Google says AI engine 'missed the mark' after race controversy

Several users shared images online that were generated by the chatbot in response to prompts about the Founding Fathers and other historical figures.
One of the images that provoked some users
One of the images that provoked some usersTwitter/@mtracey

After facing widespread criticism over bias and inaccuracy issues, Google has made the decision to temporarily pause the image generation capabilities of Gemini - its new AI-powered chatbot tool unveiled in early February.

In a lengthy blog post acknowledging the missteps, Google's Senior Vice President Prabhakar Raghavan said that the image generation feature clearly "missed the mark" in some cases by producing offensive, misleading or factually incorrect images.

The controversy around Gemini erupted earlier this month when several users shared images online that were generated by the chatbot in response to prompts about the Founding Fathers and other historical figures.

Instead of accurately portraying the figures, the AI-generated images depicted people appearing to be Native American, Black, Asian and other non-white ethnicities. According to Raghavan, Google's efforts to tune the AI system to ensure diversity in image results inadvertently led it to "overcompensate in some cases."

He stressed that Google did not intend for the tool to refuse to generate images of people from any particular group.

While apologizing for the errors, Raghavan staunchly defended AI as an "emerging technology with huge potential" that Google is trying hard to "roll out safely and responsibly."

However, he also warned that hallucinations and inaccuracies are a "known challenge" with AI systems that require constant tweaking and improvements.

For the time being, Google has completely disabled Gemini's image generation capabilities for depicting people, while its teams work on fixing the underlying issues.

Raghavan said this will involve extensive testing before eventually relaunching the feature. However, he did not provide an expected timeline for when Gemini's image generation would be turned back on.

The controversy marked one of the first significant cases of AI bias as more and more generative chatbots reach mainstream consumers. It highlighted potential flaws and vulnerabilities in the rapidly-advancing technology currently powering the latest wave of AI applications across sectors.

Related Stories

No stories found.

X
The New Indian Express
www.newindianexpress.com