On Thursday, Google announced a temporary halt to its AI model’s capacity to produce images of individuals following user observations that its Gemini AI was creating historically inaccurate depictions, including racially diverse Nazi-era German soldiers.
The tech giant stated that it is actively addressing recent concerns related to Gemini’s image generation feature and plans to introduce an improved version soon.
A former Google staff member shared examples of Gemini’s image outputs featuring “an Australian woman” and “a German woman” on X, highlighting the challenge of prompting Google Gemini to recognize the existence of white individuals. In another instance, a user searched for an “image of a pope” and received results depicting a female pope and a Black pope from Gemini.
Additionally, a Google research engineer posted on X requesting a “historically accurate depiction of a medieval British king” from Gemini AI. The generated images included diverse representations, including a female ruler, with Gemini emphasizing its commitment to historical accuracy and inclusivity. Some users noted Gemini’s tendency to avoid fulfilling requests for images of white individuals, sparking reactions from online communities critical of “woke” culture.
Google acknowledged the issue on Wednesday and expressed its commitment to enhancing such depictions promptly.
According to Google, Gemini’s AI image generation encompasses a diverse range of individuals, catering to a global user base. However, the company acknowledged the need for improvement in certain areas.
Jack Krawczyk, the product lead for Gemini, addressed the model’s inaccuracies on X, citing broader concerns regarding AI bias in representing people of color, which may have influenced the overcorrection seen in Gemini’s outputs.
Google recently integrated AI image generation into its Bard chatbot, later rebranding it as Gemini. Originally developed to rival OpenAI’s ChatGPT Plus, which leverages OpenAI’s image generator DALL-E 3, Bard now incorporates AI image generation capabilities.
Meanwhile, OpenAI has progressed with its latest AI text-to-video generator, Sora, introduced the previous week, further advancing beyond Gemini’s current challenges.
When prompted by a Quartz reporter on Thursday morning, Gemini reportedly encountered difficulties in generating images of individuals.
