COMPAS Algorithm Case Study
Metadata
Published: Feb 09, 2025
Tags: #🌐 learning-in-public artificial-intelligence ethical-ai bias-mitigation cognitive-science
The Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) is a prominent example of Algorithmic Bias in the criminal justice system. This case study demonstrates the real-world impact of biased AI systems on people’s lives.
Overview
COMPAS is a software tool developed by Northpointe (now Equivant) that:
- Assesses likelihood of defendant reoffending
- Used in various U.S. jurisdictions (New York, Wisconsin, California)
- Aids courts in decisions regarding:
- Bail
- Sentencing
- Parole
Bias Discovery
A 2016 ProPublica investigation revealed significant racial bias:
- Disproportionately labeled Black defendants as high-risk
- Higher false positive rates for Black defendants
- Lower accuracy in predictions for minority groups
Impact on Justice System
The COMPAS case highlighted several critical issues:
- Automated decision-making in sensitive contexts
- Lack of procedural justice
- Need for algorithmic fairness
- Transparency in AI systems
Key Findings
-
Predictive Accuracy:
- No more accurate than non-experts
- Demonstrated systematic bias against minorities
- Failed to achieve intended objective of fair risk assessment
-
Bias Manifestation:
- Black defendants twice as likely to be misclassified as high-risk
- White defendants more likely to be misclassified as low-risk
- Perpetuated existing systemic biases
Lessons Learned
-
Importance of Auditing:
- Regular evaluation of AI systems
- Independent third-party assessments
- Transparency in algorithmic decision-making
-
Need for Safeguards:
- Bias Mitigation Techniques
- Human oversight in critical decisions
- Clear appeals process
-
Policy Implications:
- Regulation of AI in sensitive domains
- Standards for algorithmic fairness
- Protection of individual rights