The U.S. military discovers that more training isn't better training: AI as the 80/20 filter

By Small Wars Journal · July 15, 2026.

Four authors—two Coast Guard lieutenants, an Air Force sergeant and a veteran educator from Florida's public school system, all tied to learning-technology studies at Florida State University—have published an article with a thesis that is uncomfortable for any hierarchical institution: most of the training military units do does not improve their actual readiness. Using a synthetic dataset that simulates four fictional units (the authors are explicit that they are not exposing classified operational data), they apply the Pareto principle—the 80/20 rule—to military training and find that simulation, live fire and field exercises concentrate more than 80% of measurable readiness gains, while the remaining categories (classroom instruction, administrative and maintenance tasks) barely move the needle.

The most provocative finding is not the 80/20 itself—a heuristic seen a thousand times over in business management—but what it breaks: training duration correlates weakly with readiness improvement. Even within the high-impact categories, time invested does not predict the outcome. What does explain the differences between units subjected to the same training structure is execution: realism, intensity and instructional quality. In other words, a unit can train fewer hours and be better prepared than another that trains more, if it does so better. For a military culture that tends to measure commitment by hours of instruction completed, it's a claim that strikes directly at the habit of justifying budget and schedule by volume rather than by result.

The role the authors reserve for AI is deliberately modest and, precisely for that reason, more credible than much of the noise about military AI: there is no talk of autonomous weapons or algorithmic combat superiority, but of a system that ingests training data, applies consistent evaluation criteria and flags which activities produce real readiness across multiple units, at a scale a human commander cannot sustain manually. The authors themselves insist that AI does not replace command judgment, it only accelerates pattern recognition. It's a distinction worth taking seriously, not as a defensive rhetorical gesture, but because it fits with what is really working today in organizations that adopt AI successfully: decision-support analytics, not decision autonomy.

It is worth, however, not overselling the finding. The dataset is synthetic and the authors themselves acknowledge it: it serves to demonstrate the technique, not to validate it operationally. A unit's real readiness depends on personnel rotation, equipment availability and mission pressure, variables that a dataset generated to illustrate a method cannot capture. This is, at bottom, a methodological proposal—a five-question framework any training officer could start applying tomorrow with spreadsheets, without relying on any AI system—rather than an empirical study of the real military. That does not diminish its value, but it does bound what can be asserted from the evidence presented.

Our reading: this piece is representative of where AI is really landing in military institutions in the short term, and it contrasts with the dominant narrative of the autonomous drone and the algorithmic weapon that usually dominates the debate on AI and defense. The most immediate and least glamorous application—optimizing training schedules, budgets and logistical priorities through data analytics—is also the one with the shortest path to adoption, because it does not require solving any problem of lethal autonomy or command doctrine, only better management of the information that already exists. It is consistent with a pattern we have been observing in other sectors: the first tangible benefit of AI is usually not replacing the human in the most visible task, but making visible the waste the organization had gone years without measuring. In the long run, if these techniques scale from simulation to real operational data, they point to armed forces that do more with less training time—freeing human and budgetary resources toward what does require irreplaceable human judgment—a modest but concrete example of how the abundance AI promises does not always arrive as a new product, but as the elimination of friction in processes we had taken for granted as optimal without ever having checked.