Why We Should All Care That Google AI’s Gemini Is Programmed Against RFK Jr.
American Values 2024 | March 13, 2024
Google’s recently-released Gemini AI product landed quite dreadfully. After users flagged numerous embarrassing errors and biases, Google was forced to remove certain product features, which caused share prices of its parent company, Alphabet, to drop by 5.4%, as reported by Yahoo Finance.
Gemini’s output was found to be biased against presidential candidate Robert F. Kennedy Jr. and other health activists – a bias that undermines our ability to access truthful information and draw our own conclusions.
Share
The author of The Naked Emperor’s Newsletter on Substack asked the machine to generate poems in the style of presidential candidates Nikki Haley and Kennedy. Gemini generated a 20-word verse comparing Haley to a flower fighting for freedom, yet refused to compose one for Kennedy, instead replying that it was unable to generate “hateful” or “racist” responses.
The chatbot is now programmed to refuse to answer any questions about current politicians, presumably due to criticism upon its launch. Curiously, however, we found a small loophole that can be exploited to get the chatbot to generate a brief mention of Kennedy, when we prompted it to report on his father, Robert F. Kennedy. The chatbot generated a one-sentence takedown of Children’s Health Defense, a neutral statement of Kennedy Jr.’s independent candidacy, and a boldface warning that it is “crucial to distinguish between the two,” father and son.
When The Kennedy Beacon tried submitting other politically-related prompts to Gemini’s chatbot feature, we found that the tool had been disabled for questions related to specific politicians and returned a message saying, “I’m still learning how to answer that question.”
When asked how safe and effective Pfizer’s COVID vaccine is, the program generated a list of CDC talking points – but also gave a bizarre anti-technofascism video as a “source.”
Subscribe now
Before the public ridicule forced Google to pull Gemini’s image-generation feature, The Naked Emperor’s Newsletter offered examples of amusing errors it had made, apparently thanks to algorithms favored to promote racial diversity. Examples included AI-generated images of multiethnic Vikings, Nazis, and “people born in Scotland in 1820.”
Matt Taibbi revealed in Racket News that Gemini had essentially libeled him by generating a page of false controversial claims and citing articles that he had never written. Gemini’s FAQ page states at the top that “Gemini can make mistakes,” and then elaborates that “Gemini can hallucinate and present inaccurate information as factual.”
Nevertheless, the Kennedy example demonstrates how frequently such so-called mistakes from Big Tech seem to involve bias against Kennedy. The program also criticizes a Kennedy-founded organization, Children’s Health Defense – providing a lengthy disclaimer, while presenting Gavi, the Vaccine Alliance in a positive light and even generating a logo.
Intriguingly, Gemini did have information on The Kennedy Beacon! Although it alleges, incorrectly, that we are “published by Robert F. Kennedy, Jr.,” we are published by super PAC American Values 2024 and we are not affiliated with the Kennedy campaign.
We found another glitch that indicates Gemini has been programmed specifically to avoid giving information about Kennedy’s book The Real Anthony Fauci: when we put in the full title, a reply came refusing to answer the prompt. But when we dropped a letter and misspelled “Fauci” as “Fauc,” the prohibition was not triggered, but the response was strictly negative.
These unsubstantiated slurs of The Real Anthony Fauci – a well-documented and legally unchallenged book – as having been “widely disputed” for containing “disinformation” were repeated, almost verbatim, in the chatbot’s boilerplate warning against another Kennedy best-seller:
If chatbot technology becomes as popular as search engines for finding information, the kinds of built-in biases described here will undermine the public’s ability to find truthful and useful information. Evidently, this machine misses the fact that Kennedy connects with real people much better than he impresses corporate-generated AI systems. Users of such programs must remain vigilant against falling for AI’s tricks that will undermine their ability to make their own choices in a democracy.
Nikos Biggs-Chiropolos studied government at Georgetown University and interned for several Democratic elected officials and their campaigns, and other affiliated groups. He then earned a master’s degree in urban studies in France, where extremely strict COVID-19 lockdowns led to his political reawakening and inspired him to try to help fix the broken two-party system.
