BiasMedium
iGPT, SimCLR Learned Biased Associations from Internet Training Data
- Occurred
- Jun 17, 2020
- Published
- Jan 29, 2021
- Source
- AI Incident Database
Summary
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.
Developers
- OpenAI
Deployers
- OpenAI
Harmed parties
- underrepresented groups in training data
- Minority groups
Classification
Category confidence: 95%
Severity confidence: 90%
Rationale: Research paper demonstrating embedded racial and gender biases in unsupervised image generation models.
Classifier: Qwen/Qwen3.5-35B-A3B-FP8