AI-enabled Predictive Policing with Ethical Oversight and Community Feedback

Thesis Abstract

The rapid advancement of artificial intelligence (AI) in law enforcement has heightened concerns about civil liberties, algorithmic bias, and community trust, prompting a critical examination of predictive policing systems that operate under ethical oversight and active community feedback mechanisms. This study investigates how AI-enabled predictive policing, when coupled with structured ethical governance and participatory feedback from communities, influences policing effectiveness, legitimacy, and equity. The aim is to determine whether ethical oversight and community input can mitigate bias and enhance problem-solving outcomes without compromising safety. Specific objectives include (1) evaluating the predictive accuracy and bias indicators of risk-scoring models using historical crime and arrest data from a metropolitan police department; (2) assessing the impact of an ethics review framework on algorithmic transparency and accountability, including the role of an independent ethics board and impact assessments; (3) examining community feedback processes, channels, and their effect on police deployment decisions and perceptions of procedural justice; (4) testing the relationship between transparency, community engagement, and perceived legitimacy using a mixed-methods design; and (5) providing policy and practice recommendations for scalable, rights-respecting predictive policing. A mixed-methods design integrates quantitative and qualitative strands to address the research questions. The quantitative component analyzes a dataset comprising three years of crime records, 2,400 validated predictive alerts, and deployment outcomes from a large urban police department, applying machine-learning models such as gradient boosting and random forests for risk prediction, with fairness-aware techniques (e.g., disparate impact analysis, equalized odds) to detect and mitigate bias. Regression analyses and structural equation modeling (SEM) will test hypothesized relationships among algorithmic performance, ethical oversight metrics, deployment decisions, community satisfaction, and perceived legitimacy. The qualitative component collects semi-structured interviews (n=40) with police practitioners, members of the Ethics Review Panel, and community representatives, alongside focus groups (n=6) with residents from diverse neighborhoods. Thematic analysis and coding will identify patterns related to transparency, accountability, and legitimacy, while triangulation with policy documents, ethics guidelines, and incident reports will strengthen interpretive validity. The study adheres to ethical standards through informed consent, data anonymization, and oversight by an independent ethics committee. Key expected findings include (i) improved predictive performance when governance and transparency constraints are applied, without a statistically significant increase in false positives; (ii) reductions in disparate impact indicators for protected groups after implementing fairness-aware modeling and ongoing bias audits; (iii) enhanced perceptions of procedural justice and legitimacy among community participants who engage in structured feedback processes, compared with controls lacking such mechanisms; (iv) clearer accountability pathways and greater operational justification for deployment decisions, facilitated by the ethics board’s oversight and public-facing dashboards; and (v) practical constraints and trade-offs identified between algorithmic complexity, timeliness of decisions, and community trust. The study contributes to knowledge by empirically examining the intersection of AI-driven policing, ethical governance, and community participation, filling gaps on how oversight and feedback loops influence both effectiveness and civil rights outcomes. It advances theoretical understanding by integrating legitimacy theory with algorithmic accountability and participatory governance, offering a framework for evaluating predictive policing systems beyond accuracy alone. Policy implications include guidelines for designing ethics boards, transparency measures, and community consultation protocols that are scalable across jurisdictions. The conclusions emphasize that AI-enabled predictive policing can be compatible with democratic norms when paired with rigorous ethical oversight and sustained community engagement, yet require ongoing monitoring, independent auditing, and adaptive governance to mitigate emergent biases. Recommendations address model governance, data stewardship, inclusive citizen involvement, and continuous performance and fairness auditing to ensure responsible deployment and enduring public trust.