Midjourney’s CEO David Holz suggests that the company is close to finalizing images of Donald Trump, Joe Biden, and others “for the next 12 months.”
Midjourney, the pioneering artificial intelligence image-generating company, is contemplating a ban on the use of its software for creating political images of Joe Biden and Donald Trump. This move is aimed at preventing the software from being utilized to divert attention from or spread misinformation about the 2024 US presidential election.
“I’m unsure about the extent to which political speech should be allowed on our platform in the next year,” stated Midjourney’s CEO, David Holz, last week. He mentioned that the company is on the verge of “hammering” or prohibiting political images, including those of the prominent presidential candidates, “for the next 12 months.”
During a conversation with Midjourney users in a Discord chatroom, as reported by Bloomberg, Holz further remarked, “I know creating Trump pictures can be entertaining – I create Trump pictures myself. Trump is visually captivating. However, it’s probably wiser to refrain from doing so, to step back a bit during this election. We’ll see.”
AI-generated imagery has recently raised concerns. Two weeks ago, the appearance of pornographic images featuring Taylor Swift’s likeness prompted lawmakers and her supporters (known as the Swifties) to call for stronger protections against AI-generated images.
The origin of the Swift images was traced to 4chan, a community message board frequently associated with the dissemination of sexual, racist, conspiratorial, violent, or otherwise antisocial material, with or without the use of AI.
Holz’s remarks coincide with ongoing efforts by image-generator operators to outmaneuver users in preventing the creation of questionable content.
There is growing concern about AI’s role in politics, although the MIT Technology Review recently observed that discussions about how AI might jeopardize democracy “lack imagination.”
The review highlighted concerns about campaigns that use fake images, audio, or video to attack opponents, noting that we have extensive experience dealing with manipulated images over decades. It suggested that attributing a surprising election outcome to a specific AI intervention would be challenging.
However, in October, the image-generation company Inflection AI announced that its chatbot, Pi, would not endorse any political candidate. Co-founder Mustafa Suleyman stated at a Wall Street Journal conference that chatbots “likely need to remain a human element of the process,” even if they operate flawlessly.
Meta’s Facebook announced last week that it intends to label posts created using AI tools as part of a broader effort to combat misinformation during election years. Additionally, Microsoft-affiliated OpenAI stated that it would include watermarks on images created with its platforms to counter political deepfakes generated by AI.
In a blog post last month, the company stated, “Safeguarding the integrity of elections demands cooperation from all aspects of the democratic process, and we aim to ensure that our technology is not employed in a manner that could jeopardize this process.”
At a recent event, OpenAI chief executive Sam Altman expressed his primary concern, stating, “I am most worried about the advancement of AI capabilities leading to the creation of more sophisticated deepfakes than those seen in 2020.”
In January, a manipulated audio call, falsely attributed to Joe Biden advising New Hampshire voters to stay home, highlighted the potential for AI-driven political manipulation. Subsequently, the FCC implemented a ban on the use of AI-generated voices in robocalls.
David Ryan Polgar, president of the non-profit All Tech Is Human, previously told the Guardian, “We’re increasingly recognizing the growing gap between rapid innovation and our societal capacity to collectively comprehend best practices, behavioral norms, the necessary actions, and the need for new legislation. Progress in these areas is still painfully slow.”
Midjourney’s software generated a fake image depicting Trump being handcuffed by agents, among others that have surfaced online. These include images of Biden and Trump as elderly men knitting sweaters together, Biden smiling while firing a machine gun, and Trump meeting Pope Francis in the White House.
The software already incorporates several safeguards. Midjourney’s community standards guidelines prohibit images that are disrespectful, harmful, portray public figures/events misleadingly, or have the potential to mislead.
Bloomberg noted that permissibility varies depending on the software version used. An older version of Midjourney produced an image of Trump covered in spaghetti, while a newer version did not.
However, if Midjourney decides to ban the generation of AI-generated political images, consumers— including voters—will likely be unaware.
“We’ll probably just prohibit it without announcing anything,” Holz remarked.
