Bias In, Bias Out? Evaluating the Folk Wisdom

TL;DR

This paper investigates how biases in human decision-making influence algorithmic fairness, revealing conditions where algorithms can either inherit or reverse biases present in training data.

Contribution

It provides a theoretical framework explaining bias reversal phenomena and clarifies conditions affecting bias inheritance or reversal in algorithms.

Findings

Bias reversal occurs when biased decision-makers influence training data.

Bias inheritance or reversal depends on how decision-makers affect data and labels.

Simulation confirms theoretical predictions using NYC Stop and Frisk data.

Abstract

We evaluate the folk wisdom that algorithmic decision rules trained on data produced by biased human decision-makers necessarily reflect this bias. We consider a setting where training labels are only generated if a biased decision-maker takes a particular action, and so "biased" training data arise due to discriminatory selection into the training data. In our baseline model, the more biased the decision-maker is against a group, the more the algorithmic decision rule favors that group. We refer to this phenomenon as "bias reversal." We then clarify the conditions that give rise to bias reversal. Whether a prediction algorithm reverses or inherits bias depends critically on how the decision-maker affects the training data as well as the label used in training. We illustrate our main theoretical results in a simulation study applied to the New York City Stop, Question and Frisk dataset.