Google CEO Calls Unacceptable Gemini AI Image Bias
Google CEO Sundar Pichai labels the recent biased image generation of the Gemini AI tool as completely unacceptable. The company pauses the feature and promises structural changes before relaunching it in the coming weeks.
Google CEO Sundar Pichai tells employees that the recent performance of the company's Gemini AI tool is completely unacceptable. The backlash starts after users share viral examples of the image generator producing historically inaccurate depictions, such as America's Founding Fathers as black, a female Pope, and a dark-skinned Nazi soldier. Furthermore, the tool routinely refuses to create images of white people, leading to widespread online accusations of anti-white bias.
In response to the controversy, Google pauses the image generation feature of Gemini while it works on a fix. Pichai states in an internal memo that there is no excuse for the problematic and offensive results, noting that the tool shows clear bias. He promises a clear set of actions to address the failure, including structural changes, updated product guidelines, improved launch processes, and better technical evaluations.
Google executive Prabhakar Raghavan explains in a blog post that the issue stems from a fine-tuning error designed to prevent violent or explicit images. Developers intentionally program the system to generate diverse images for general requests so the tool works well for everyone globally. However, this safety measure backfires because the AI fails to recognize historical contexts where a diverse range of people is factually incorrect or inappropriate.