Establishing Governance Models for Bias and Fairness Management in Dynamic AI Analytics Pipelines

Organizations are increasingly adopting AI analytics pipelines for decision-making in critical areas, both for real-time and historical data. These dynamic pipelines present ever-changing ‘risk surfaces,’ which static governance frameworks are unlikely to manage effectively. This study focuses on governance frameworks for identifying, addressing, and maintaining bias and fairness throughout the entire life cycle of dynamic AI analytics pipelines, including those for predictive analytics and AI models. Using descriptive analytics of current practice, effectiveness perception, common frameworks, and gaps which lead to bias, gaps which organizations and their data scientists, ML engineers, and governance professionals have, and an integrated quantitative survey using 100 population which targeted at data scientists, ML engineers, governance officers, and AI governance professionals, the study applies descriptive analytics to unify the diverse conceptualizations of bias and privileges. Most organizations' governance frameworks are geared toward the analytics lifecycle; hence, the proposed governance layers for dynamic AI comprise policy, tooling, monitoring, accountability, and remediation to address gaps in the established framework, including continuous presence, dynamic governance, and the operationalization of these layers.