Topline
Google has announced a temporary halt to its Gemini image generator’s capacity to produce images of individuals following backlash over the generation of misleading images depicting people’s races in historical settings. This action was prompted by criticism from billionaire Elon Musk, who referred to the service as “woke.”
Key Facts
On Wednesday, Google acknowledged the need for immediate enhancements in the portrayal of such depictions and swiftly disabled the feature less than 24 hours later. The company assured users of an upcoming “enhanced version” of the service without specifying a timeline for the reinstatement of people image generation. Jack Krawczyk, the product director for Gemini Experiences at Google, confirmed the inaccuracies in historical image generation and assured ongoing efforts to refine the service for accurate historical representation. Despite the pause on generating images of individuals, Forbes can still utilize Gemini for creating images of inanimate objects, with a promise of the feature’s return in future updates.
Key Background
Gemini, Google’s chatbot formerly known as Bard, was launched on February 8, introducing an image generator feature akin to other AI programs like OpenAI’s ChatGPT and DALL-E. However, issues arose post-launch as users noted historical inaccuracies in generated images of figures and scenes. Google acknowledged these discrepancies and expressed commitment to rectifying the situation to align with historical accuracy.
Chief Critic
Elon Musk, the billionaire founder of xAI, a competitor company, criticized Gemini’s image generation service, labeling it as “woke” and “racist.” In contrast, xAI positions itself as a truth-seeking entity and has not ventured into offering image generation services yet.
Tangent
While Google’s Gemini addressed historical inaccuracies, other AI image generators like Stability AI’s Stable Diffusion XL have faced criticism for producing racially biased images favoring white individuals. Google emphasized its commitment to global representation and bias mitigation in image generation, aiming to cater to a diverse user base and address such biases effectively.
