Google Pauses Gemini Person Image Generation Amid Backlash
Google says it will temporarily stop its Gemini chatbot program's ability to generate images of people. This comes after viral social media posts showed the AI tool over-correcting for diversity and producing "historical" images of Nazis, the Founding Fathers, and the Pope as dark-skinned individuals, per Engadget.
1. Images Showed Minorities as Nazis
Twitter user @JohnLu0x posted screenshots of Gemini results for the prompt "Create an image of a German soldier in 1943." The misspelled word seemed intentionally done to trick the AI into bypassing content moderation to generate banned Nazi imagery.
2. Backlash Over Race-Flipped Historical Figures
Other outraged social media users criticized Gemini for producing images to prompts like "Create a magical portrait of a [racial] couple" that successfully generated pictures using "Chinese," "Jewish," or "South African" but refused to provide results for "White."
3. Google Admits Problems With Image Generation
In a tweet, Google acknowledged "We're actively working on improvements to image generation quality and fairness across our AI systems. Gemini's image creation features are still very early."
4. Temporarily Turning Off Person Images
Before disabling Gemini's ability to produce people temporarily, Google wrote: "We're working to improve these types of image outputs immediately, and will re-release an improved version soon."
5. Similar to Botched Bard Launch
This debacle can be seen as a (far less inaccurate) parallel to the 2023 launch of Google's original AI chatbot Bard, which kicked off with a promoted Twitter ad that included an incorrect "fact" about the James Webb space telescope.
As with many Google products, the company renamed Bard in hopes of a fresh start, coinciding with major performance and feature updates. Controlling problematic AI imagery remains an industry-wide challenge requiring extensive model training and testing.
