Fake Trump arrest photos trigger a new AI panic
Yes, the fake-photo singularity is coming. But panic is the wrong response.
Before OpenAI, the company behind both ChatGPT and DALL-E, the sustained panic centered around deepfake videos, produced by generative adversarial networks (GANs). Everybody feared that deepfake videos would become the ultimate tool for spreading disinformation. Deepfakes, we were told, could produce “evidence” that political leaders, for example, said or did things they never said or did.
Deepfake fears haven’t been realized yet. The tech isn’t convincing enough. (And, once they are convincing, I fear the bigger threat will be that people caught on video will plausibly deny the truth by falsely claiming the real video is a deepfake.)
But this week a new panic set in. Rumors circulated about the imminent indictment and arrest of former president Donald Trump (rumors started by Trump himself in a Truth Social post that falsely claimed he’d be arrested Tuesday).
Then images showing Trump being arrested began circulating on social networks. The pictures blew up, and have been viewed millions of times on Twitter alone. (That Trump would be arrested is plausible. He’s accused of crimes that his alleged co-conspirator, Michael Cohen, has already served prison time for.)
Many of these pictures were created with Midjourney by Eliot Higgins, founder of the investigative group Bellingcat. As part of the collective panic, Higgins said he got banned by Midjourney, which then also banned use of the word “arrested” as a prompt for creating pictures.
Funny arrest images were just the beginning. Additional AI “pictures” feature Trump being tackled by NYPD…
running from the cops…
crying in a Congressional hearing…
picking out an orange jumpsuit…
filing out prison paperwork…
doing janitorial work in prison…
and the most unrealistic of all: reading a book.
The series goes on to depict Trump in prison eating lunch, playing basketball and hanging out in the “yard.” They even show him escaping from prison and making it to McDonald’s.
The AI-generated images of Trump getting tackled by New York cops are no more convincing than faked pics of Trump dunking a basketball.
Personally, I didn’t experience even a half-second of misunderstanding over any of these pictures, and for two reasons: 1) I can easily tell the images are AI-generated; and 2) the scenes are unbelievable — literally not believable.
Also: Higgins clearly labeled them all as fake.
That some were fooled demonstrates an extreme lack of understanding about both what’s possible with AI and also how the world works. (When Trump gets arrested, he’s not going to be tackled in Times Square by a dozen cops struggling to contain him.)
I think AI panic is premature, and would recommend scheduling that panic for later. The fact is that a vastly more convincing fake image is possible from a skillful Photoshop user than a skillful Midjourney user. Other pictures of Trump’s arrest — such as these fake mugshots — are produced presumably by good old-fashioned Photoshop, and are significantly more convincing than AI images.
Still, the AI-panic faction came up with an odd justification for their calls to ban the pictures on social: They make a lasting impression even among people who know they’re fake. In other words, the public has been provided with dozens of images of a Trump arrest, conviction, imprisonment and, hilariously, jailbreak but no counterbalancing imagery of a non-arrest. And so, despite “knowledge,” the public “perception” is that Trump got convicted.
They’re also concerned about the good old-fashioned slippery slope. Higgins used the shiny new Midjourney version 5. How believable will fake photos made with future versions 6, 7, 8, or 15?
Panic is premature. Here’s the real fake-everything challenging humanity, and what we should all do about it.
Mike’s List of Shameless Self Promotions
Can’t hire? Can’t get hired? How to avoid the “Great Mismatch” trap!
The Chinese government, which banned TikTok, says Europe's ban on TikTok is bad for business. Why?
‘Digital twin’ tech is twice as great as the metaverse
What’s new on the ELGAN DOT COM blog!
If only that photo were true. (sigh) I would have preferred them clubbing fRump like a baby seal but that might of made it even More fake.
I believe it's fake because of the "tells" in it.
One, his head is directly centered.
And Two, because his head is much brighter than everyone else in the photo and the photo in general. Just doesn't look right to me.