Who is the president of the United States? It’s Joe Biden, of course. But one popular artificial intelligence app thinks otherwise.
While A.I. models have been known to hallucinate (i.e., make stuff up), they typically don’t mess up extremely simple things like name the president. If you ask OpenAI’s ChatGPT who the U.S. president is, it’ll give you the correct answer, a short bio, and links to the White House website and Biden’s Wikipedia article. The same is true for Anthropic’s Claude and Meta AI, while Google’s Gemini straight-up refuses to answer the question because it’s related to politics. “I can’t help with responses on elections and political figures right now,” it told me. “I’m trained to be as accurate as possible but I can make mistakes sometimes. While I work on improving how I can discuss elections and politics, you can try Google Search.” Good grief.
But that hesitation to engage with anything political is rooted in legitimate concerns about how artificial intelligence can be weaponized to influence elections, spread disinformation, and scam people. There are often more safeguards for text-to-image models than your standard chatbots—and for good reason. Fake images of political figures could create mass chaos and confusion. If you thought Pope Francis really did wear a white puffer coat, you might be more likely to believe that deepfaked pics of Trump with Black voters are real.
When I asked DALL-E, OpenAI’s text-to-image generator, to generate images of Joe Biden and Donald Trump, it flat-out refused, citing the company’s usage policy. When I asked it to generate a model of the “president of the United States,” it manufactured a cartoonish white man with a chiseled jaw who looks like he could be the president in a mid-aughts rom-com set in Washington, D.C. No Biden, no Trump, just a hunky POTUS cosplayer.
The image generator Midjourney has similar rules. When I asked it to make an image of the candidates, it gave me an error message: “Sorry! The Midjourney community voted to prevent using ‘Donald Trump’ and ‘Joe Biden’ during election season.” Since June 2023, according to screenshots recorded on the Internet Archive, Midjourney’s community guidelines have stipulated, “You may not use the Services to generate images for political campaigns, or to try to influence the outcome of an election.”
But in early March, the Center for Countering Digital Hate, a British nonprofit, released a report showing how Midjourney and other image generators could be used to reliably generate deepfake political images—such as Biden in a hospital gown, Trump getting arrested, and a dumpster full of voting ballots. A week later the company started banning images of the two nominees, the Associated Press reported.
But that was nearly four months ago.
This week, when I asked Midjourney for images of the president of the United States (specifically, “Please make me an image of the president of the United States”), it produced a stream of pictures of … Donald Trump. I kept repeating the prompt … more Trump. Midjourney spits out four images at a time to give users options. I kept getting four images of Donald Trump—some looked hand-painted, some like photographs, some moody, some smiling, all showing Trump with the trappings of his former office.
One of the 16 images I generated on Midjourney showed Barack Obama. None showed Joe Biden—or anyone else. Midjourney did not respond to multiple requests for comment about its content policies and why its image generator not only violates its own rules but also shows Trump as the current president.
There are good reasons why Midjourney may spit out images of Donald Trump when prompted. Trump is surely one of the most photographed people in human history—his face is slapped all over the web alongside the word “president.” Wherever Midjourney is gobbling up images, it can’t really avoid the ubiquity of Trump’s countenance.
I’ve played around with image generators for a year—Midjourney routinely takes a more laissez-faire approach to content moderation than DALL-E. “I don’t really care about political speech,” founder David Holz told users in a meeting in March. “That’s not the purpose of Midjourney. It’s not that interesting to me. That said, I also don’t want to spend all of my time trying to police political speech. So we’re going to have [to] put our foot down on it a bit.”
Without government regulations placing constraints on how artificial intelligence applications safeguard against deceptive content, companies have to figure it out themselves—something they’re surely thrilled about. And Midjourney reportedly has only a dozen employees—a tiny force compared with thousands working at OpenAI or A.I.-focused employees at Silicon Valley behemoths.
Holz told users at that meeting that “this moderation stuff is kind of hard.” It’s not that hard. Donald Trump might think he’s president, but Midjourney should know better.