When the Algorithm Says No: Artificial Intelligence, Welfare Delivery, and Accountability in the Global South

Abstract

Artificial intelligence has entered development practice across most of the Global South through one channel above all; the automation of welfare delivery. Biometric authentication, reduplication algorithms and eligibility scoring now sit on top of what is called digital public infrastructure, or DPI. This article draws on doctrinal legal analysis and published field evidence from India to examine what happens when these systems come between the poorest citizens and the entitlements owed to them. India's regulatory framework, including the “Digital Personal Data Protection Act 2023”, governs data flows. Five safeguards are proposed here, each implementable without new legislation: a human fallback guarantee, exclusion audits, an explainability floor, grievance parity, and procurement transparency. As India exports its DPI model abroad, its accountability gaps travel with it. The question, in the end, is not innovation versus regulation. It is who bears the burden when an algorithm gets it wrong.