Google is rushing to repair its new AI-powered image creation tool in response to allegations that it overcorrected images to prevent the appearance of racism.
Users alleged that the company’s Gemini algorithm generated images that historically portrayed an assortment of ethnicities and genders, despite the fact that such depictions were inaccurate.
An inquiry for visual representations of the founding progenitors of the United States, for instance, returned results featuring women and individuals of color.
The organization claimed its instrument fell short of expectations.
“The AI image generation of Gemini does produce a diverse array of individuals. Additionally, this is generally a positive development due to its widespread usage across the globe. “However, it falls short in this regard,” senior director of Gemini Experiences Jack Krawczyk stated on Wednesday.
He added, “We are working diligently to improve these types of depictions immediately.”
Later, Google announced that while it worked on a fix, the tool’s ability to generate images of individuals would be temporarily disabled.
This is not the initial instance in which artificial intelligence has encountered practical challenges pertaining to diversity.
Nearly a decade ago, Google infamously issued an apology for labeling a photo of a black couple as “gorillas” in its photos app.
OpenAI, a competitor in the AI industry, was also accused of perpetuating detrimental stereotypes after users discovered that its Dall-E image generator produced predominantly white male images in response to queries for the chief executive, for instance.
In an effort to demonstrate that it is not lagging behind in artificial intelligence advancements, Google unveiled Gemini, its most recent iteration, last week.
The algorithm generates images in reaction to textual inquiries.
Critics rushed to the bot’s defense, alleging that the company had instructed it to be lamely awake.
Debarghya Das, a computer scientist, wrote, “Getting Google Gemini to acknowledge the existence of white people is embarrassingly difficult.”
“Come on,” said author and humorist Frank J. Fleming, who has contributed to publications such as the right-wing PJ Media, in reaction to the responses he received requesting an image of a Viking.
In right-wing circles in the United States, where numerous major technology platforms are already encountering censure for purported liberal bias, the claims gained momentum.
The company, according to Mr. Krawczyk, took representation and bias seriously and desired for its results to accurately reflect its global user base.
“We will continue to fine-tune to accommodate the nuance of historical contexts,” he wrote on X, formerly Twitter, where users were discussing the questionable outcomes they had obtained.
“This constitutes an iterative process of alignment based on feedback.” Continue providing such gratitude.
