*This AI-generated summary is based on reporting by TaiwanPlus and verified by newsroom fact checkers. Learn more about our policy here.
A new report from Taiwan-based fact-checking group FactLink warns that generative artificial intelligence is ushering in a new era of influence operations that may pose an even greater threat to democracy than traditional disinformation.
The study, which examined Taiwan’s recent mass recall elections, found that AI-generated “mini clips” were deployed online to sway emotions and muddy the line between fact and fiction. These videos, just a few seconds long, lack traditional narratives but pack a powerful punch.
“In our recent report, we are just focused on a very new phenomenon, uh, AI generated mini clips,” said Summer Chen, founder of FactLink. “It was very short, like just a few seconds. And, uh, there is no text, no story, but it has a very strong, uh, emotion. It tries to trigger our emotion. And it's spreading fast on reels and shorts. Because the algorithm.”
Unlike earlier waves of disinformation, which relied heavily on long-form text and videos, these clips are stripped-down bursts of emotional content that spread quickly on platforms such as Facebook. “Mostly on the Facebook reels, like show radio. It's just like you just you just scroll the Facebook and you just pump up to my screen,” said Mary Ma, FactLink’s head of investigations.
The team’s findings suggest that the nature of influence operations is shifting from arguments over factual accuracy to an assault on emotional stability. “It's trying to trigger our emotion. It's pretty targeted. It's bypassed the critical thinking and has no room for audience or for fact checkers to analyze this,” Chen said. “And it's just trying to, uh, reinforce our stereotype.”
That shift, researchers argue, makes AI-driven campaigns more insidious than traditional disinformation. “Because I think people should know, not just focus on is this AI or not? They just don't care about this AI. They just know this is AI. But they want to like, like share this video because this video is like propaganda,” Ma said.
FactLink investigators also traced many of the clips to tools developed in China. “We've found those mini clips has watermark, and the watermark turned out to be very important or strong clues for us to trace,” Chen said. “And we found most of those mini clips was produced by Chinese AI tools.”
Still, Chen stopped short of directly attributing the campaign to Beijing. “What we can prove is those, uh, mini clips was produced by Chinese AI tools, but we can't make sure it was conducted by China,” she said.
The emotional targeting raises fresh challenges for democracy and for fact-checkers, who traditionally focus on verifying information. “Just like someone say, like our methodology must be shift. It's not just focus on debunking. We need to talk the public,” Ma said. “To let them know that AI is using by influencing operation.”
Despite the challenges, FactLink’s team expressed some optimism. “I have more like, optimistic about about these questions because I think we always say that AI we must detect AI. But AI is not just that bad, it is for good too,” Ma said. “For example, they need to use the ChatGPT accurately. Certainly, and they need to learn about it.”
While FactLink could not confirm whether the AI-driven campaigns altered the results of Taiwan’s recall elections, Chen said the group’s aim was to highlight the growing use of such tactics. “So what do we see is how about in the digital world and what's going on on the digital world, and what kind of propaganda happened on the digital world? So that's why we focused on what's going on in the digital world,” she said.
FactLink plans to continue investigating AI in information operations, including suspicious YouTube channels. “We just found some, like, AI generated YouTube channel. So we still are working on it and investigating it,” Ma said.
As generative AI tools become cheaper and easier to use, experts warn Taiwan may be the testing ground for tactics that could spread worldwide.
14:48
14:47
14:48
14:48
14:48
13:39
14:49
14:48
13:43
14:07
14:18
14:47
14:48
14:17
14:51
14:53