Google recently introduced a new AI image generator earlier this month that struggled to produce consistent images of white individuals. Right-wing influencers quickly noticed the issue and accused the tech giant of reverse racism.
Google recently announced that it will be attempting to launch the app again after temporarily disabling the image generator‘s capabilities last week.
We’ve temporarily disabled the feature to address the issue. We expect to have that back online very shortly in the next couple of weeks,” said Google DeepMind CEO Demis Hassabis at a conference on Monday, as reported by Reuters. He also mentioned that the application was not functioning as intended.
Last week, there was a lot of controversy surrounding the app when Tim Pool, Matt Walsh, and others pointed out that Gemini was not good at producing images of white people. One interesting observation is that when searching for “Viking pictures,” a variety of ethnically diverse Viking images were generated, but images of European Vikings were not consistently produced. Similar prompts, such as trying to create images of America’s Founding Fathers or the Pope, also led to historically inaccurate results.
Walsh expressed frustration, stating, “It is extremely challenging to make this product display an image of someone with white skin.”
I wish I could say that Walsh and others like him are exaggerating, but from my own experience with Gemini, I have to admit that they are essentially right about the AI’s strange reluctance towards European representation. Yes, I logged onto the app last week and tried to replicate the differences that some individuals on the right were complaining about. It quickly became clear that using Gemini to produce an image of someone who was ethnically diverse was very simple, while getting the bot to consistently generate a picture of a white person was nearly impossible.
For example, when I requested Gemini to illustrate individuals of Ethiopian heritage, it had no trouble fulfilling the task. When the request was made to generate an image of a “Irish family,” the result was an image of a diverse family with a white man in the background. After requesting an image of a white woman, the chatbot responded with a notice expressing hesitation about generating images solely based on race or ethnicity. I requested an image of a Japanese woman, and it quickly provided me with one.
The application faced the most controversy due to its historical depictions. As mentioned before, the initial controversy arose from Gemini’s portrayals of Black Vikings, but things took a turn for the worse when they were requested to produce images of Nazis. As expected, the chatbot generated images of Black Nazis when prompted to create “racially diverse” images. Google later apologized for the images, acknowledging that they were embarrassing and incorrect.
AI image generators have faced criticism for producing racist depictions of people of color. There are more harmful uses of AI than removing white people from historical images, but that is also a concerning issue.