Meta AI is facing criticism for its image generator’s inability to accurately depict diverse prompts. Reports from *The Verge* and Engadget reveal that prompts such as “Asian man with white wife” result in images of same-race couples, indicating a bias towards depicting people of the same race. Even prompts for a diverse group of people generated mostly white faces with occasional token representation of diversity.
Noteworthy signs of bias include making Asian men look older and Asian women appear younger, as well as adding culturally specific attire without prompt. This issue with diversity portrayal is not unique to Meta AI, as Google’s Gemini image generator similarly faced scrutiny for overcorrecting diversity to strange results when depicting historical figures.
Meta AI, currently in a “beta” phase, has not provided immediate feedback on these concerns. The company’s struggles extend beyond image generation, as it has also faced challenges in accurately responding to simple questions about current events and public figures. The root cause of Meta AI’s bias issues remains unclear, highlighting the ongoing challenges in developing unbiased generative AI platforms.
HEALTH
WORLD AFFAIRS
TECHNOLOGY
CREATIVE
