Interpreting Criminal Charge Prediction and Its Algorithmic Bias via Quantum-Inspired Complex Valued Networks

The study addresses a critical transparency problem in criminal justice AI: Predictive policing systems often lack explainability, making it difficult to identify whether decisions are biased. This paper uses "post-hoc explanations" and attention mechanisms to make the decision-making process more understandable to users and stakeholders.

Criminal history, not demographic characteristics, should drive predictions: The researchers found that criminal histories are "statistically significant factors" in predicting future charges, while demographic identifiers like race and age are not—suggesting the model can be designed to avoid perpetuating racial and age-based disparities in the criminal justice system.

The model reveals gradual rather than sudden escalation in crime severity: The analysis indicates that suspects tend to gradually increase crime severity over time rather than suddenly committing more serious offenses, which could inform more nuanced intervention and rehabilitation strategies.

Technical innovations can improve both accuracy and fairness: By combining Bi-LSTM networks, attention mechanisms, and quantum-inspired complex-valued networks, the researchers achieved "consistent and reliable prediction precision and recall" while simultaneously improving interpretability—demonstrating that accuracy and transparency need not be trade-offs.

Explainability is essential for public trust in criminal justice AI: The paper emphasizes that showing why an algorithm makes a prediction (through feature importance analysis) is crucial for addressing public concerns about bias and building confidence in automated decision-making systems used in the justice system.