Beyond Automation: Understanding Fairness, Ethics, and Human Discretion in AI-driven Societal Decisions

Abstract

My doctoral research investigates the ethical readiness and societal implications of deploying AI in high-stakes resource allocation. This work bridges empirical analysis of AI capabilities with computational modeling of human decision-making and theoretical explorations of long-term fairness. Through a multi-method approach, I first evaluate the reliability of Large Language Models (LLMs) in a real-world homelessness services context, revealing critical inconsistencies. I then use interpretable machine learning to model the sophisticated discretion of human caseworkers, demonstrating that their non-formulaic judgments are systematic and effective. Finally, I use agent-based simulations to show how repeated, ostensibly "fair" allocations can entrench group-level inequities over time. Collectively, these findings argue for caution in autonomous AI deployment, highlight the value of human-in-the-loop systems, and call for a more dynamic understanding of fairness in sociotechnical systems.