Seeming Pattern of Gemini Bias and Sociotechnical Training Failures Harm Google's Reputation

In February 2024, Google's Gemini AI chatbot (formerly Bard) faced significant backlash after users discovered its image generation feature was producing historically inaccurate depictions. The AI system generated images showing racially diverse people in contexts where they would be historically inaccurate, such as depicting Nazi-era German soldiers as people of color, showing Black Vikings, female popes, and diverse representations of the US Founding Fathers. Users reported that the system appeared to avoid generating images of white people even when historically appropriate. On February 22, 2024, Google acknowledged the issues and temporarily paused Gemini's ability to generate images of people. Google CEO Sundar Pichai called the responses 'completely unacceptable' in an internal memo. The controversy appeared to stem from Google's attempts to address AI bias by promoting diversity in image generation, but the system overcorrected and applied diversity requirements even to historical contexts where accuracy was important. The incident attracted widespread criticism on social media, with some critics calling the AI 'woke' and historically revisionist. Google promised to release an improved version after extensive testing and fixes.