Gemini AI Accused of Racism Against White People

Google’s latest AI image generation capabilities on Gemini have faced criticism from X (formerly Twitter) users recently. The tool, designed to produce images based on text prompts, has been criticised for its tendency to prioritize 'wokeness,' often generating images featuring individuals of various ethnicities even in scenarios where historical accuracy is compromised.

These 'glitches' have sparked controversy on X, with some individuals, particularly those on the right, accusing Google of perpetuating racism against white people. For example, when an X user asked Gemini to depict America’s founding fathers, the AI included women and people of color, presumably to enhance diversity representation. However, in instances where historical accuracy is essential, this "inclusive" approach results in inaccuracies and discomfort.

"We are aware that Gemini is producing inaccuracies in certain historical image generations," stated a Google announcement on X. "We are actively working to rectify these inaccuracies. While Gemini’s AI image generation encompasses a diverse range of individuals, which is generally positive given its global user base, it falls short in this context."

This isn't the first time an AI has struggled with diversity balance. Years ago, Google faced backlash and issued an apology after its photo app notoriously mislabeled a picture of a Black couple as "gorillas," as reported by the BBC. More recently, OpenAI's Dall-E image generator consistently depicted CEOs and other authority figures as white males unless gender or race were explicitly specified by users.

However, overcorrecting for bias by indiscriminately injecting diversity into every image risks perpetuating harmful stereotypes and technical challenges. Gemini's current issue underscores the complexities associated with implementing simplistic inclusion filters.

For any enquiries or information, contact ask@tlr.ae or call us on +971 52 644 3004. Follow The Law Reporters on WhatsApp Channels.