Author: Michael Kearns, Aaron Roth
Publisher: Brookings
Publication Year: 2020
Summary: The following article is focused on frameworks for thinking about algorithmic bias and how to address the “unanticipated consequence of following the standard methodology of machine learning: specifying some objective (usually a proxy for accuracy or profit) and algorithmically searching for the model that maximizes that objective using colossal amounts of data.” The article argues that in order to properly address algorithmic bias there must be both 1). A systematic approach for discovering “bad behavior” by algorithms before they can cause harm and 2). A rigorous methodology for correcting it. The crux of the argument is that most issues that we face with algorithmic bias today are only addressed in an ad-hoc way after the fact.