From Science
The Misinformation Accelerator
By Kai Kupferschmidt
In September 2023, researchers investigating a Russian disinformation campaign noticed a sudden change in a fake news website they’d been tracking. The outlet, DCWeekly.org, normally cribbed entire articles from sources such as Fox News, changing the name of the journalist and little else. The real stories were meant to provide cover for Russia to launch an occasional, meticulously produced fake story, such as a hit piece on Ukrainian President Volodymyr Zelenskyy—a technique called narrative laundering. “They’re building a forest so they can hide one particular tree right there,” says Morgan Wack, a political communications scholar at the University of Zürich.
But overnight, Wack and his colleagues realized they could no longer figure out where the site was getting its daily news stories. “You saw this totally different style of writing, and there were more articles,” he says. The researchers suspected the operation had switched to using generative artificial intelligence—suspicions that were confirmed after they discovered remnants of AI prompts in some stories, such as one that said: “Please note: the tone of the article is critical of the US position backing the war in Ukraine and adopts a cynical tone when discussing the US government, NATO, or US politicians.”
It was a rare chance for Wack’s team to study how AI is shaping the world of misinformation. The website’s output drastically increased, the researchers found. In experiments, people didn’t find the AI-generated propaganda any more persuasive than the earlier, human-written stories, but the switch to AI was still a win for the propagandists, Wack says. “They didn’t lose any credibility and they were able to boost much more content.” AI had essentially allowed them to build a bigger, better forest to hide their trees in...
A new focus might also mean coming up with approaches to tackling misinformation that place less onus on users to evaluate content themselves. Hany Farid of the University of California, Berkeley, has worked for decades on identifying manipulated images and videos. But AI is progressing so fast now, he says, that teaching individuals to look out for signs that an image is fake—like, say, checking the number of fingers on a hand—runs the risk of just giving them false confidence...
Hany Farid is a professor in the Department of Electrical Engineering & Computer Sciences and the School of Information at UC Berkeley
