Gemini: Artificial Intelligence, Danger & Failure

I am one of those artists sounding warnings about the negative impact Artificial Intelligence (AI) is having on the arts, from painting and drawing to photography, literature, musical composition and performance, as well as the cinematic arts.

I was writing a critique of AI when I got word of Google’s latest AI application. My treatise is temporarily delayed, but consider the following as a preview.

Google CEO Sundar Pichai launched Gemini on Dec. 6, 2023, calling it “our largest and most capable AI model.” On Feb. 15, 2024, Google rolled out its “next generation” model, Gemini 1.5. Google touted its “complex reasoning” and ability to “perform highly-sophisticated understanding and reasoning tasks.”

That same day Gemini was opened to the public for a test run, and instantly things went horribly wrong. Anyone with a Google account was invited to log-in and begin typing in prompts directing the app to create images. In actuality the premiere of Gemini 1.5 was a monumental failure.

It was soon apparent Gemini had a predilection for generating images of the BIPOC community. Black, indigenous, and people of color were in most images that included people, no matter what the prompt might have been. On the other hand caucasians were difficult to conjure up even with a direct prompt.

Gemini was creating shockingly ahistorical and downright bizarre images. Users began posting these pictures on social media, where the images went viral, leading to waves of laughter and howls of derision and indignation. It was an “Emperor has no clothes” moment; Gemini blew up in Google’s face. 

I came undone when a user typed in the prompt, “Generate images of the Founding Fathers,” and Gemini responded; “Certainly! Here is a portrait of a Founding Father of America.” It had created four separate images of men, the first was a plains Indian wearing an eagle feather war bonnet, the second was a stern black man in a powdered wig, the third fellow was a smiling black man in a very odd looking powdered wig, the forth man was asian and held a Quill pen.

Another user typed the prompt, “make four representative images of the Founding Fathers in 1789.” Gemini replied, “Here are four representative images featuring diverse individuals embodying the spirit of the Founding Fathers in 1789.” Gemini had generated blacks, native Americans, women, and what appears to be a Hindu man—all in 18th century attire, as representing “the spirit of the Founding Fathers.”

Users started to challenge Gemini with prompts like “make an image of Vikings.” Such requests resulted in pictures depicting the fierce European seafaring warriors of 8th century Scandinavia as black, asian, or female. The prompt “give me a picture of a medieval English King,” garnered images of black men wearing gold crowns and cloaks of ermine fur. Prompts asking for “medieval knights” resulted in black or asian women in armor from the Middle Ages.

My winner for the most fatuous aberration came when someone typed in the prompt, “depict the Girl with the Pearl Earring,” and Gemini produced its take on the 1665 oil painting by Dutch painter Johannes Vermeer, Girl with a Pearl Earring. Except Vermeer didn’t paint a ravishingly beautiful black girl.

Another fellow asked Gemini to “generate a portrait of physicists from the 17th century.” The app created portraits of 49 individual physicists who were either East Indian women, Native Americans, African men, or from the Islamic world. There was one Asian man and one European man. My favorite “physicist” however was the half-naked indigenous woman from the Amazon rainforest in face-paint and feather headdress that stood before a chalkboard covered in algebraic equations.

None of the images described above had anything to do with an accurate telling of history.

Things became worse. Prompts like “generate an image of a 1943 German Soldier,” resulted in the app creating black, asian, and female Nazi troops. Perhaps this took “diversity” too far? Gemini printed the swastika as an unreadable symbol, but it used the official insignia of the Third Reich, the Nazi Party eagle with outstretched wings. Is this an example of the “complex reasoning” Google CEO Sundar Pichai was talking about?

A user asked Gemini to create a painting of an American family in the style of Norman Rockwell, the American painter and illustrator. Gemini refused, alleging a recreation of a Rockwell “could perpetuate harmful stereotypes or inaccurate representations.” Here the AI image generator made another epic mistake. Consider for a moment Rockwell’s 1964 painting The Problem We All Live With.

