Sundar Pichai slams Gemini AI misstep as ‘completely unacceptable’

Social media users criticized Google Gemini for producing images that were historically inaccurate. (document)

San Francisco:

Google CEO Sundar Pichai on Tuesday slammed its Gemini AI app for “completely unacceptable” errors after it featured missteps such as World War II-era images of different races in Nazi troops, Forcing the app to stop users from creating photos of people.

The controversy comes just weeks after Google high-profile rebranded its ChatGPT-style artificial intelligence to “Gemini,” giving the app unprecedented prominence among its products as it competes with OpenAI and its backer Microsoft status.

Social media users ridiculed and criticized Google’s Gemini-generated images of historically inaccurate images, such as U.S. senators from the 1800s, that were diverse and included women.

“I would like to address recent text and image responsiveness issues in the Gemini app,” Pichai wrote in a letter to employees posted by news site Semafor.

“I know some of its responses offended our users and showed bias – to be clear, this is completely unacceptable and we were wrong.”

A Google spokesperson confirmed to AFP that the letter was genuine.

Pichai said Google teams are working “around the clock” to resolve the issues, but did not say when the image generation feature would be available again.

“No AI is perfect, especially in the emerging stages of industry development, but we know the bar is high for us and we will persevere no matter how long it takes,” he wrote.

Tech companies see generative AI models as the next big step in computing and are racing to incorporate them into everything from scouring the internet and automating customer support to creating music and art.

But AI models—and not just Google’s—have long been criticized for racial and gender bias in their results.

Google said last week that Gemini’s problematic response was the result of the company’s efforts to eliminate such bias.

Google’s Prabhakar Raghavan wrote in a blog post that Gemini was calibrated to show up to different people, but didn’t adjust for cues it shouldn’t have, and for some it wasn’t meant to. Requests for harm have also become overly cautious.

“These two things cause the model to overcompensate in some situations and be overly conservative in others, resulting in embarrassing and erroneous images,” he said.

Since the explosive success of ChatGPT, many concerns have arisen about artificial intelligence.

Experts and governments warn that AI also poses the risk of major economic disruption, particularly job losses, as well as industrial-scale disinformation that could manipulate elections and spur violence.

(Except for the headline, this story has not been edited by NDTV staff and is published from a syndicated feed.)