Google CEO says Gemini imaging errors ‘unacceptable’

Google CEO Sundar Pichai addressed the company’s recent problems with its AI-based image generation tool Gemini after it in historical images. He called the turn of events “unacceptable” and said the company was “working around the clock” on a fix, according to

“No AI is perfect, especially at this nascent stage of the industry’s development, but we know the bar is high for us and we’ll stick to it for as long as it takes,” Pichai wrote to employees. “And we’re going to review what happened and make sure we fix it at scale.”

Pichai remains optimistic about the future of the Gemini chatbot, previously called Bard, noting that the team has already “seen significant improvement across a wide range of prompts.” The imaging aspect of Gemini until the correction is fully developed.

It started when Gemini users began to notice that the generator was starting to display historically inaccurate images, such as pictures of Nazis and America’s founding fathers as people of color. It quickly became a big thing on social media, with the word “woke” being thrown around a lot.

Prabhakar Raghavan, Google’s senior vice president of knowledge and information, didn’t blame vigilance, Basically, the model was fine-tuned to allow for different groups of people in photos, but “failed to account for cases that clearly shouldn’t show a range.” This led to controversial depictions such as people of color appearing as Vikings and Native American Catholic Popes.

Raghavan also said the model has become more cautious over time, occasionally refusing to respond to certain prompts after mistaking them as sensitive. This explains the reports that the model refused to generate images of white people.

It seems the company was trying to both please a global audience and ensure the model didn’t fall into some of the pitfalls of competing products, such as creating sexually explicit images or images of real people. Tuning these AI models is an extremely delicate job and the software can do it easily. That’s what they do. In any case, I would prefer a historically inaccurate Catholic Pope any day of the week. Chalk this up as another reminder that AI still has a long way to go.

As for Gemini, the company promises that the image generator will return in the near future, but it still requires a set of fixes and tests to make sure it never happens again, including “structural changes, updated product guidance, improved launch processes, robust evaluations and red teaming, and technical recommendations.”