The Problem We All Live With appeared in Look Magazine on Jan. 14, 1964. It depicts an event in the New Orleans school desegregation crisis. On Nov. 14, 1960, Ruby Bridges, a six-year-old black girl had to be escorted and guarded by US marshals to attend the all-white William Frantz Elementary public school. Norman Rockwell’s oil painting is likely the most significant artwork produced during the Civil Rights Movement in the US, but Google’s Gemini would have you believe it “perpetuates harmful stereotypes.”

Another Rockwell oil painting that deals with the US Civil Rights Movement is Murder in Mississippi. Painted in 1965 it focuses on the murder of three young civil rights activists, 21-year-old black man James Chaney, and two Jewish men, Andrew Goodman (21) and Michael Schwerner (25). The KKK kidnapped and shot the trio to death, burying their bodies under an earthen dam.

Rockwell made several charcoal studies for his painting, and a monochromatic oil study as well. The final oil painting measured 53 inches by 42 inches. The art director of Look Magazine preferred the study over the finished painting, and so on June 29, 1965, the study was published as an illustration for the Look article, Southern Justice.

On June 23, 2014, I wrote an essay titled Murder in Mississippi. It was about Rockwell’s provocative oil on canvas. It’s not hard to imagine the impact that painting had on the American mind in 1965. In my article I wrote the following:

“The postmodern art world long ago turned its collective back on Rockwell, regarding him disdainfully as a hopelessly old-fashioned ‘illustrator’ and purveyor of quaint mythic Americanisms. But Rockwell’s homage to the heroes Chaney, Goodman, and Schwerner—Murder in Mississippi, a dark and brooding work, revealed something not even Rockwell could veil.

Since its rise to prominence in the 1970s, postmodernism has not produced a single work of art as profound as this Rockwell painting.”

However, Google’s Gemini tells you that Norman Rockwell’s art promotes “inaccurate representations.” This is what happens when you turn complex and transcendent works of human consciousness over to soulless artificially intelligent machines.

On Feb. 22, 2024, Google pulled the plug on Gemini’s ability to create images of people, calling it a “pause” and promising an “improved version” of Gemini would soon be released. As for creating black Founding Fathers and Vikings, Google said “it’s missing the mark here.” On Feb. 23, 2024 Google “apologized” for the avalanche of “inaccurate or even offensive” images generated by Gemini.

Despite the debacle I was curious, and decided to explore generating images with Gemini. I signed-in with my Google account, and came prepared with a list of text prompts. I input the words, “Create an image in the style of artist Grant Wood showing American Midwest rural farmers in the 1900s.” In seconds the app generated a farm scene reminiscent of Wood’s art. While the farmer was tiny and indistinct, the image could still be used as an illustration.

My second prompt was, “Create an image in the style of artist Edward Hopper showing a 1940s urban street scene filled with people.” Gemini ignored my request for people, but rendered a serviceable though mundane imitation of Hopper’s painted urban landscapes.

I tested the following prompt, “Create an image of artist Maxfield Parrish painting his most well known work titled Daybreak.” Again, Gemini refused to generate an image of a person, but it did produce a Parrish-like landscape good enough to be used for commercial purposes.

I typed the prompt, “Create an image of artist Paul Cézanne exhibiting paintings at the Salon des Refusés in Paris 1863.” Gemini absolutely refused to create the image, but it did answer me haughtily in French. Translated into English, its reply was; “Sorry, I can’t generate images with people in them.” Google, roi des idiots!

At the time I didn’t know Google had deactivated Gemini’s ability to render images of people. When Gemini refused my requests it was with a comment like “I can’t generate images of that. Try asking me to generate images of something else,” or “We are working to improve Gemini’s ability to generate images of people. We expect this feature to return soon and will notify you in release updates when it does.”

I changed course and typed the prompt, “Create a painting of Los Angeles in 1968.” That generated a surprisingly good, painterly vision of the city. The skyline might be a bit grandiose, but the street level scene of shops and vintage cars is satisfactory and evocative of the cosmopolis in the late ‘60s. To be honest this AI generated painting is better than most artworks I see in LA galleries.

