Amazon’s Experimental Hiring Tool Allegedly Displayed Gender Bias in Candidate Rankings

August 10, 2016

Between 2014 and 2017, Amazon developed an AI-powered recruiting tool that reportedly displayed gender bias in candidate rankings. Trained on resumes largely from men for a decade, the system learned to favor male candidates over female ones, penalizing terms such as 'women's' and graduates from certain all-women's colleges. Despite attempts to remove these biases, fairness was not guaranteed. The project, which recruiters never solely relied upon, was eventually abandoned.

For those interested in shaping the future of safe and secure AI practices, we invite you to join HISPI Project Cerebellum and contribute to our efforts in Govern, Map, Measure, or Manage AI incidents such as this one. JOIN US. Let's work together to establish guardrails for AI and ensure harm prevention.

Matched TAIM controls

Suggested mapping from embedding similarity (not a formal assessment). Browse all TAIM controls

GOVERN 3.1 — similarity 0.623, rank 1. TAIM detail and related incidents →
MEASURE 2.5 — similarity 0.620, rank 2. TAIM detail and related incidents →
MAP 1.6 — similarity 0.617, rank 3. TAIM detail and related incidents →

Alleged deployer: amazon
Alleged developer: amazon
Alleged harmed parties: amazon-applicants, women-applying-to-amazon

Source

Data from the AI Incident Database (AIID). Cite this incident: https://incidentdatabase.ai/cite/37

Data source

Incident data is from the AI Incident Database (AIID).

When citing the database as a whole, please use:

McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. In Proceedings of the Thirty-Third Annual Conference on Innovative Applications of Artificial Intelligence (IAAI-21). Virtual Conference.

Pre-print on arXiv · Database snapshots & citation guide

We use weekly snapshots of the AIID for stable reference. For the official suggested citation of a specific incident, use the “Cite this incident” link on each incident page.