Google CEO Sundar Pichai has said that the company got it wrong when it generated racially biassed image results in response to user queries earlier this year. He also said that the image generation capabilities in Google Gemini will be re-released in a few weeks.
“We got it wrong,” Pichai said in an interview with Bloomberg, indicating that the incident was an example where good intentions backfired.
Google quickly paused Gemini’s image-generating feature after social media users complained it was creating ‘wrong’ images. In the image that caused much uproar, the chatbot included people of colour while depicting Nazis and the US Founding Fathers.
Pichai, who dubbed AI tool’s controversial responses ‘completely unacceptable’, also ordered a complete rebuild.
“From the ground up we are retraining these models, just to make sure we are also making the product better. As soon as it’s ready, we will get it out to people,” he said.
What Pichai previously said on biassed image generation
Previously, Pichai said that Gemini’s answers align with the answers that users find on Google Search.
“The work we have done over many many years is making sure, from a Search standpoint, we try to reflect what’s out on the web and we want to give trustworthy, high-quality information,” Pichai said.
“What I’m saying it’s not about [just] giving an answer. There are certain times you give an answer [for questions like] ‘What’s the population of the United States?’ Yes, it’s an answerable question. [But] there are times you want to surface a breadth of opinions out there on the web, which is what search does and does it well,” the top executive added.
“Just because you’re saying we are summarising it on top doesn’t mean we are from those principles, the summary can still point you to the range of opinions out there,” Pichai explained.
The company previously acknowledged the issue, saying that this happened due to “limitations in the training data used to develop Gemini.” Even Google co-founder Sergey Brin also said at San Francisco’s AGI House that the AI chatbot “definitely messed up”.
Originally Appeared Here
