Responsible AI Adoption in the Public Sector: A Data-Centric Taxonomy of AI Adoption Challenges

Data quality is foundational, not optional: The paper identifies poor data quality as a key challenge preventing responsible AI adoption in government. Without addressing this technical foundation first, fairness and accountability frameworks remain "largely aspirational" rather than practically achievable.

Responsible AI requires more than good intentions: Current responsible AI frameworks emphasizing fairness, accountability, and transparency often overlook the institutional and governance structures needed to implement them. The study found that weak governance and misaligned human-AI decision-making processes are critical barriers that must be addressed alongside ethical principles.

Government needs AI-ready infrastructure before deployment: The taxonomy identifies limited AI-ready infrastructure as a major obstacle. Public sector organizations attempting to adopt AI without adequate technological foundations risk high-risk deployments that could harm citizens and erode public trust.

A diagnostic tool can help identify risky AI projects early: The paper's taxonomy of 13 challenges across technological, organizational, and environmental dimensions serves as a practical diagnostic tool. Policymakers can use it to surface "symptoms" of problematic AI deployments before they cause harm, rather than addressing problems after implementation.

Sustainability and economic viability matter for responsible AI: Beyond fairness and transparency, the study highlights that economic and environmental sustainability concerns are integral to responsible AI adoption in government, suggesting that truly responsible AI must be economically viable and environmentally conscious.