Fake photos: AI deepfakes spread online
new york –
Former US President Donald Trump is attacked by New York City police in riot gear. Russian President Vladimir Putin wears a gray suit behind bars in a dimly lit concrete cell.
The highly detailed and sensational imagery recently flooded Twitter and other platforms amid news that Trump faced possible criminal charges and that the International Criminal Court had issued an arrest warrant for Putin. There is
But neither visual is realistic. The images, and the various variations that are scattered across social media, were created using increasingly sophisticated and widely accessible artificial intelligence-powered image generators.
Misinformation experts warn that these images portend a new reality. Waves of fake photos and videos flood social media after major news events, further confusing fact and fiction at a crucial time for society.
Professor Jevin West of the University of Washington in Seattle, which focuses on the spread of misinformation, says, “It adds noise during crisis events. It also increases the level of cynicism.” Told. “You start to lose faith in the system and the information you are getting.”
The ability to manipulate photos to create fake images is not new, but AI image generation tools such as Midjourney, DALL-E are easier to use. It can quickly generate realistic images with detailed backgrounds at scale with little more than a simple text prompt from the user.
Some of the recent images are due to the release of a new version of Midjourney’s text-to-image synthesis model this month. The model can now produce compelling images that mimic the style of news agency photography.
In one widely circulated Twitter thread, Eliot Higgins, founder of the Dutch-based investigative journalism collective Bellingcat, used the latest version of the tool to create a dramatic image of Trump’s fictional arrest. reminded me a lot of
The visual, which has been shared and liked tens of thousands of times, showed a mob of uniformed cops grabbing the Republican billionaire and violently dragging him down the sidewalk.
Higgins, who was behind the string of images of Putin being arrested, tried and then imprisoned, says he posted the images with malicious intent. In his Twitter thread, he clearly states that the image was generated by his AI.
Still, according to Higgins, the image was enough to lock him out of Midjourney servers. The San Francisco-based independent laboratory did not respond to an email seeking comment.
In an email, Higgins said, “The Trump arrest image just subtly shows how good or bad Midjourney is in rendering the actual scene.” As I began to form stories, I decided to piece them together into a story and finish the story.”
He noted that the images are far from perfect. In some, Trump is seen wearing a police utility belt, oddly enough. is distorted.
But it wasn’t enough for users like Higgins to explicitly state in their posts that the images were generated by AI and were just for entertainment, according to the New York-based human rights group. says Shirin Anlen, a media tech at the collective Witness.
Too often, she said, visuals are quickly re-shared by others without significant context. The Instagram post has garnered over 79,000 likes.
“You’re just looking at an image, and once you see something, you can’t erase it,” Anren said.
In another recent example, a social media user shared a composite image that appears to show Putin kneeling and kissing the hand of Chinese President Xi Jinping. The image, which circulated when the Russian president welcomed Xi to the Kremlin this week, quickly became a crude meme.
It is not clear who created the images or what tools were used, but several clues have revealed the forgery. It did not match the room where the actual meeting took place.
Experts say the best way to combat visual misinformation is to raise public awareness and education, as synthetic images become increasingly difficult to distinguish from the real thing.
“It’s becoming so easy and inexpensive to create these images that we should do whatever we can to make the public aware of how good this technology is,” said West. said.
Higgins suggests that social media companies could focus on developing techniques to detect AI-generated images and integrate them into their platforms.
Twitter has a policy prohibiting “synthetic, manipulated, or out of context media” that could deceive or harm. A note from Community Notes, Twitter’s crowdsourced fact-checking project, was attached to include context.
When asked for comment on Thursday, the company emailed back only an automated response.
Facebook and Instagram parent company Meta declined to comment. Through a third-party fact-checking program in which AP participates, some of the fabricated Trump images have been labeled as either “false” or “lacking context.”
Arthur Holland Michel, a fellow at the Carnegie Council on International Affairs and Ethics in New York who focuses on emerging technologies, said he was worried the world was unprepared for the impending deluge.
He wonders how deepfakes involving members of the public, such as harmful fake photos of ex-partners or co-workers, for example, would be regulated.
“From a policy standpoint, I’m not sure we’re ready to deal with disinformation on this scale at all levels of society,” Michelle said in an email. will require technological breakthroughs that were previously unimaginable.”
Associated Press correspondent David Klepper in Washington contributed to this article.