🤖 AI Summary
This paper addresses *performativity* in machine learning—where predictions actively alter the underlying system (e.g., user behavior), inducing dynamic distributional shifts that violate standard i.i.d. and stationarity assumptions. To formalize this phenomenon, the authors introduce the first systematic definition of performative prediction and propose a “learning–steering” dichotomy to distinguish between passive model adaptation and active system intervention. They further define *performative power*, a novel metric quantifying a platform’s capacity to influence user behavior through its predictions. Integrating tools from game-theoretic equilibrium analysis, distributional shift modeling, and optimization theory, the paper develops a provably convergent learning algorithm framework and characterizes necessary and sufficient conditions for stable equilibria. The work establishes a rigorous theoretical foundation for analyzing the interplay between predictive models and institutional power in digital markets, with implications for algorithmic governance and policy design.
📝 Abstract
Predictions in the social world generally influence the target of prediction, a phenomenon known as performativity. Self-fulfilling and self-negating predictions are examples of performativity. Of fundamental importance to economics, finance, and the social sciences, the notion has been absent from the development of machine learning. In machine learning applications, performativity often surfaces as distribution shift. A predictive model deployed on a digital platform, for example, influences consumption and thereby changes the data-generating distribution. We survey the recently founded area of performative prediction that provides a definition and conceptual framework to study performativity in machine learning. A consequence of performative prediction is a natural equilibrium notion that gives rise to new optimization challenges. Another consequence is a distinction between learning and steering, two mechanisms at play in performative prediction. The notion of steering is in turn intimately related to questions of power in digital markets. We review the notion of performative power that gives an answer to the question how much a platform can steer participants through its predictions. We end on a discussion of future directions, such as the role that performativity plays in contesting algorithmic systems.
