The AI Stratification Spiral: How Predictive Systems Turn Historical Inequality into Statistical Common Sense

This preprint develops the AI Stratification Spiral, a sociological framework for explaining how predictive systems can reproduce and intensify existing inequalities over time. The framework identifies a recursive process linking unequal data production, algorithmic interpretation, institutional allocation, and confirmatory data generation. When predictions influence access to opportunities, resources, scrutiny, or support, the resulting outcomes can generate new data that appear to validate the original prediction. In this way, historically contingent inequalities may gradually acquire the appearance of objective statistical evidence. The article connects debates on artificial intelligence, algorithmic governance, social stratification, cumulative disadvantage, and institutional decision-making, while emphasizing that these processes are not technologically inevitable but depend on how predictive systems are designed, deployed, and governed.