NATO wants AI to speed up learning. The result must be a changed capability.

Artificial intelligence can process an exercise report faster. The military value comes when somebody changes what happens in the next exercise. That distinction sits at the centre of NATO's latest discussion of institutional learning.

In a 14 October account of the Lessons Learned Conference in Lisbon, Allied Command Transformation highlighted AI, realistic training and interoperability. JALLC hosted the meeting on behalf of Supreme Allied Commander Transformation. ACT described AI as a way to handle information and identify patterns while retaining human judgement.

The important measure is the change produced by a lesson. Our analysis is that this creates a more demanding test for defence software than producing a fluent summary: the user needs to follow the evidence through to a decision and a subsequent check.

A pattern needs a cause

Repeated complaints in reports can reveal a real problem. They can also describe different causes with similar wording. A tool that groups them should preserve the route back to each observation, including its context and limits.

Consider a training report that says information arrived late. The relevant issue might be a technical interface, a procedure, or how the exercise was arranged. A summary alone cannot distinguish those explanations. An analyst needs enough evidence to investigate and select the appropriate change.

This matters to companies offering dual-use AI for document analysis. The useful product is a reviewable link between what was observed and what the organisation decides to do.