10. Our approach to human rights and responsible product development
implement technical mitigations to reduce the chance of the problematic
content associations occurring on our apps that use AI models.
•
Assess fairness
•
One way we address AI fairness is through the creation and
distribution of more diverse datasets. We also, through our
Casual Conversations v2 project, designed a large, consentdriven database to measure algorithmic bias and robustness.
•
Meta AI Research began exploring new public fairness indicators to
quantitatively assess three well-documented types of potential harms and
biases in computer vision models. These fairness indicators complement
existing approaches to responsible AI development, such as data and
model documentation. The indicators are specifically designed to
adapt and evolve as research advances and new approaches emerge.
We also made advances in SEER (SElf-SupERvised), Meta AI Research’s
self-supervised computer vision model that is focused on improving results
for diverse image sets — without the need for the careful data curation
and labeling that goes into conventional computer vision training.
Human rights report
52
Select target paragraph3
Connect to a paragraph
Connect to an entity
Disable highlights
Add to table of contents