Incident 367: iGPT, SimCLR Learned Biased Associations from Internet Training Data

Description: Unsupervised image generation models trained using Internet images such as iGPT and SimCLR were shown to have embedded racial, gender, and intersectional biases, resulting in stereotypical depictions.

Tools

New Report New Response DiscoverView History

Entities

View all entities

Alleged: OpenAI and Google developed and deployed an AI system, which harmed gender minority groups , racial minority groups and underrepresented groups in training data.

Incident Stats

Incident ID

367

Report Count

Incident Date

2020-06-17

Editors

Khoa Lam

Incident Reports

Reports Timeline

An AI saw a cropped photo of AOC. It autocompleted her wearing a bikini.

technologyreview.com

technologyreview.com · 2021

Ryan Steed, a PhD student at Carnegie Mellon University, and Aylin Caliskan, an assistant professor at George Washington University, looked at two algorithms: OpenAI’s iGPT (a version of GPT-2 that is trained on pixels instead of words) and…

Variants

A "variant" is an incident that shares the same causative factors, produces similar harms, and involves the same intelligent systems as a known AI incident. Rather than index variants as entirely separate incidents, we list variations of incidents under the first similar incident submitted to the database. Unlike other submission types to the incident database, variants are not required to have reporting in evidence external to the Incident Database. Learn more from the research paper.