Google's Gemini AI Image Generator Is Now On Lockdown After This Embarrassing Mistake
Over the past couple of days, social media sites like Reddit and X, the platform formerly known as Twitter, have been inundated with unusual images generated by Google's Gemini AI. Even Elon Musk, developing his own self-described "anti-woke" and "maximum truth-seeking AI," took to mocking Google's generative AI chatbot troubles.
Google AI. Notice a difference? pic.twitter.com/cUhwrnKxpu
— DennisVinyard 🇦🇱🇺🇸 (@vinyard_dennis) February 21, 2024
In a turn from how other AIs have been criticized for biases toward white people in their generated images, Gemini did the opposite, creating images like dark-skinned Vikings and America's early leaders as Native American or Black when no race was specified. Former Google engineer Debarghya Das also highlighted a few cases where Gemini can be seen failing repeatedly along the same lines.
It's embarrassingly hard to get Google Gemini to acknowledge that white people exist pic.twitter.com/4lkhD7p5nR
— Deedy (@debarghya_das) February 20, 2024
Several X accounts that have opposed modern pushes toward diversity used Gemini's skin color flubs to push a narrative that the AI was racist against white people. Other posters alleged that Gemini outright blocked image generation for prompts where white people are specifically mentioned and asked the user to write a more diverse prompt.
Quite an embarrassment for a high-flying product
Google's official communications account posted on X that the company is aware of Gemini "missing the mark" with certain historical depictions. Earlier today, Google confirmed that it will pause Gemini's image generation capabilities until the issues are solved. "We will re-release an improved version soon," said the company.
We're already working to address recent issues with Gemini's image generation feature. While we do this, we're going to pause the image generation of people and will re-release an improved version soon. https://t.co/SLxYPGoqOZ
— Google Communications (@Google_Comms) February 22, 2024
Google hasn't explained exactly how or why Gemini's text-to-image function started making such errors. Still, the missteps are quite notable, considering Google's industry-leading work in the field of AI. It seems that Gemini may have been trained to avoid accusations of bias that have plagued other AIs, but something went wrong in its implementation.
The stakes for Gemini are particularly high because it's not just another ChatGPT wannabe: Google aims to push Gemini integration as much as possible, from phones to the entire Google Workspace suite of productivity tools.
The company recently launched a standalone Gemini app on Android and released it on iPhones via the iOS Google app. Gemini is not necessarily meant to replace Google Assistant, but it has already started offering integrations with some of its biggest products. As for the image generation functionality, Google hasn't offered a timeline for when exactly it will roll out the revisions.