Racial bias in artificial intelligence (AI) models continues to be a persistent problem, even as technology advances. A recent incident involving an Asian-American student illustrates this issue.
The student uploaded her picture to an AI portrait generator, requesting a more professional look. However, instead of enhancing her attire, the AI transformed her into a Caucasian woman with blue eyes.
was trying to get a linkedin profile photo with AI editing & this is what it gave me 🤨 pic.twitter.com/AZgWbhTs8Q— Rona Wang (@ronawang) July 14, 2023
Rona Wang, a 24-year-old Asian-American MIT graduate, recently decided to try out AI portrait generators. She gave Playground AI a try, asking it to create a more professional LinkedIn profile photo.
Unfortunately, the AI model surprised her with a significant alteration to her appearance. It gave her a fairer complexion, dark blonde hair, and blue eyes. Rona shared the result on Twitter, causing a stir among users.
As Twitter users debated the issue, it seemed that the algorithm misinterpreted the request for a “professional” look, changing Rona’s ethnicity from Asian to white. “I was like, ‘Wow, does this thing think I should become white to become more professional?'” Rona said in an interview with Boston.com.
The models aren't instructable like that so it'll pick any generic thing based on the prompt. Unfortunately, they're not smart enough.— Suhail (@Suhail) July 14, 2023
Happy to help you get a result but it takes a bit more effort than something like ChatGPT.
She initially found it amusing but realized it highlighted a recurring problem with AI tools: racial bias.
Rona openly discussed whether AI models are inherently racist. She found the transformation in the AI-generated image somewhat offensive. However, she also acknowledged the need to refrain from jumping to conclusions about AI’s racist intentions.
Suhail Doshi, the founder of Playground AI, responded directly to Rona’s tweet. He stated that the models were not designed to be instructable in that way and would generate results based on generic prompts. Unfortunately, the models lack the intelligence needed to fully grasp the nuances.
He expressed a willingness to assist Rona in achieving the desired result, but mentioned that the process required more effort than a simple task like using ChatGPT.
Doshi also emphasized that Playground AI was dissatisfied with the incident and expressed a desire to find a solution.
Responses to Rona’s tweet varied, with some accusing AI of embodying racist tendencies, while others attempted to explain the underlying issues behind the transformation.
If I understand correctly, the problem here is not in the biased datasets as most people could think. The even distribution amongst all types of faces wouldn't help, it'll reduce the overall model quality and will not fix the issue here – the result will still be random depending…— 0xprincess (@0x9212ce55) July 14, 2023
One user pointed out that the problem may lie in the general perception of AI being highly intelligent. They argued that AI often struggles to understand nuances and fully grasp the context of an image.
Another user challenged this perspective and highlighted the bias in the AI model’s transformation. They questioned why the model changed Rona’s eye color to blue, a trait possessed by only about 10% of the global population when her original photo featured brown eyes. While acknowledging that the model might be flawed, they questioned the absence of any bias.
I may not possess an extensive understanding of how AI models work, but it is evident that this incident is not an isolated occurrence of racial bias in AI. It will likely not be the last either. The critical question remains: How will the AI industry address this problem? It is an issue that cannot be ignored.