Performative Power

Performative power describes how predictive models, once deployed, influence the very data they aim to predict, creating feedback loops that can lead to inaccurate or biased outcomes. Current research focuses on understanding and mitigating these effects, particularly in machine learning, exploring techniques like model retraining with regularization and developing methods to anticipate and correct for performative shifts in various contexts, including time-series forecasting and social prediction problems. This research is crucial for improving the reliability and fairness of AI systems across diverse applications, from algorithmic fairness to economic modeling, by addressing the inherent power dynamics created by these feedback loops.