TLDR: This paper introduces a data-centric taxonomy of 13 challenges for responsible AI adoption in the public sector. Based on a systematic literature review and expert validation, it categorizes these challenges across technological, organizational, and environmental dimensions, integrating the Technology-Organization-Environment (TOE) framework and Institutional Theory. The study highlights that responsible AI adoption depends heavily on robust data ecosystems and strong governance, serving as a diagnostic tool for policymakers to address structural and institutional barriers.
Artificial Intelligence (AI) holds immense promise for transforming public sector services, enhancing decision-making, and boosting administrative efficiency. However, its adoption remains inconsistent, largely due to a complex interplay of technical, organizational, and institutional hurdles. A recent study delves into this critical area, proposing a data-centric taxonomy to understand the challenges of responsible AI adoption in government.
The research, titled Responsible AI Adoption in the Public Sector: A Data-Centric Taxonomy of AI Adoption Challenges, highlights that while responsible AI frameworks emphasize fairness, accountability, and transparency, they often fall short by overlooking the foundational realities of data and governance. The authors, Anastasija Nikiforova, Martin Lnenicka, Ulf Melin, David Valle-Cruz, Asif Gill, Cesar Casiano Flores, Emyana Sirait, Mariusz Luterek, Richard Michael Dreyling, and Barbora Tesarova, argue that without addressing data limitations, truly responsible AI adoption is unattainable.
The study introduces the concept of “data-AI challenges” – systemic frictions that emerge at the intersection of AI deployment and Public Data Ecosystems. These are not merely technical issues but are multidimensional, socio-technical, and deeply embedded within institutions, directly influencing whether AI can be adopted responsibly.
A Comprehensive Approach to Understanding Challenges
To address this gap, the researchers developed a taxonomy based on a systematic review of 43 academic studies and validated it through evaluations from 21 experts. This robust methodology allowed for the identification and structuring of key challenges. The taxonomy is framed using two powerful theoretical lenses: the Technology-Organization-Environment (TOE) framework and Institutional Theory.
The TOE framework helps categorize challenges across three dimensions:
- Technology: This includes internal and external infrastructure, tools, data quality, and capabilities that enable or restrict AI use.
- Organization: This covers characteristics of the public entity, such as its size, structure, leadership, culture, human capital, and resources, which shape its capacity for change.
- Environment: This encompasses external pressures like regulatory frameworks, professional norms, inter-organizational dynamics, and public policy.
Institutional Theory further enriches this understanding by explaining why certain challenges persist. It identifies three types of pressures:
- Coercive pressures: Stemming from formal legal, regulatory, or financial mandates (e.g., data privacy laws).
- Normative pressures: Arising from shared professional ethics and norms (e.g., fairness, transparency).
- Mimetic pressures: Referring to the imitation of perceived best practices from other agencies, often under uncertainty.
By combining these frameworks, the study offers an integrated view of both the structural and institutional roots of data-related challenges in public sector AI adoption.
The 13 Key Data-AI Challenges
The refined taxonomy identifies 13 key challenges, categorized by their dominant TOE dimension and influenced by various institutional pressures:
- Technological Challenges: These include difficulties in accessing suitable data sources, limited computing and data storage resources, challenges in acquiring and maintaining modern AI systems and platforms, poor AI-ready data quality (inconsistency, bias, lack of metadata), and issues with algorithm transparency and standardization.
- Organizational Challenges: These involve weak governance and management of data and AI systems, misalignment between human judgment and machine logic in decision-making, uncertainty over the fairness and reliability of AI-generated decisions, and a lack of awareness and technical competence (data and AI literacy) among public officials and stakeholders.
- Environmental Challenges: These encompass resistance to inter-organizational data sharing, inadequate data and AI security and privacy measures, high economic costs and uncertain financial returns, and often-overlooked environmental sustainability concerns related to AI’s energy consumption.
The study also identifies cross-cutting challenges, such as data and AI security, privacy, and ethics, and the impacts on decision-making and stakeholder literacy, which bridge organizational and environmental domains, requiring both internal safeguards and compliance with external norms.
Also Read:
- AI Accelerates Public Funding Application Reviews in Europe
- Rethinking AI Ethics: Why Current Evaluation Methods Fall Short in Measuring Systemic Harms
Implications for Responsible AI Adoption
This taxonomy serves as a practical diagnostic tool for policymakers and public administrators. It helps them pinpoint systemic obstacles, prioritize interventions, and align AI initiatives with existing organizational capabilities and diverse stakeholder needs. Rather than asking whether AI systems themselves are inherently responsible, the study shifts the focus to whether public administrations are positioned to adopt AI responsibly – in ways that uphold legality, accountability, fairness, sustainability, and inclusion.
By making visible the preconditions and trade-offs, the taxonomy complements ongoing debates on trustworthy AI, fair AI, and sustainable AI. It highlights that these high-level principles risk remaining aspirational unless connected to the practical realities of data ecosystems and public institutions. The research underscores the need for holistic strategies that embed governance, collaboration, and ethical data stewardship in the design and implementation of AI initiatives.
