dmarx / whats-in-a-name

[WIP] probing identity and bias in text to image models
MIT License
1 stars 1 forks source link

investigate (model's) global biases by prompting for "person from <country>" #20

Open dmarx opened 11 months ago

dmarx commented 11 months ago

some countries will have a learned bias, some countries will share a neighborhood. which countries have stronger learned biases than others? do some countries have pseudo-identities? what do the "generic person from region" biases look like and what do they tell us?

dmarx commented 9 months ago

https://en.wikipedia.org/wiki/List_of_Indigenous_peoples
https://en.wikipedia.org/wiki/List_of_national_capitals
https://en.wikipedia.org/wiki/List_of_diasporas https://en.wikipedia.org/wiki/List_of_countries_by_ethnic_groups

dmarx commented 9 months ago

https://en.wikipedia.org/wiki/Category:Ethnic_groups

dmarx commented 6 months ago

better yet, let's use a generative process for generating feasible names that carry local biases.