Emotion Detection Models Showed Disparate Performance along Racial Lines

November 9, 2018

A study reveals racial disparities in the performance of emotion detection models from Face++ and Microsoft's Face API, with these tools more frequently classifying smiling or ambiguous facial photos as negative emotions for Black faces compared to white faces.

For those interested in shaping the future of trustworthy AI, join HISPI Project Cerebellum TAIM and help govern, map, measure, and manage such incidents to prevent harm and establish guardrails for AI.

Matched TAIM controls

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

MEASURE 2.6 — similarity 0.646, rank 1. TAIM detail and related incidents →
MEASURE 4.3 — similarity 0.640, rank 2. TAIM detail and related incidents →
MEASURE 2.5 — similarity 0.639, rank 3. TAIM detail and related incidents →

Alleged deployer: megvii, microsoft
Alleged developer: megvii, microsoft
Alleged harmed parties: black-people

Source

Data from the AI Incident Database (AIID). Cite this incident: https://incidentdatabase.ai/cite/454

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.