A Framework for Fairer Machine Learning in Organizations

TL;DR

This paper introduces a comprehensive framework for organizations to select and implement fair machine learning algorithms, addressing sources of unfairness, fairness tradeoffs, and behavioral ethics to prevent bias over time.

Contribution

It provides the first integrated framework combining ethics, fairness criteria, and organizational considerations for fair machine learning implementation.

Findings

Identifies sources of unfairness in organizational machine learning

Reviews fairness criteria and their tradeoffs

Proposes a practical framework for fair algorithm deployment

Abstract

With the increase in adoption of machine learning tools by organizations risks of unfairness abound, especially when human decision processes in outcomes of socio-economic importance such as hiring, housing, lending, and admissions are automated. We reveal sources of unfair machine learning, review fairness criteria, and provide a framework which, if implemented, would enable an organization to both avoid implementing an unfair machine learning model, but also to avoid the common situation that as an algorithm learns with more data it can become unfair over time. Issues of behavioral ethics in machine learning implementations by organizations have not been thoroughly addressed in the literature, because many of the necessary concepts are dispersed across three literatures: ethics, machine learning, and management. Further, tradeoffs between fairness criteria in machine learning have not been addressed with regards to organizations. We advance the research by introducing an organizing framework for selecting and implementing fair algorithms in organizations.