Fair decision making in modern AI systems

AI systems increasingly help decide how scarce resources are distributed: who gets matched to an opportunity, which regions receive service, whose needs are prioritized, or how limited capacity is used. These decisions are rarely isolated. They unfold over time, under constraints, and across people and communities with different needs. This talk examines fair and responsible decision making in resource allocation and in language models, drawing on examples from multiple domains including ridesharing, homelessness prevention and mental health. I will discuss how unfairness and bias can emerge from data, models, and allocation mechanisms, with examples from my own research on how we detect and mitigate such unfairness.