Someone's race is a sensitive subject, and Meta's AI image generator is under fire because of it.
Meta has its own generative AI to generate images, and that the only limit it users' imagination. The thing is, it has at least one apparent weakness: it struggles to properly generate images of couples, or friends from different racial backgrounds.
For example. asking the AI to create a picture of an Asian man with a White wife, the tool will come up with a slew of images of an East Asian man with an East Asian woman.
The same thing happened when prompted to create an image of an Asian woman with a White husband.
When asked to create an image of a Black woman with a White husband, the AI is only able to generate images featuring Black couples.
And when the tool also generated images of two Asian women when asked to create an image of an Asian woman and her Caucasian friend.
And when requested for an image of a Black woman and her Asian friend, the AI came up with photos of two Black women.
At certain times, the AI failed to work properly.
"This image can’t be generated. Please try something else," the AI responded.
It's worth noting that the repeating the request over and over again, did eventually have the AI to create proper images.
Long story short, the AI from Meta sparked controversy because it couldn't properly represent different races, but only when they're together.
When the AI is asked to create one person, no issue is found.
But when it's asked to create two or more people in different races, the request somehow confuses the AI.
The news about this first surfaced when The Verge reported the issue with race, highlighting how the AI tool from Meta "can’t imagine" an Asian man with a White woman.
Meta said that its team are working to fix these problems, and try to make the tool less biased.
But it's worth noting that the issue is not limited to just Meta.
Other AI tools from other companies have had issues with race in the past.
This is because an AI is only as good as the datasets it used for training.
Since tech companies get many of their AI's training materials by literally scraping the web and social media for photos and videos, their AIs inherited many of the web's biased information, including those they gather from White dominant-communities.
This shows that making AI fair and accurate is hard work.