Skip to main content


Showing posts from July, 2019

Mismeasuring algorithmic fairness #pop

This post belongs to LoPSE popularization (#pop) series written both in English and in Polish.  ENGLISH VERSION [POLISH VERSION BELOW] Say we want to treat people equally - what does it even mean? This question pops up quite often in political science, economics, policy considerations, or local government decisions. Since AI is often involved in the decision-making process, a puzzle for a mathematical philosopher arises: how do you define fairness so that the definition is precise enough to be applicable to predictive algorithms and so that philosophically it still seems plausible? It turns out the task is non-trivial. Three attempts of such explications are discussed in The Measure and Mismeasure of Fairness: A Critical Review of Fair Machine Learning   by Sam Corbett-Davies and Sharad Goel: anti-classification is the requirement to exclude protected attributes such as gender or race in the decision process. classification parity  requires that common measures of pr