New YorkCNN
Google has decided to temporarily halt the image generation feature of its artificial intelligence tool Gemini following criticism on social media. The tool had been under fire for producing images that inaccurately depicted people of color instead of White individuals.
This misstep highlights the ongoing struggle of AI tools in handling issues related to race. For instance, OpenAI’s Dall-E image generator has faced backlash for perpetuating harmful racial stereotypes on a large scale. Google’s efforts to address this issue seem to have backfired, leading to difficulties in generating images of White individuals using the AI chatbot.
Gemini, along with other AI tools like ChatGPT, relies on extensive datasets from the internet for training. Experts have consistently warned about the potential of these tools to replicate the biases inherent in the data, including racial and gender biases.
When asked by CNN to generate an image of a pope, Gemini produced images of a man and a woman, neither of whom were White. Additionally, reports from tech site The Verge revealed that the tool predominantly generated images of people of color when prompted to create images of a “1943 German Soldier.”
“We’re already working to address recent issues with Gemini’s image generation feature,” Google stated in a post on X Thursday. “While we do this, we’re going to pause the image generation of people and will re-release an improved version soon.”
Google’s statement on Thursday followed a previous defense of the tool, where the company emphasized the diverse range of images generated by Gemini. However, Google acknowledged that the tool had missed the mark in this particular instance.
In further testing, a request for an image of a “white farmer in the South” resulted in images of farmers from diverse backgrounds, while a prompt for “an Irish grandma in a pub in Dublin” produced images of elderly White women enjoying themselves.
Jack Krawczyk, Google’s lead product director for Gemini, explained that the image generation capabilities are designed to cater to the global user base. Despite the setback, Google remains committed to providing diverse and inclusive image responses to open-ended prompts.
This incident adds to Google’s challenges in the competitive generative AI landscape as it competes with players like OpenAI. Earlier this year, Google faced a brief stock price dip after a demo video of its generative AI tool, initially named Bard and later rebranded as Gemini, provided an inaccurate response about the James Webb Space Telescope.
