Computation and the Search for Fairness in Representative Democracy

Abstract We‘ve all heard that algorithmic assistance is
increasingly used for high-stakes decisionmaking, from policing and sentencing
to medical diagnosis to resource allocation, sometimes with manifestly unjust
results. Often, these critical conversations treat algorithms as consummate
black boxes, either because the algorithms are proprietary or because they are
built on such complicated neural networks (say) that their interpretability is
severely limited. But what if we wrote the algorithms? Electoral redistricting
is an excellent problem domain for this inquiry it seems perfectly clear that
something is wrong and unfair about the way gerrymandered maps divide people
for the purposes of voting, but it's quite hard to locate the precise harm and
even harder to reason about remedies. I'll use the case of redistricting to
tell overlapping stories model design; Constitutional logic; and metrics of
fairness.

Bio Moon
Duchin is an associate professor of Mathematics and Senior Fellow in the Tisch
College of Civic Life at Tufts University. She serves as director of the
interdisciplinary program in Science, Technology, and Society and as
collaborating faculty in the Department of Race, Colonialism, and Diaspora
Studies. Her mathematical subfields are geometry, topology, group theory, and
dynamical systems. Her current research focus is in the study of electoral
redistricting in the U.S., using Markov chain Monte Carlo and other randomized
algorithms to understand relationships between community, partisanship, race,
and representation.