Reflexive Prediction

There is a class of prediction that contains within it the seed of its own falsification. A forecast about a bank's solvency, once public, can trigger the run that makes the bank insolvent. A prediction that a given route will be congested can empty it. A model that identifies which neighborhoods will see crime, and directs policing accordingly, generates more recorded crime in those neighborhoods, confirming its prediction by producing what it forecast. What makes these failures distinctive is that they are not errors of calculation but consequences of the prediction's own influence: the predicted thing is an intelligent agent that can learn the prediction and respond. Oskar Morgenstern identified this structure in the 1930s, in his analysis of the impossible regress of perfect economic foresight, and he formalized it as the foundational problem that motivated game theory itself. The insight is now one of the most important and least appreciated constraints on what