Observing and Recommending from a Social Web with Biases

TL;DR

This paper investigates how to detect and mitigate unlawful discrimination in black box algorithms used in social web platforms, focusing on biases related to protected characteristics like gender and race.

Contribution

It introduces methods for observing and recommending adjustments to identify and reduce biases in opaque algorithms on social web platforms.

Findings

Algorithms can exhibit significant biases against protected groups.

Proposed methods improve detection of discriminatory patterns.

Recommendations help mitigate bias in algorithmic decision-making.

Abstract

The research question this report addresses is: how, and to what extent, those directly involved with the design, development and employment of a specific black box algorithm can be certain that it is not unlawfully discriminating (directly and/or indirectly) against particular persons with protected characteristics (e.g. gender, race and ethnicity)?