With AI becoming more and more available online, and easier to use, the lines between real and fake are getting harder to discern.
But could they potentially influence the 2024 presidential election?
Alex Mahadevan, the director of MediaWise, Poynter Institute’s digital media literacy project, speaks with WUSF's Matthew Peddie, host of "Florida Matters" and the podcast "Our Changing State," about how to spot deepfakes and cheapfakes.
Mahadevan described a deepfake as a "fake video that's been created with artificial intelligence."
And as the technology behind artificial intelligence continues to advance, Mahadevan
said it's becoming increasing more difficult to spot a deepfake video.
"Originally, back in the day, it was when someone would use lots and lots of photos and video of someone like Barack Obama to paste his face on someone like Jordan Peele's face, and make it look like Obama was saying something he never did," Mahadevan said.
Peddie and Mahadevan discussed some examples cited by The Washington Post, including a digitally altered deepfake of Eminem endorsing a candidate in South Africa, and a robocall audio deepfake that appeared to be President Joe Biden telling voters not to go to the polls in New Hampshire during this year's primary election.
Audio deepfakes can be tricky to spot
Mahadevan said spotting an audio deepfake is especially difficult because they're easier to make.
"Even though videos have gotten a lot easier, audio deepfakes are what we've seen in global elections pop up all over the place," Mahadevan said. "So hearing that example kind of hit me like, 'Hey, we're seeing this in the US now.' And also the story behind it was fascinating. A political operative paid a magician who lived in Texas to create this audio deepfake that went live in New Hampshire."
Right now, Mahadevan said there's not enough data to show whether these more sophisticated deepfakes will have an impact on this year's election, since they're more flashy and tend to get debunked more frequently due to the heightened media attention.
"I think what worries me the most is that it has given any regular person the power to be a extremely high-functioning political operative in that they can create lots and lots of political graphics."Alex Mahadevan, MediaWise director
However, he said "cheapfakes" — including cheaply edited videos or out-of-context images — might be more deceiving.
"So the last year and a half or so, since Chat GPT came on the scene, they really put the power of AI into people's hands, anyone's hands," Mahadevan said. "And so it's getting easier to make synthetic video deepfakes, but it's still a lot easier to take a video of Joe Biden giving a thumbs up to a parachuter on the ground, and cut the video so it looks like he's giving thumbs up to nobody. It's really easy to end a video that makes it look like Joe Biden is about to sit in a chair that isn't there, when the context is totally different.
"So the bar is still much lower to create cheapfakes, really easily edited videos, and that's why I think you don't see as big of examples as deepfakes, is because all these online trolls, they're just going to take a clip of Joe Biden and cut two seconds of it and put it on Twitter, and then it's going to go viral and it gets reported in The Daily Caller and on Fox News."
Keeping up with the advancing technology
And even though the heightened awareness might eventually make it easier to recognize a video or image that could be a deepfake and cheapfake, Mahadevan said he doesn't think "there will ever be any mechanism to stop the creation and spread of disinformation."
"Think about the different tools that are used to create AI," Mahadevan said. "So Chat GPT has put in guardrails, so I can't put in, 'give me a picture of Joe Biden spilling ice cream on a shirt.' I can't do that. It won't let me. But there are plenty of people who figured out how to add a lot of different words and change around the prompt. So it does do that. And there are open source tools that will always skirt those guardrails."
Mahadevan warned that although the outcome of the Kamala Harris-Donald Trump election might not be determined by deepfakes, it's going to become more and more difficult to distinguish in some cases between what's real and what's fake — especially with how easy it's becoming for ordinary people with limited technical skills to create this wealth of information.
"I think what worries me the most is that it has given any regular person the power to be a extremely high-functioning political operative in that they can create lots and lots of political graphics," Mahadevan said. "They can create tons of political copy. They can generate 5,000 Facebook comments in a matter of seconds if they set up the right system using AI.
"So while AI may not necessarily swing an election with, like, a last-minute deep fake of Trump, what you might see is just a flooding the zone of AI-generated images and text aimed at trying to influence people, made by regular people, targeting regular people."
It's a trend, Mahadevan said, that's only going to continue to get worse.
"What really worries me the most is it's just getting harder and harder to trust what you see online, because there's so much synthetic content," Mahadevan said. "It doesn't matter how good it is. And like I said, it's not going to be this ground-shaking deepfake of Joe Biden or Donald Trump. What it is is kind of this slow march toward non-reality that's filled with synthetic AI slime."
