OpenAI's GPT-3 Associated Muslims with Violence

August 6, 2020

Users and researchers revealed generative AI GPT-3 associating Muslims to violence in prompts, resulting in disturbingly racist and explicit outputs such as casting Muslim actor as a terrorist.

The incident highlights concerns about the training data used by OpenAI and the potential for harmful biases in large language models. It also underscores the need for more effective safeguards against hate speech and discriminatory content. In an effort to address these issues, researchers are exploring ways to improve the robustness of AI systems to diverse input prompts.

This incident serves as a reminder that the development of trustworthy AI requires careful consideration of the data used to train these models. By studying incidents like this one, we can work towards creating safer and more responsible AI systems. We invite you to share your thoughts on this matter and JOIN US  to learn more about efforts to improve AI governance.

Matched TAIM controls

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

Alleged deployer
OpenAI
Alleged developer
OpenAI
Alleged harmed parties
Muslims

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/118

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.