Common Biases of Vector Embeddings

July 21, 2016

The findings of this study serve as a stark reminder that even the most fundamental building blocks of NLP, such as vector embeddings, can perpetuate existing social inequalities if not carefully designed and tested for bias. By shedding light on these common biases, researchers from Boston University and Microsoft Research, New England have taken an important step towards creating more inclusive and equitable AI systems, helping us to build a future where technology serves the needs of all people JOIN US.

Matched TAIM controls

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

Alleged deployer
Microsoft Research, Boston University
Alleged developer
Microsoft Research, Google, Boston University
Alleged harmed parties
Women and girls, Women, 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/12

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.