Google claims to be significantly improving its new AI-powered image creation tool and is striving to refine it in order to reduce the possibility of bias.
Users complained that the company’s Gemini bot displayed pictures of people of different genders and nationalities, which has previously been incorrect.
For instance, when looking for pictures of America’s founders, women and people from many ethnic origins turned up.
According to the manufacturer, their tool had “missed the mark.”
Senior Director of Gemini Experience Jack Crawzic stated, “A wide spectrum of persons are created by Gemini’s AI image production. Additionally, the fact that it is used by people worldwide is usually a positive thing. However, its footprint is not here.”
“We’re working to immediately improve these types of depictions,” he stated.
For instance, Google encountered criticism roughly ten years ago when it called a picture of a black couple in its photo app “gorillas.”
When users saw that OpenAI, a rival AI company, primarily produced photos of white men, they too came under fire for feeding damaging prejudices.
Google unveiled the most recent version of Gemini last week in response to pressure to demonstrate that it is not falling behind in the development of AI.
In response to written inquiries, the bot generates images.
It attracted detractors right away, who charged that the business had trained the bot to reply humorously in order to soften its position.
In response to the outcomes of a Viking image request, humorist and writer Frank J. Fleming, who publishes on publications such as the right-wing PJ Media, exclaimed, “Come on.”
In America, where a number of significant digital platforms are already under fire for allegedly liberal prejudice, accusations of bias have gathered momentum.
According to Mr. Krawczyk, the business values fairness and representation and wants its output to accurately represent its diverse user base worldwide.
Users were publishing dubious findings on Ex, a platform that was formerly Twitter. “There are more nuances in historical contexts, and we’ll work to organize them further,” he said.
“Feedback loops are a component of the iteration process. Thank you, and please don’t stop.”