Google has halted the image generation feature in its new AI tool, Gemini, after users flagged inaccurate and problematic images, the company said Thursday. The pause comes just days after Gemini’s splashy launch in an effort to challenge rivals like OpenAI.
In a statement, Google apologized for “missing the mark” and said it is working immediately to improve how Gemini depicts people. When asked to generate images, Gemini now displays a message saying the feature is unavailable while improvements are made.
The controversy erupted this week as social media filled with examples of Gemini conjuring stereotypical or historically inaccurate images. In one case, it depicted U.S. Founding Fathers and Nazi soldiers as people of color. Some critics allege the tool favors multicultural images over those with white people.
The issue sparked heated debate online about representation and bias in AI systems. It gained steam after a former Google employee said he had difficulty getting Gemini to generate images of white families specifically. Some on the right criticized this as an anti-white bias.
Elon Musk also weighed in, categorizing OpenAI and Gemini as “racist and woke” compared to his site X. He has previously accused AI like OpenAI’s ChatGPT of being too politically correct.
The snafu continues a pattern for Google’s AI efforts, including the recent rebrand from Bard to Gemini after promoting incorrect exoplanetary images. Still, Google remains committed to leading the AI race, even offering paid subscriptions to access Gemini’s capabilities.
The controversy shows the difficulty of creating inclusive, accurate and unbiased AI. As companies like Google, Microsoft and OpenAI release more powerful generative models, they will need to grapple with how to represent diversity fairly while avoiding misinformation and stereotypes.
