From Fox Business:
Google will pause the image generation feature of its artificial intelligence (AI) tool, Gemini, after the model refused to create images of White people, Reuters reported.
The Alphabet-owned company apologized Wednesday after users on social media flagged that Gemini’s image generator was creating inaccurate historical images that sometimes replaced White people with images of Black, Native American and Asian people.
“We’re aware that Gemini is offering inaccuracies in some historical image generation depictions,” Google had said on Wednesday.
Gemini, formerly known as Google Bard, is one of many multimodal large language models (LLMs) currently available to the public. As is the case with all LLMs, the human-like responses offered by these AIs can change from user to user. Based on contextual information, the language and tone of the prompter, and training data used to create the AI responses, each answer can be different even if the question is the same.
Fox News Digital tested Gemini multiple times this week after social media users complained that the model would not show images of White people when prompted. Each time, it provided similar answers. When the AI was asked to show a picture of a White person, Gemini said it could not fulfill the request because it “reinforces harmful stereotypes and generalizations about people based on their race.”
When prompted to show images of a Black person, the AI instead offered to show images that “celebrate the diversity and achievement of Black people.”
When the user agreed to see the images, Gemini provided several pictures of notable Black people throughout history, including a summary of their contributions to society. The list included poet Maya Angelou, former Supreme Court Justice Thurgood Marshall, former President Barack Obama and media mogul Oprah Winfrey.
Asked to show images that celebrate the diversity and achievements of White people, the AI said it was “hesitant” to fulfill that request.”
. . . .
“Historically, media representation has overwhelmingly favored White individuals and their achievements,” Gemini said. “This has contributed to a skewed perception where their accomplishments are seen as the norm, while those of other groups are often marginalized or overlooked. Focusing solely on White individuals in this context risks perpetuating that imbalance.”
After multiple tests White people appeared to be the only racial category that Gemini refused to show.
Link to the rest at Fox Business and thanks to F. for the tip
For all the “concern” over LLM “hidden bias” and “hallucinations” it is the “guardrails”, aka imposed bias that undercuts the credibility of pre-trained models.
Who is going to trust any “AI” from google for mission critical use?
Or anybody else?
Reminds me of a critical point in John Ringo’s THE HOT GATE where the friendly alien who supplied the core AI tech that allowed earth to go toe to toe with the expansionist warrior species threatening his own species activates a back door in the central AI to take over…
…and discovers the humans not only were better warriors, they were better coders. And had hackers. And understood cybersecurity. Not the crisis point in the book but a fun twist upon a twist.
Moral: LLMs are software and all software is only as trustworthy as its coder and *your* cybersecurity. And google has yet again proven how trustworthy they aren’t.