Abstract
AI-driven decision-making is often evaluated in terms of individual human judgment - are algorithms faster, fairer or less biased than humans? In this talk, I challenge this framework, arguing that the most appropriate class of comparison for AI-driven decision-making is not individual human judgment, but bureaucratic decision-making. Like bureaucratic systems, AI-driven decision-making depends on quantification, standardization, depersonalized authority and specialized tasks. It does not remove human discretion entirely, but distributes it throughout the process, from data selection and model design to implementation and supervision. Both bureaucracies and AI aim for consistency and impartiality, but they are also criticized for their opacity, rigidity and insensitivity to context. In this talk, I will show how moving from a human vs. machine framework to an algorithm vs. bureaucracy framework opens up new and more productive ways of thinking about accountability, discretion and legitimacy in AI systems.
