Performative modeling

Performative modeling is the design and deployment of algorithmic systems that do not merely describe a pre-existing reality but actively reshape the social, economic, or institutional conditions they purport to measure. The term extends the economic concept of performativity — the idea that theories and models can make the world they describe — to the domain of machine learning and algorithmic governance. A credit-scoring model that causes lenders to deny credit to certain zip codes is not describing a pre-existing risk profile; it is creating one by redlining. A predictive policing model that concentrates patrols in certain neighborhoods is not describing crime patterns; it is producing them through increased surveillance.

Performative modeling is the special case of reflexive prediction in which the system's effects are not merely statistical but structural: they alter the distribution of resources, opportunities, and power in the world. The concept challenges the epistemological foundations of machine learning by showing that the distinction between description and intervention collapses when the described system is composed of agents who can react to the description. The question is not whether a model is 'accurate' in a static sense. The question is whether the world it produces is one we want to live in.