Unmasking Algorithmic Bias in Predictive Policing: A GAN-Based Simulation Framework with Multi-City Temporal Analysis

This paper examines racial bias in predictive policing systems by building a simulation framework that uses AI (specifically GANs) to track how bias flows through the entire police enforcement process, from crime prediction to actual police contact. Using crime data from Baltimore and Chicago combined with demographic information, the researchers found extreme racial disparities—for example, Black residents were dramatically under-detected in some cases—and showed that while AI debiasing techniques can help somewhat, they cannot solve the problem without policy changes. The analysis reveals that officer deployment levels have the biggest impact on these disparities, confirming that racial composition of neighborhoods strongly predicts policing intensity.