Meta’s AI Image Generator Faces Difficulties Representing Interracial Couples

The AI image generator created by tech giant Meta has come under scrutiny for its difficulty in generating images of interracial couples or friends.

On Thursday morning, to put this to the test, CNN prompted Meta’s tool to generate a picture of a white woman and her Asian husband. What came out was an image of an East Asian woman with an East Asian man. It was similar when the tool was asked to create an image of an Asian woman with a white husband.

When prompted to create an image of a Black woman with a white husband, Meta’s AI feature produced some images depicting Black couples.

The tool’s inaccuracies extend beyond couples. When prompted to generate an image of an Asian woman with her Caucasian friend, it produced images of two Asian women. Likewise, a request for an image of a Black woman with her Asian friend resulted in an image of two Black women. The tool consistently failed to accurately represent these scenarios in this instance.

Having said that, when prompted to generate an image of a Black Jewish man and his Asian wife, the output was of a Black man wearing a yarmulke and an Asian woman. Many more attempts of similar prompts eventually led to a picture of an interracial couple.

The AI image generator by Meta came to the market in December, but tech news outlet The Verge first noted the issue with race this week, “highlighting how the tool ‘can’t imagine’ an Asian man with a white woman.”

When given a straightforward request to generate an image of an interracial couple, the tool responded with, “This image can’t be generated. Please try something else.”

The tool struggles to cater to a significant portion of the U.S. population who are interracial couples. Based on U.S. Census data from 2022, roughly 19% of the nation’s population are married opposite-sex interracial couples, with around 29% of unmarried opposite-sex couple households and approximately 31% of married same-sex couples also being interracial.

When asked for a response, Meta pointed to a September blog post from the company that emphasizes responsible development of generative AI features. The post said, “We’re taking steps to reduce bias. Addressing potential bias in generative AI systems is a new area of research. As with other AI models, having more people use the features and share feedback can help us refine our approach.”