The Challenge
Public sector adoption of artificial intelligence remains uneven and fraught with risk despite its transformative potential. Current responsible AI frameworks emphasize high-level principles like fairness, accountability, and transparency, yet remain largely aspirational. They consistently overlook critical technical, organizational, and institutional realities, particularly foundational data structures and governance mechanisms. This study addresses this strategic gap by tackling the disconnect between abstract ethical principles and the ground-level data realities of government agencies, investigating why institutional pressures and inadequate infrastructure routinely impede safe, equitable, and efficient AI deployment in public administration.
Core Findings
The study develops a comprehensive data-centric taxonomy of AI adoption challenges through a rigorous systematic literature review of 43 studies combined with 21 expert evaluations. Operating across technological, organizational, and environmental dimensions, the resulting taxonomy identifies 13 key challenges. These encompass acute operational hurdles such as poor data quality, limited AI-ready infrastructure, weak data governance, and misalignments in human-AI decision-making, alongside broader economic and environmental sustainability concerns. Furthermore, the taxonomy is annotated with institutional pressures, functioning as an applied diagnostic tool designed to surface early warning symptoms of high-risk public sector AI deployments.
Strategic Takeaway
For digital leaders and socio-technical scholars, this research shifts the AI governance discourse from philosophical ethics to operational engineering. It provides organizational leaders with a concrete diagnostic framework to audit their institutional readiness before procuring or scaling algorithmic systems. By mapping data-centric bottlenecks against environmental and organizational pressures, digital leaders can systematically bridge the gap between high-level policy mandates and front-line data execution. The taxonomy serves as an actionable roadmap for building resilient data infrastructure and governance conditions essential for sustainable, responsible AI integration within complex public sector bureaucracies.