Artificial intelligence, judicial archives, and transitional justice: analysis of the Quiñones Case in Chile

Abstract

This article analyzes the use of Artificial Intelligence (AI) for reconstructing historical truth from the archives of Transitional Justice (TJ) in Chile, focusing on the Quiñones Case – a forced disappearance of 1976 documented across 13,283 judicial records. Through a sociotechnical approach, it proposes the Graph-based Retrieval-Augmented Generation (GraphRAG) methodology to transform fragmented case files into navigable epistemic infrastructures. By integrating large language models (LLMs) with knowledge graphs, the system identifies patterns of state violence with an accuracy of 91.7% based on expert-validated data. The study develops a deep critique of techno-solutionism, warning about epistemological risks such as the illusion of understanding and the algorithm of silence, where algorithmic fluency could obscure evidentiary gaps. In response, it advocates for a Human-in-the-Loop (HITL) model as a form of epistemic governance: AI does not replace expert interpretation but enhances it, allowing human validation to focus on relational evidence. Operating as a situated infrastructure for knowledge production from the Global South, GraphRAG transforms the archive into a traceable and contestable object, thereby contributing to broader processes of memory, accountability, and reparation in Latin American TJ.