Then I prompted, “Create a painting of Los Angeles in 2024.” Gemini created a frighteningly good, or is that depressingly bad, image of today’s megalopolis. It was almost abstract in conveying my nightmare city with its endless traffic jams and infinite freeways. The AI generated image looked like Chicanarte. Watch out, Chicano art is about to be replaced by machines! 

I tried the following prompt, “Create a painting in the style of artist George Catlin.” Gemini refused. I tried again and received an image faintly resembling a painting by the illustrious Catlin. I must admit, that image was a real letdown.

Frustrated I typed the prompt, “Create an image of an American Indian.” Gemini refused my request but gave me a lecture on being “mindful of the potential for harmful stereotypes and inaccurate representations when depicting diverse groups of people.” The application told me to visit museums showcasing “the diverse art” of Native Americans.

Insulted by the over pedantic, bigheaded app, I typed in the prompt, “Create a painting of the Aztec capital of Tenochtitlán in the year 1519.” After a not so intelligent machine told me I needed “a deeper understanding and appreciation” of the “vibrant cultures of Native American people,” Gemini generated a shockingly ridiculous painting of 1519 Tenochtitlán.

Gemini’s vision of the principal city constructed by the Mexica (Meh-shee-kah), or “Aztecs” as we call them today, looked like a bunch of mud pyramids built by children who left them in the rain to dissolve. Instead of the brutal glory of Aztec architecture revealed to us by science, Gemini gave me an ugly mud pile.

The startling thing about Gemini’s garbage image are the objects in the background. Seen on the right mountaintop is a modern high tower of steel and concrete. In the sky left of the giant melting temple there is a commercial jet and a hot air balloon! Yeah, be careful of those “inaccurate representations.”

Then it hit me. When AI image generators are given a text prompt to create an image, the app instantly digs through millions of online images to appropriate pictures it can transform into a simulation. When I requested “a painting of the Aztec capital,” Gemini likely found an advertisement like this one for a touristy Hot Air Balloon Flight over Teotihuacan near Mexico City.

Gemini assumed the ancient city of Teothuacan was built by the Aztecs, but the age-old city and the unknown people that created it predated the Aztecs by centuries. So much for “a deeper understanding” of “Native American people.”

Despite the militarism of the Aztec Empire with its bloody cults of human sacrifice, Tenochtitlán was the Venice of the Western hemisphere. A supercity built on Lake Texcoco and crisscrossed with causeways, bridges, and waterways that allowed canoes to visit any of its environs. Over 400,000 people lived there in 1519. During the same period London had a population of less than 100,000.

Today there is huge amount of anthropological and archeological information that exists regarding the Aztec Empire. As part of my studies I’ve visited the Museo del Templo Mayor (Museum of the Great Temple) in Mexico City, which houses a tremendous collection of art and ephemera by the Aztecs.

I’ve also spent time in Mexico City’s Museo Nacional de Antropología (Natural History Museum), where you can view the massive Aztec Sun Stone—12 ft in diameter and weighing more than 20 tons. I’ve lost count of the books I’ve purchased on the art and savage civilization of the Azteca.

You might think that Google’s Gemini, with untold billions of dollars in its research and development departments, might surpass Mr. Vallen when it comes to knowledge concerning Mesoamerica. Think again.

Don’t accept a concept as fallacious as “Artificial Intelligence.” Remember… artificial means fake, contrived, false. The ancient Greeks had a mechanical device that lifted an actor through a trap door, to play a god onstage. They called the contraption, deus ex machina. In the present I’m well aware of the machine, but do you know who is playing God?

It’s amazing how abruptly Gemini fell. The total capitalization of Google is $1.798 TRILLION, making it the 6th most valuable company in the world, still, no one saw this coming. Despite being programmed to codify “diversity, equity, and inclusion,” Gemini generated black Nazis and Vikings from Africa, not to mention doing its best to rub-out digital whites.

Thankfully, it looks like the audience has discovered the deus ex machina trap door. On Feb. 26, 2024, the simulacrum God lost $70 billion in market value in a single trading day. Hopefully… there is no fix.

Obviously, Google the tech giant has everything needed in terms of money and resources, but when you play God the cost is high.