Deepfakes fool more than half of Americans, UVU study shows

SALT LAKE CITY — More than half of Americans who encounter an artificial intelligence deepfake on social media don’t realize they’re looking at a deepfake, according to a new study by Utah Valley University.

More than half, 56%, of the over 200 test subjects from around the US couldn’t tell the difference between deepfakes and real content. That’s something Senior Project Analyst Hope Fager says was a surprise.

“One of the questions we’ve been asking is when deepfakes are going to get good enough that they’re actually convincing,” Fager said “That day is today.”

Fager said the speed at which deepfakes can be produced, and the ease, is a concern.

“I created this deepfake in a weekend on my own personal laptop with software I found online for free,” Fager said.

Related: What is real? AI, elections, and deepfakes

“The question is no longer whether deepfakes can get that good. It is a question of who is capable of creating deepfakes that are that good,” Fager said.

Knowing when something is fake

Another outcome from the UVU study on deepfakes illuminates how the speed of information dissemination and consumption impacts a deepfake’s believability.

In other words, the study found that when people know they’re looking at a deepfake, their subconscious picks up that something is off. But if a viewer encounters a deepfake “in the wild,” or unknowingly, the viewer can believe it is true.

Related: FCC votes to outlaw scam robocalls that use AI-generated voices

“In the current state of affairs, the buck stops with the end user, with the voter,” said Brandon Amacher, an instructor the UVU Center for National Security Studies. “And so all of us as voters and as consumers of media, we need to try to proceed in a heightened sense of awareness and take a couple of extra steps to verify things that we see that may influence our opinions on an election.”

Who is fooling whom

When you think of an artificial intelligence deepfake, you may picture political figures like Vice President Kamala Harris or former President Donald Trump as the deepfake target. And it’s true, they have been targeted this election year.

So was Utah governor Spencer Cox, who had to tackle this deepfake that was spreading misinformation about his gubernatorial race earlier in 2024. 

The following is a deepfake created by UVU students. The subject was former Utah Governor Gary Herbert. (The deepfake audio begins at 51 seconds into the video. It demonstrates the manipulation of audio to imply that the former governor was throwing his hat into the 2024 presidential race.)

 

UVU Researchers say they created that in just a few minutes with free software on the internet.

Boiling deepfakes down for smaller audiences

While deepfakes can target state politicians, they can be even more local than that.

Utah County Commissioner Amelia Powers Gardner said some deepfake creators are interested in smaller targets. 

“They’ll make a deepfake of a fake local reporter reporting news,” Powers Gardner said.

While disinformation, election interference, and fake news have all deepfake targets, the problem goes beyond politics.  Bad actors can use deepfakes for blackmail, bullying, harassment, and nonconsensual pornography. They are also used in financial fraud and scams, according to a report from the US Department of Homeland Security.

How to fight deepfakes, and what’s next in the fight

Utah County Commissioner Amelia Powers Gardner said knowing what to look for helps fight deepfake manipulation. That’s especially true if the question involves a local candidate or local election. 

“This web gets deeper and harder. So my advice as an elected administrator would be: go to the source. Go to your county clerk’s website,” Powers Gardner said.

Election Specialist for the office of the Utah Lieutenant Governor, Hayden Loftus, has a prediction. He said the big question lawmakers will be asking ahead of the next election is how to prosecute deepfake videos.

“The people that are making the deepfakes are really good at it. And they’re really good at hiding themselves as well. I would imagine that in the next four years, we see some kind of law being worked on or produced that tries to crack down on the mis- and disinformation … coming out of these deepfakes,” Loftus said.

Sam Herera and Simone Seikaly contributed.