Making Models that Matter: Computation, Gerrymandering, and the Search for Fairness, by Class of 1960s Speaker, Prof. Moon Duchin, Tufts University, Joint Math/Stat Kick Off Colloquium, Wednesday, September 11, 1-1:45 pm, Thompson Chemistry, Wege Auditorium, Room 123
Abstract: The algorithmic age is upon us! Complex, opaque algorithms are increasingly invoked in decision making, sometimes with clearly high stakes like policing, sentencing, medical diagnosis, and resource allocation and sometimes with seemingly low stakes like music and movie recommendation or driving routes. Many authors have raised the alarm about relying on algorithms for fairness and objectivity, pointing to manifestly unjust results. But these critical conversations tend to treat their algorithms as consummate black boxes, either because they are secret or because they are built in such a complicated fashion that their interpretability is limited. But what if we wrote the algorithms ourselves? Redistricting is a great example for this, because the unfairness of certain gerrymandered maps seems obvious, but figuring out precisely why and how to fix it requires a mix of different kinds of interdisciplinary work and thoughtful mathematics. Can a metric capture racial injustice or vote dilution? Can a computational or statistical model show you the relevant alternatives? I’ll use my team’s current work in redistricting to tell overlapping stories about model design and the logic of fairness.