Image: Maria Diaz/ZDNET.
With AI image generators, you can create any illustration, including historical scenes. But when users last week asked Gemini, Google’s model, to draw them certain historical figures, the AI’s response surprised them. As a reminder, this functionality is not yet available in France.
Indeed, when a person asked Gemini last week to generate an image with historical figures, or with people of different nationalities or ethnicities, the AI’s response did not correspond to the request at all. For example, the AI was no longer able to represent white people, even when the request involved a white person.
Witnesses to this problem have shared their images on social networks, notably showing people of color in response to requests for the Pope, the founding fathers of the United States or even a Viking. Debarghya Das, a former Google employee, shared a thread on these inaccuracies on X, compiling examples from his own experience and those of other users.
Google is working to improve its model
As a result, Google Communications acknowledged the issue via
Google then announced yesterday that it was temporarily suspending the generation of images depicting real people while the company works to resolve the issue. She specifies that an improved version should see the light of day soon.
If you try to ask Gemini to generate an image of a person today, you should get an error message that looks something like this: “I’m not able to do this at the moment.” And the AI won’t generate an image for you.
While Google’s AI-powered chatbot was launched last year, it only had image generation capabilities since the beginning of the month. With the web giant waiting so long to build this capability into its chatbot, the fact that it’s already making mistakes is surprising.
Google had already had a similar experience during the initial launch of Gemini, then called Google Bard: the chatbot had communicated inaccurate information about the James Webb space telescope during a demonstration.
Source: ZDNet.com