We describe an approach for dynamically generating asymmetric tactics that can drive adversary behaviors in synthetic training environments. GAMBIT (Genetically Actualized Models of Behavior for Insurgent Tactics) features a genetic algorithm and tactic evaluation engine that - provided a computational specification of a domain and notional representation of the trainee’s tactics - will automatically generate a tactic that will be effective given those inputs. That tactic can then be executed using embedded behavior models within a virtual or constructive simulation. GAMBIT-generated tactics can evolve across training exercises by modifying the representation of the trainee’s tactics in response to his observed behavior.
