AI Prediction as Domination in Public Institutions - A Neorepublican Perspective

This paper examines the political implications of AI-based predictive optimization systems in public institutions through the lens of neorepublican political theory. While current debates about algorithmic decision-making mainly focus on fairness and bias from an egalitarian liberal perspective, this analysis argues that example-based machine learning systems present a fundamental challenge to democratic governance and political freedom, necessitating an additional perspective. Drawing on neorepublican conceptions of freedom as non-domination, the paper demonstrates how predictive optimization constitutes a novel form of uncontrolled power that evades traditional mechanisms of democratic accountability and contestation. The argument proceeds in three steps. First, it distinguishes between rule-based and example-based reasoning in algorithmic decision-making, highlighting how machine learning systems derive authority from statistical patterns rather than explicit rules. Second, it establishes that while all forms of profiling by public authorities create power asymmetries, example-based predictive optimization represents a distinctly problematic form of authority due to its inherent opacity and resistance to meaningful contestation. Third, it identifies three specific challenges these systems pose to neorepublican ideals: the unavoidable absence of common knowledge regarding decision rationales, the technocratic depoliticization of governance, and the practical impossibility of meaningful contestation. The paper concludes that from a neorepublican perspective, rule-based decision-making systems should be preferred over example-based predictive optimization in public institutions, particularly for high-stakes decisions. It suggests that recent work on interpretable machine learning, specifically the possibility of reducing complex models to simpler rule-based algorithms, offers a promising technical direction more aligned with neorepublican democratic values. However, it cautions that rule-based algorithms may also not meet the full requirements of legitimate democratic governance, especially given the fundamental unpredictability of human behaviour.