This paper focuses not on those predictive analytics systems that attempt to predict naturally occurring phenomenon. Rather, it draws attention to a potentially troublesome area where AI systems attempt to predict social phenomenon and behavior, particularly in the national security space. This is where caution must be advised for the policy crowd. This paper discusses human behavior in complex, dynamic, and highly uncertain systems. Using AI to predict ever more complex social phenomena, and then using those predictions as grounds for recommendations to senior leaders in national security, will become increasingly risky if we do not take stock of how these systems are built and the “knowledge” that they produce. Senior leaders and decisionmakers may place too much reliance on these estimations without understanding their limitations.