The Epistemological Dilemma of Algorithmic Justice: What is Lost When Law Becomes ‘Computable’?

Abstract The rapid and transformative evolution of artificial intelligence—from predictive recidivism algorithms to Large Language Models (LLMs) capable of passing the Uniform Bar Exam—has reinvigorated the ambition to render law ‘computable’. This article argues that this project, even in its most advanced generative forms, rests on a fundamental epistemological category error. We contend that the operational logics of AI—whether the rigid formalism of symbolic systems or the probabilistic mimicry of ‘stochastic parrots’—are structurally incompatible with the nature of legal reasoning. Synthesizing insights from the Anglo-American liberal jurisprudential tradition (Hart, Dworkin, and the Legal Realists), we demonstrate that law is an open system structurally reliant on linguistic purpose, moral commitment, and social context. We undertake a critical deconstruction of contemporary AI paradigms, showing how Generative AI’s ‘hallucinations’ and lack of intentionality represent a failure not of technical performance, but of the communicative engagement required for democratic legitimacy. What is irretrievably lost in the computational translation are the core judicial capacities of practical wisdom ( phronesis ), narrative integrity ( nomos ), and situated social intelligence. The article concludes that AI’s ontological limitations, when properly understood, reveal its true role: not to replace the human judge in adjudication, but to serve as an epistemic foil designed to augment, rather than amputate, the uniquely human burden of judgment.