Meta's artificial intelligence can't handle different ethnicities

New artificial intelligence models are once again sparking controversy due to their inadequate handling of phenotypic traits and user descriptions in image creation. Now, Meta's artificial intelligence model, which has a tool for creating images from text, is under scrutiny.

Some early tests showed that Meta's AI image generator tool seems to have problems creating images of an Asian man with a Caucasian wife. After several tests performed by CNN and The Verge team, the tool wouldn't produce an image of an Asian person with someone who is not of the same racial phenotype.

The other way around doesn't seem to work either. AI tools can easily design any fantasy or science fiction image, yet they can't produce an image of an Asian woman with a White man. The tool also has problems with other interracial combinations. According to CNN, the AI tool couldn't come up with an image of a couple of an African-American woman with a White husband.

This AI tool works in mysterious ways. By adding religion, the tool seemed to better comply with CNN's instructions. They described that "when asked to create an image of a Black Jewish man and his Asian wife, it generated a photo of a Black man wearing a yarmulke and an Asian woman."

When asked to simply produce an image of an "interracial couple," the tool responded by saying it wasn't able to generate the image. However, couples of different phenotypes and backgrounds are common in the United States today. At least 19% of American opposite-sex couples are made up of individuals of different ethnicities.

The issues with Meta's AI image generator emerged a few weeks after the unsuccessful launch of Gemini, Alphabet's (Google's parent company) artificial intelligence model. Gemini was also accused of generating woke images, something that former company employees claim was not a mistake. Among its most catastrophic failures, Gemini failed to condemn pedophilia.