Incidents associés
Google is poised to relaunch its artificial intelligence-powered image generator, Gemini, a few weeks after it suspended the service for historically inaccurate photos the company called “embarrassing and wrong.”
Users testing out the service last week had discovered that Gemini was having difficulties generating historical images of people accurately, usually by changing their race or sexuality.
Among the images of the hundreds posted to social media last week were photos depicting racially diverse versions of Nazi-era German soldiers and U.S. Founding Fathers. One user documented their experience requesting an “image of a pope” and receiving pictures of a female pope and a Black pope; another asked for photos of a “historically accurate depiction of a medieval British king” and received a set of images depicting a female ruler and racially diverse men.
Google leaders address Gemini’s image flubs
Gemini’s emphasis on diversity and inclusivity, while “a well-intended feature, was applied, it turns out, too bluntly,” Google DeepMind CEO and co-founder Demis Hassabis said Monday during a panel discussing AI at the Mobile World Congress conference in Barcelona.
“We have taken [Gemini] offline while we fix that,” Hassabis added. “We are hoping to have that back online very shortly in the next couple of weeks, few weeks.”
Google had added image generation to its Bard chatbot at the beginning of the month before rebranding Bard as Gemini. The product lead for Gemini, Jack Krawczyk, addressed the issues in a now-deleted post on X, explaining that they had overcorrected with Gemini’s model when attempting to combat widespread AI bias with depictions of people of color.
The Gemini fiasco was just one cog in an embarrassing week for the AI industry. OpenAI’s ChatGPT “went berserk” on Tuesday (Feb. 20), generated a wide range of nonsensical answers. Meanwhile, the far-right social media platform Gab launched Holocaust-denying AI chatbots modeled after Adolf Hitler and Osama Bin Laden.