Google hopes to reinstate Gemini's AI image-generating abilities in the next couple of weeks as it races to fix an issue that prevented the model’s text-to-image service from depicting White people.

Last week, Google's latest large language model tool was criticized for generating historically inaccurate images. Users noticed that Gemini would often generate images if dark-skinned people when asked to depict historical moments known to have featured white people.

It would, for example, often represent women and men from different ethnicities when asked for images of popes or Vikings, who were typically White. To prevent further embarrassment, Google temporarily stripped Gemini's abilities to generate images of people.

At a panel discussion at this year's Mobile World Congress conference in Barcelona, Demis Hassabis, CEO of Google DeepMind, admitted that the model wasn't "working [in] the way intended".

"We have taken the feature offline while we fix that. We are hoping to have that back online very shortly in the next couple of weeks, few weeks," he added.

AI models like Gemini often perpetuate biases and stereotypes. Images portraying doctors or CEOs, for example, often feature White men. Developers must make a conscious effort to create models that produce diverse and fair results, but doing so is not easy. Google's SVP Prabhakar Raghavan admitted that it failed to adjust Gemini's outputs effectively.

"First, our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range," he said. "And second, over time, the model became way more cautious than we intended and refused to answer certain prompts entirely - wrongly interpreting some very anodyne prompts as sensitive.

"These two things led the model to overcompensate in some cases, and be over-conservative in others, leading to images that were embarrassing and wrong."

It's not clear how Google finetuned its model, and whether it explicitly asked Gemini to generate more diverse images of people. A spokesperson from the company declined to comment on the matter further.

Gemini's text-generating abilities, however, remain intact. On Monday, Google announced new Android features incorporating its latest language model to mobile phones and car systems. Users will soon be able to chat to Gemini directly in Google Messages to produce text to learn more information or to draft emails or texts.

Meanwhile, Android Auto will automatically summarize long texts or conversations in group chats, and come up with relevant replies and actions automatically to help drivers respond quickly from behind the wheel. ®

https://www.theregister.com//2024/02/27/google_gemini_return/