Conference Program

The digital transformation of public administration is often promoted in terms of efficiency and modernization, yet its ethical and democratic implications remain insufficiently examined. This paper presents an experimental project on the ethical deployment of Artificial Intelligence (AI) across public sector decision-making, a field exposed to integrity risks, procedural complexity, and fragmented data infrastructures.

The project proposes a model of “assistive AI” designed to enhance transparency, anomaly detection, and accountability without replacing human decision-making. The framework incorporates natural language processing (NLP) for automated document analysis, interpretable red-flagging systems for identifying irregularities, distributed ledger technologies for traceability, and public-facing dashboards to support civic oversight. Mandatory human supervision, algorithmic registries, and dual-layer rights to explanation—technical and civic—ensure AI operates as a support tool rather than a substitute for officials.

A comparative reflection considers the case of “Diella,” an avatar introduced in Albania as a digital minister for administrative oversight. While symbolically innovative, this example raises concerns about the personalization and depoliticization of algorithmic authority. In contrast, the proposed model emphasizes distributed institutional responsibility, structured oversight, and participatory governance, aligning with international initiatives such as the UK Ministry of Justice’s collaboration with the Alan Turing Institute on responsible AI in judicial systems.

Empirical evidence is drawn from a panel italian survey of public administration professionals and stakeholders (n=408), designed to assess perceptions of AI’s role in governance. Results indicate moderate trust in institutions, high demand for transparency in automated decisions, and strong opposition to fully automated decision-making without human oversight. While AI is perceived as a tool for improving integrity and efficiency, legitimacy is conditional upon explainability, auditability, and robust accountability mechanisms. Public acceptance correlates positively with transparency, contestability, and reversibility.

Preliminary pilot outcomes suggest improved anomaly detection, enhanced traceability of administrative processes, and increased accessibility of information for stakeholders. Persistent challenges include fragmented data systems, organizational resistance, and potential technological dependencies.

The study concludes that democratic renewal under algorithmic governance depends less on computational sophistication than on the design of accountability architectures. AI can strengthen integrity, transparency, and public trust only if embedded in open, auditable, and participatory frameworks. The key challenge is not whether algorithms can make decisions more efficiently, but whether institutions can govern algorithmic power in ways that preserve human responsibility, plural oversight, and democratic legitimacy.