align fires and maneuver across component commands. When automated systems misinterpret foundational constructs such as landpower versus the land domain, the error propagates into operational design under Joint Publication 3-0. In high-tempo operational environments, unverified command post summaries distort battle tracking across joint task force headquarters.

The friction mechanism becomes evident during target nomination workflows within Project Maven. Language models that misapply operational doctrine alter the target prioritisation criteria used by division staff during the military decision-making process. During Project Convergence experimentation, unvalidated text generation systematically introduced processing delays inside XVIII Airborne Corps fire support elements.

Strategic Question for Discussion

If automated target nomination tools in Project Maven misapply doctrinal terms, which operational mechanism breaks down first — command post battle tracking or joint fire support coordination?

The pattern suggests that joint fire support coordination absorbs the immediate impact of doctrinal errors. In automated targeting pipelines, semantic inaccuracies alter strike prioritisation before battle tracking displays update at corps level. Field evidence from division-level command posts indicates that target processing units face verification bottlenecks that directly negate machine-driven speed advantages.

Share your assessment in the comments below.

To read in details click here