Predictive Policing Bias

The Challenge

Predictive policing and risk assessment algorithms are increasingly used in criminal justice systems worldwide to forecast crime hotspots, allocate police resources, and assess individuals' risk of reoffending. However, these systems often encode and amplify existing biases in criminal justice data and practices. Key concerns include:

• Data Bias Amplification: Algorithms trained on historically biased policing data perpetuate and potentially amplify discriminatory patterns in enforcement and sentencing.
• Feedback Loops: Predictive systems create self-reinforcing feedback loops where increased policing in predicted "high-risk" areas leads to more arrests, which further reinforces the algorithm's focus on those areas.
• False Risk Labeling: Risk assessment tools can incorrectly label individuals as high-risk based on correlations with demographic factors rather than causal relationships to criminal behavior.
• Opacity and Unaccountability: Many predictive systems operate as "black boxes," making it difficult for defendants, communities, and even system operators to understand or challenge their decisions.

Our Approach

The Global Tech Governance Institute takes a justice-centered approach to addressing predictive policing bias:

• Bias Auditing: Developing methodologies and tools to detect, measure, and mitigate bias in predictive policing and risk assessment algorithms.
• Transparency Frameworks: Creating standards and guidelines for algorithmic transparency and explainability in criminal justice contexts.
• Community Governance: Researching and promoting models for community oversight and governance of algorithmic policing systems.
• Rights-Based Approaches: Developing legal and policy frameworks that protect due process, equal protection, and other fundamental rights in algorithmic criminal justice systems.

Current Initiatives

Our work in this area currently includes:

Matrix Integration

Key Publications