iGPT, SimCLR Learned Biased Associations from Internet Training Data

June 17, 2020

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. The research highlights the need for responsible AI practices, including careful evaluation of training data and deployment strategies, to prevent harm from biased AI models. And to help build a safer AI ecosystem, visit JOIN US.

Matched TAIM controls

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

Alleged deployer
OpenAI, Google
Alleged developer
OpenAI, Google
Alleged harmed parties
underrepresented groups in training data, Minority groups

AI governance case studies

For forensic AI governance failure analysis (TAIMScore™ case studies), browse Human Signal’s Failure Files™.

Source

Data from the AI Incident Database (AIID). Cite this incident: https://incidentdatabase.ai/cite/367

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.