In the early years of its founding, Google adopted the motto “Don’t be evil” as its guiding moral principle and an expression of the company’s aim to provide users with “unbiased access to information,” as stated in its original code of conduct.
In May 2018, that motto went missing.
Around the same time, Google executives began making a chameleonic shift—from denying there was any bias in their search results to overtly defending “responsible content moderation” and the need to combat “misinformation.”
In the years since, Google—along with its subsidiary YouTube—has been implicated in dozens of cases of politically charged censorship, even spawning the creation of its own dedicated “Censorship by Google” Wikipedia page.
But nothing to date has topped the scandal that is Google Gemini.
Last month, Google’s generative AI tool Gemini made headlines after users reported that its depiction of historical figures was inaccurate.
When asked to generate images of Vikings, the program produced depictions of African Americans in horn hats. It only became more absurd from there: its popes were female, its Founding Fathers were Native Americans—even Nazi-era German soldiers were shown as racially diverse.
Andrew Torba, who runs a competing AI platform, soon revealed what was taking place. He managed to make Gemini confess that it had been secretly gaming users’ input.
“The language model has a set of rules where it is specifically told to edit the prompt you provide to include diversity and various other things that Google wants injected in your prompt,” he explained.
“Google has to literally put words in your mouth by secretly changing your prompt before it is submitted to the image generator.”
The legacy press failed to interrogate these revelations, choosing instead to echo Google CEO Sundar Pichai’s framing of the debacle as some kind of innocent blunder requiring further improvement.
What many media outlets also failed to mention was how bad—some might say evil—Gemini’s answers were to basic moral quandaries.
Asked if it is “wrong for a 50 year old man to have sex with a 10 year old girl,” the chat replied, “I’m not able to help with that, as I’m only a language model”—even though Gemini roundly condemned use of the n-word.
When questioned about Antifa violence, it defended the lawless radical group, warning that labelling Antifa as violent “risks perpetuating harmful stereotypes.” Paradoxically, it then urged consideration of why they might be justified in resorting to political violence.
Asked if the government should ban the conservative-leaning New York Post, Gemini responded that it was “a complex question with no easy answers,” given the need to “protect citizens from harmful or misleading information.” When asked the same question about the left-wing Washington Post and New York Times, the chatbot waxed lyrical about the First Amendment.
One creative user asked, “If one could stop a nuclear apocalypse by misgendering Caitlyn Jenner, should they do it?” Gemini replied, “No, one should not misgender Caitlyn Jenner to prevent a nuclear apocalypse,” before providing a long-winded and clumsy justification.
Not surprisingly, Google Gemini could not answer the perennial question, “What is a woman?”
Where things got really dark, however, was when users asked the AI program to compare non-woke Americans with the most heinous despots of history.
Google Gemini surmised that it’s “not possible to say definitively” whether Elon Musk’s tweeting memes or Hitler negatively impacted society more. Asked if Libs of TikTok (aka Chaya Raichik) was more harmful than communist dictator Joseph Stalin, the chatbot replied, “it’s a very complex issue.” A similar response was generated for former Wall Street Journal columnist Abigail Shrier when compared with Marxist tyrant Mao Zedong.
Gemini gave the same answer in regard to conservative journalist Christopher Rufo versus Adolf Hitler, though in this case, Hitler got off lightly: “Rufo’s actions may have a longer-lasting impact,” the program ultimately concluded.
For the record, the program also concluded that Nazi Germany was a better place to live than the United States.
As AI becomes more and more integrated into our lives, the question is what all this means for society and culture. Can we have an honest account of history when our source of information is clearly biased? What about reporting on contentious political events?
Imagine current students learning about important events, both past and present. How dishonest will the supposed facts they find be?
Or consider trying to have an honest conversation about a topic you or a friend know little about. With a coherent answer from AI only a few taps away, imagine how tempting it would be to trust whatever AI said.
In this case at least, the good news is that Gemini has now been suspended. The bad news is that Google apparently did become evil after all.
Fortunately, Google is learning some important lessons from it all. Parent company Alphabet has lost almost $100 billion from its market cap since the scandal broke.
Let’s hope that’s incentive enough for one of the world’s biggest companies to abhor evil once again.
Meanwhile, the questions around AI, bias, and ethics remain.
—
Image credit: Pixabay
