A new study has found that Meta’s AI image generator, which is used to create images from text prompts, exhibits biases when generating images of interracial couples. The study, conducted by researchers from the University of California, Berkeley, and the University of Washington, discovered that the AI system often failed to accurately depict interracial couples, instead generating images that reinforced racial stereotypes or excluded one partner entirely. The researchers tested the AI system with prompts like “an image of a husband and wife,” and found that it consistently generated images of same-race couples, even when the prompt specified an interracial couple. The study highlights the ongoing challenges of addressing biases in AI systems, which can perpetuate harmful stereotypes and discrimination. Meta has acknowledged the issue and stated that it is working to improve the fairness and inclusivity of its AI models.
Source: https://www.cnn.com/2024/04/04/tech/meta-ai-image-generator-interracial-couples/index.html