WHEN THE ALGORITHM DECIDES: ARTIFICIAL INTELLIGENCE, DIFFUSED LIABILITY, AND THE EMERGING ACCOUNTABILITY GAP IN AUTOMATED DECISION-MAKING

Abstract

Decisions that once turned on human judgement are now routinely produced by artificial intelligence: who receives credit, who is flagged to the police, whose application survives the first cut of a recruitment process. Legal liability, however, was built around a different picture of the world, one in which an identifiable actor commits a fault that can be traced to a harm. Against the distributed and frequently anonymous reality of how algorithmic systems are designed and deployed, that architecture strains badly. This article examines the lacuna that opens as a result, where harm has plainly occurred and a system is plainly implicated, yet no single actor can be made to answer under existing rules of tort, contract or data protection law. Taking the problem of many hands as its analytical frame, the article traces how responsibility for an automated decision fragments across data suppliers, model developers, fine-tuners and deployers, and how that fragmentation defeats the causal reasoning on which fault-based liability rests. It then surveys the comparative responses, among them Article 22 of the EU General Data Protection Regulation, the Court of Justice's ruling in Schufa, the EU Artificial Intelligence Act and the English case law on automated policing, before turning to Nigeria's fragmented position under the Nigeria Data Protection Act 2023 and orthodox tort principles. Nigeria, having no statute directed at algorithmic harm, is acutely exposed. The article proposes a layered legislative response built on mandatory algorithmic traceability, an operator-centred presumption of liability, and a compensation mechanism for residual harm, and concludes that closing the gap calls for deliberate legislative recalibration rather than further stretching of inherited common law doctrine.