Keywords: Demographic Parity, Equal Accuracy, Equal Opportunity, Fairness, Google, Group Thresholds, Group Unaware, Machine Learning
Author: David Weinberger
Publisher: What-If Tool
Publication Year: N/A
Summary: The following article discusses how Google created a “What-If” visualization tool to test machine learning systems and demonstrate how they perform in regards to different definitions of “fair”. This article outlines how the what-if tool assesses 5 different types of fairness: Group unaware, demographic parity, equal opportunity, equal accuracy, and group thresholds. Each concept is explained using a case study and a group of experts debating which definition of “Fair” to use. This resource outlines ideas that will come up during debates when using machine learning in the future and can help us understand what options we can weigh when making those decisions.