Fairness, Accountability, and Transparency in Predictive Models for Criminal Justice

Explore a comprehensive lecture on fairness, accountability, and transparency in predictive modeling within criminal justice systems. Delve into Kristian Lum's research from the University of Pennsylvania, examining crucial examples that highlight these concepts' significance. Investigate the representativeness of police records in crime data, analyze drug crimes in Oakland through simulations and demonstrations, and understand the importance of fairness in law enforcement practices. Examine pre-trial risk assessments, including the Public Safety Assessment, and evaluate empirical risk estimates within decision-making frameworks. Discover the implications of risk assessment in supervised release programs and the benefits of hand selection. Gain insights into the validation process of predictive models and their impact on different racial groups. Conclude with a final summary emphasizing the critical nature of transparency in criminal justice applications of predictive modeling.

Intro
Are police records a representative sample of crime?
Drug Crimes in Oakland
Demo on Oakland Data
Simulation using Oakland Data
Why fairness?
Overbooking
Pre-Trial Risk Assessment
Public Safety Assessment
Empirical Risk Estimates
Decision-Making Framework
The Plan
Validation of our PSA reproduction code
Accounting for Discrepancies
Results by race
Why accountability?
Risk Assessment for Supervised Release Program
Who did the hand selection benefit?
Why transparency?
Final summary
The Process