Progressive Regression: How AI Systems Reverse Social Progress

This paper introduces and defines Progressive Regression: the process by which AI systems trained on historically biased data encode past discrimination as learned pattern, automate it at scale, feed the distorted outputs back into institutional systems as ground truth, and through successive cycles amplify rather than reduce the original inequity. The concept identifies a four-stage cycle (Encoding, Amplification, Automation, New Institutional Data Fed Back as Ground Truth) through which AI deployment reverses decades of social progress without political intent, through the structure of training data and automated feedback alone. The paper distinguishes Progressive Regression from adjacent concepts including model collapse, dataset drift, and bias amplification, provides an evidence base drawn from peer-reviewed research and documented real-world cases across hiring, criminal justice, child welfare, and healthcare, and positions the concept within a broader theoretical framework alongside AI Meta-Bias and AIDE. A research proposal for empirical validation is included. This is a pre-print working paper submitted to establish a citable, DOI-indexed record of the concept definition and research proposal. The full empirical paper is in preparation. Related deposits: AI Meta-Bias (https://doi.org/10.5281/zenodo.19187752) and AI Iterative Distortion Effect / AIDE (https://doi.org/10.5281/zenodo.19204858).