97 Things About Ethics Everyone in Data Science Should Know

Chapter 53. Causality and Fairness—Awareness in Machine Learning

Scott Radcliffe

Managing Director, MS in Business Analytics Program, Emory University

It has become axiomatic that addressing fairness and bias in machine learning models is not optional. However, the race to deploy learning models has outpaced the development of standards and methods for detecting and systematically avoiding bias. This situation is due in some part to the fact that machine learning practice is typically not concerned with causality but rather is based on observational criteria. The focus is on prediction, classification, and identification. Observational criteria are fundamentally unable to determine whether a predictor exhibits unresolved discrimination.

A long history of data analysis in the social science and medical fields has shown that fairness should be studied from the causal perspective. In order to be fairness-aware, special emphasis is placed on the assumptions that underlie all causal inferences, the languages used in formulating those assumptions, the conditional nature of all causal and counterfactual claims, and the methods that have been developed for the assessment of such claims.

What is a “causal model”? Wikipedia provides a useful definition. A causal model (or structural causal model) is a conceptual model that describes the causal mechanisms of a system. Causal models can improve ...