Author: Alexandra George
Publisher: Carnegie Mellon University
Publication Year: 2018
Summary: The following article discusses how some variables can be proxies for variables that are protected classes such as race or gender. It gives an example of how zip code can be a proxy for race and how gang membership can be a proxy for both race and gender. The author gives their solution for combating these in algorithms. They created a proxy detection algorithm that goes through and analyzes the variables in the model for correlation with protected features like race or age then returns its results to a person. They chose to incorporate a human in the last step because not all instances of proxy use are negative and they can be justified in certain circumstances.