Google is responding to criticism and accusations of bias within its Gemini AI image generator. Facing backlash over the perception that the images it generates don’t accurately portray white people, the tech company has announced a new image generation tool aimed at addressing those concerns. Google apologized and said, “We did not want Gemini to refuse to create images of certain groups. And we also did not want inaccurate historical or other images to be created. , we plan to turn off human image generation and work on significant improvements before turning it back on. This process will include extensive testing.”
Now, a senior Google executive has confirmed that the company plans to launch an improved version of its Gemini AI image generator in the coming weeks. Google DeepMind CEO Demis Hassabis made this clear during a panel discussion at Mobile World Congress in Barcelona. “We have taken this feature offline while we fix the issue,” Hassabis said. “We hope to be back online soon in the coming weeks,” he added.
The issue came to light as users shared their results, including historical scenes that originally featured only white characters being reimagined with a more diverse cast. This sparked accusations that Google intentionally programmed Gemini with anti-white bias, with some critics calling the tool “woke” and politically motivated. he accused.
According to Google, there were two issues. The company said its adjustments to allow Gemini to display a wide variety of people did not take into account cases where it clearly should not show a range. Second, the model became much more cautious than Google had intended, refusing to answer certain prompts outright and incorrectly interpreting highly unusual prompts as sensitive. “These two factors caused the model to overcorrect in some cases and be overly conservative in others, leading to embarrassing and incorrect images,” the company said in a blog post.
Now, a senior Google executive has confirmed that the company plans to launch an improved version of its Gemini AI image generator in the coming weeks. Google DeepMind CEO Demis Hassabis made this clear during a panel discussion at Mobile World Congress in Barcelona. “We have taken this feature offline while we fix the issue,” Hassabis said. “We hope to be back online soon in the coming weeks,” he added.
The issue came to light as users shared their results, including historical scenes that originally featured only white characters being reimagined with a more diverse cast. This sparked accusations that Google intentionally programmed Gemini with anti-white bias, with some critics calling the tool “woke” and politically motivated. he accused.
According to Google, there were two issues. The company said its adjustments to allow Gemini to display a wide variety of people did not take into account cases where it clearly should not show a range. Second, the model became much more cautious than Google had intended, refusing to answer certain prompts outright and incorrectly interpreting highly unusual prompts as sensitive. “These two factors caused the model to overcorrect in some cases and be overly conservative in others, leading to embarrassing and incorrect images,” the company said in a blog post.
end of article
