Google search boss says employees worked 120 hours a week on Gemini's image tool, fixed most issues in 10 days (2024)

New Delhi,UPDATED: Apr 24, 2024 16:19 IST

In February this year, Google's AI tool Gemini drew flak for refusing to generate images of white people. The AI tool was accused of being "too woke" by many social media users after it portrayed important historical figures as people of colour even though they were originally white. Google then paused Gemini's ability to generate images of human beings. Now, we all know how Google was criticised openly and how some experts and netizens reacted to the incident. However, we weren't aware of what went down behind Google's office doors. And as per a CNBC report, Google employees had to work tirelessly to improve Gemini's image generator.

Google search boss on employees working 120 hours a week

Prabhakar Raghavan, Google's search boss, addressed an all-hands meeting last month. CNBC accessed an audio of the event and reported that Raghavan talked about the direction that the company is going to take moving forward and how it is investing in machines in the age of generative AI. During the meeting, Raghavan also praised Google employees for putting in extra hours to work on Gemini's image generation tool.

As per CNBC, Raghavan was all praises for employees working on Gemini. He added how their commitment had intensified, pushing from a standard 100-hour workweek to an extraordinary 120 hours in order to correct Google's image recognition tool. This helped the team fix most issues (about 80 percent) in 10 days, Raghavan revealed.

It is to be noted that Google's abilities to generate images of people ability is still on hold. The company is yet to make the feature available.

Returning to Raghavan's statement, he added that Google's image generation mishap was not a result of insufficient effort. "I want to be clear, this wasn’t some case of somebody slacking off and dropping the ball," he was quoted by CNBC.

This wasn't the first time that Google relied on its employees to fix an AI tool's mistakes. Last year, when Bard was new and was criticised for being inaccurate, Google had asked its employees for help in improving its responses.

What happened with Google Gemini?

Gemini faced criticism for generating historically inaccurate and biased images. The AI system was accused of being "too woke" and getting details about "white people" wrong. Users reported that when they requested images of figures like the pope, English kings, Vikings, or even Nazi soldiers, the AI produced pictures of individuals with darker skin tones. In the midst of ongoing discussions about the representation of race in history, Google acknowledged the problem and pledged to improve the accuracy of its generative AI tools.

Jack Krawczyk, the Senior Director of Product Management for Gemini Experiences, had also recognised the concerns. He assured that Google takes representation and bias seriously, aligning with their AI principles, which focus on reflecting the diverse global user base. Krawczyk explained that while the image generation capabilities were designed with these principles in mind, the AI needed fine-tuning, especially for prompts related to historical figures.

Krawczyk had said, "We are aware that Gemini is offering inaccuracies in some historical image generation depictions, and we are working to fix this immediately. Historical contexts have more nuance to them, and we will further tune to accommodate that."