After trying artificial intelligence-powered avatar application Lensa, MIT Technology Review Senior Reporter Melissa Heikkila said it produced nonconsensual explicit photos of her. Heikkila reported Lensa generates avatars using open-source AI model Stable Diffusion that was built from LAION-5B, which is a data set containing scraped images from the internet. University of Washington assistant professor Aylin Caliskan said the data the AI model is trained on can be skewed toward sexualization because of the volume of such images on the internet.
13 Dec. 2022
Popular AI avatar creator can produce nonconsensual explicit images of users
Related stories
Notes from the IAPP Canada: Privacy watchdogs — What the 2024–25 annual reports reveal
A view from DC: Competing Republican visions for tech policy in the 118th Congress
A view from Brussels: Addressing data retention discrepancies
Notes from the Asia-Pacific region: Privacy matters to customers, individuals, society
European Commission receives final version of General-Purpose AI Code of Practice