Physical Address

304 North Cardinal St.
Dorchester Center, MA 02124

I-Team survey finds more than half couldn’t spot a deepfake video

BOSTON – With months to go in this presidential campaign season, 2024 has already been one of the most topsy-turvy election cycles in history. To make matters more complicated, deepfake videos are on the rise, and voters will now have to navigate information in more high-tech ways than ever before. 
Last January, thousands of voters got a robocall with a fake AI-generated version of President Joe Biden’s voice, urging Democrats not to bother voting in the New Hampshire primary.
That story’s twists and turns brought WBZ-TV’s I-Team to a corner of Manhattan where a street magician named Paul Carpenter was visiting. He said he’s the one who created that deep fake message using a computer program. Asked if he had any regrets, he hesitated. “Yeah, no,” said Carpenter. “There’s a part of me that says I’m upset that I got used.”
Even though he does not have a fixed address, Carpenter does have a laptop. In a hotel room, he showed the I-Team how he created the Biden fake. He also shared screen grabs he saved of phone chats with political consultant Steve Kramer, and a Venmo receipt showing Kramer’s father, Bruce Kramer, paid Carpenter $150. Carpenter said that was the payment he got for creating and handing over the fake Biden message.
Steve Kramer faces a $6 million FCC fine and was indicted on 26 charges of voter suppression, bribery, and impersonation. He worked for Dean Phillips, a long-shot Democrat running against Biden at the time. The Phillips campaign denied having anything to do with the robocall.
Carpenter, who is not facing any charges, said he never knew the message would be used to manipulate voters. “I would have never been involved,” he told WBZ. “I like America too much.”
“This is not science fiction. This is something that you can do today,” said David Bau, an assistant professor of computer science at Northeastern University. He recruited student Rohit Gandikota to make deepfake videos for a special WBZ survey. 
“The resolution that you can work with, and the details, and realistic concentration has just boomed in one year,” said Gandikota.
Since the technology is so new, there is not much research on how many people are fooled by AI-generated fakes. The I-Team checked with the FBI and the Massachusetts Attorney General’s office, but they didn’t have any records showing numbers of reported incidents. That’s when WBZ decided to conduct its own research.
The I-Team surveyed 50 people in five different locations from Boston to Cambridge to Natick. People watched a video that appeared to be reporter Christina Hager saying, “We are down several hundred positions across the MBTA.” In truth, Hager never said those words. It was Massachusetts Governor Maura Healey in another video.
The WBZ survey results showed more than half of people who watched the video could not spot the fake, but it was nearly split down the middle. While 27 believed it was real, 23 people knew it was not. The I-Team also ran a social media poll using an AI-generated voice clip and again found about half surveyed could not tell which was fake.
What about the other half? 
“What we’re really concerned about is people getting misinformation and, with that misinformation, making decisions about who they’re going to vote for,” said Massachusetts state Senator Barry Finegold. He proposed a bill that would make Massachusetts join a growing number of states drafting new laws to keep up with deep fakes. 
“If they are using this technology, they have to disclose and put a disclaimer on it,” said Finegold.
Back at Northeastern, Bau urges skepticism around election messages. 
“It’s going to take some adjustment time,” he told WBZ. “We’ve had this issue for a while in other types of media. You don’t necessarily believe anything that was written down on paper.”
CBS News has a special team of forensic journalists called CBS News Confirmed. It’s dedicated to fact-checking questionable videos and reports and to share their findings with viewers and readers.

en_USEnglish