Characterizing and predicting agents via multi-agent evolution is an interesting patent filed by DARPA. In a wargame simulation, this invention claims to be able to predict the behaviour of the actors based on their past behaviour, and the state of the game at the particular moment.
Wargames have been designed, developed and played to understand and study combat scenarios. The ability to predict the opponent's move in a combat situation is fraught with many imponderables. The claim of the patent although promising need to pass the real test of real wargames.
However, this is an interesting development. Digital pheromone infrastucture described in the patent indicates the use of Ant Colony Optimization and learning. The author's thesis can be downloaded as a pdf from here.