David Robinson and Harlan Yu
The key decisions that shape people’s lives -- decisions about jobs, healthcare, housing, education, criminal justice and other key areas -- are, more and more often, being made by machine learning systems. As a result, a growing number of important conversations about civil rights, which focus on how these decisions are made, are also becoming discussions about machine learning. Policymakers and the public increasingly want to understand how machine learning systems work in general, how they reach particular decisions, and (in some cases) how their operation might be altered to advance social goals.
Some political requirements for automated decisions -- such as a requirement that decisions be reached using a consistent procedure, or be intelligible to human observers -- may lend themselves to technical solutions. But in the area of non-discrimination, U.S. law often avoids bright line rules, proceeding instead on a holistic, case-by-case basis. The kind of rules that engineers may need, in other words, may not exist today. Those rules may need to be developed anew, in a policy discussion that is informed by a deeper understanding of technical methods. This is both a challenge and an opportunity for the field of machine learning to help shape the future of civil rights.