AI-DRIVEN DECISION MAKING IN PUBLIC ADMINISTRATION: ETHICS, ACCOUNTABILITY, AND GOVERNANCE

Abstract

Artificial intelligence is rapidly transitioning from a peripheral tool to a central decision-making actor within public administration systems globally. Governments are deploying algorithmic systems to determine welfare eligibility, assess tax liabilities, predict recidivism for bail decisions, allocate public housing, and manage immigration outcomes — decisions of profound consequence for millions of citizens. Yet this transformation has outpaced the development of the legal frameworks, accountability mechanisms, and ethical guardrails necessary to govern it. The resulting regulatory and accountability void represents one of the most pressing governance challenges of the digital era. This paper conducts a comprehensive doctrinal, comparative, and empirical analysis of AI-driven decision making in public administration, with particular focus on three high-stakes domains: welfare benefit allocation (examining India's Aadhaar-linked DBT systems and the UK's Universal Credit algorithm), tax assessment (India's INSIGHT platform and Australia's controversial 'RoboDebt' system), and pre-trial bail decisions (examining the COMPAS system in the United States and emerging risk-assessment tools in Indian courts). Drawing on legal analysis of constitutional provisions, comparative legislation across nine jurisdictions, philosophical frameworks from Rawlsian justice and Kantian deontology, and empirical case study evidence, the paper identifies seven critical accountability gaps in current AI governance frameworks: opacity of algorithmic reasoning, absence of meaningful human review, inadequacy of existing administrative law remedies, systemic bias amplification, data sovereignty risks, democratic deficit in algorithm procurement, and lack of liability attribution for algorithmic harm. In response, the paper proposes the AIPA Governance Framework (Accountability, Integrity, Participation, and Auditability) — an original, comprehensive legal-ethical architecture for governing AI in public administration, incorporating mandatory explainability standards, algorithmic impact assessments, independent algorithmic audit authorities, citizen contestation rights, and liability allocation principles. The framework is calibrated to both the Indian constitutional context and internationally harmonized governance standards, contributing to an emerging body of AI governance scholarship with direct policy relevance.