Discerning the deepfakes

MARY REICHARD, HOST: It’s Thursday the 22nd of February, 2024. Thank you for listening to WORLD Radio. Good morning, I’m Mary Reichard.

MYRNA BROWN, HOST: And I’m Myrna Brown.

First up on The World and Everything in It. Discerning the deepfakes.

Deepfake technology is artificial intelligence that digitally alters someone’s appearance, voice, or background to make them appear… to be doing or saying something they didn’t.

It used to be easy to spot fake videos of public figures, but new data shows that so-called “deepfakes” have gotten better at fooling people.

REICHARD: A recent campaign in New Hampshire raises questions about how AI-generated content might influence voters in an election year.

So how can Christians meet this new tech with discernment?

Here’s WORLD Radio’s Mary Muncy.

MARY MUNCY: Before the New Hampshire primary in January, some registered Democratic voters received a call from President Joe Biden, supposedly.

Sound from Boston 25 News.

ROBO CALL/BIDEN: It’s important that you save your vote for the November election. We’ll need your help in electing Democrats up and down the ticket. Voting this Tuesday only enables the Republicans in their quest to elect Donald Trump again.

The New Hampshire Department of Justice traced the source of possibly tens of thousands of these calls to two Texas-based telecommunications companies and an individual. They supposedly used artificial intelligence to generate a recording that sounds like President Biden. The state filed cease-and-desist orders saying the calls were attempts to suppress voters.

It’s unclear whether the calls affected voter turnout, since more voters came to the polls than the state expected.

But this isn’t the only incident.

THE TELEGRAPH: [ZELENSKYY SURRENDERING]

Last year, a deepfake of Ukrainian President Volodymyr Zelenskyy surrendering to Russia came out while the countries were in peace talks, and a video of Elon Musk supposedly confessing to using drugs went viral

AI SENTINAL, ELON MUSK: I just dropped like 150 milligram edible.

Then a few weeks ago, a Hong Kong finance worker was tricked out of millions by fraudsters using deepfake technology.

It’s been used to create everything from filters on social media, to fake videos of election candidates, to pornography using a real person’s likeness.

The tools are ubiquitous, and that makes it hard to regulate.

Earlier this month, the Federal Communications Commission made AI-generated voices in robo-calls illegal. And several states have passed or are considering laws requiring labels for AI-generated videos.

Tech and social media companies are also trying to come up with their own internal regulations.

Two weeks ago, Facebook and Instagram owner Meta said it would promote standards across the social media industry for identifying and labeling deepfakes. Those standards include marking any AI-generated content with an embedded watermark or marker.

But that’s currently only possible for Meta if the AI-generated content is created on its platform.

Meta President Nick Clegg told PBS News Hour:

NICK CLEGG: The technology doesn’t yet exist across the industry to apply the same kind of common standards on audio or video content.

So as the U.S. general election gets closer, voters can expect to see at least a few deepfakes of their favorite candidate. But if technology and regulation can’t weed out fakes, will voters be able to catch them themselves?

MARKUS APPEL: We presented deepfakes and around 50 percent took notice of it and reported it.

Markus Appel is professor of the Psychology of Communication at the University of Wurzburg. He did a study on how well people can identify deepfakes.

MARKUS APPEL: We showed deep fake videos, and then we ask people, “What went through your mind when you watch these videos? ” And we were thinking that when people identify a fake video, they will like, notice it right away and also let us know about it.

But many didn’t.

The way the researchers asked the questions is important. “What did you think about these videos” not “was that a deepfake?”

It’s closer to the way people process media in their everyday lives. Generally, no one is asking a person scrolling through Instagram whether the video they just liked and shared was a deepfake. The person scrolling has to ask that question themselves.

APPEL: So the results were that people who are inclined to process information more thoroughly, like as a kind of general tendency, they were better at identifying a deepfake video to be a deepfake video.

That means someone could improve his or her chances of spotting a deepfake by creating habits of verification and discernment.

Some things to look for in potential deepfakes include unnatural human movement, lighting that doesn’t match the setting, and background noise that doesn’t match the background. Some apps and sites can even identify them for you. But as deepfakes improve, even subtle imperfections could be resolved.

Jason Thacker is an Assistant Professor of Philosophy and Ethics at Boyce College.

THACKER: We're kind of entering into a phase where just because you saw it online doesn't make it true. But also just because you saw it within a video or just because you heard it, or just because you have a photo of it doesn't actually make it true.

Not only that, but bad actors can also cast doubt on things that really happened.

THACKER: You have the ability to almost deny reality in that sense. And to use that for political gain.

Thacker says it’s essentially propaganda. And while the concept is not new, it is new for amateurs to be able to create convincing propaganda.

He says regulation is needed, but using AI to control or fix AI is an imperfect solution. Instead, slowing down and using discernment is how Christians can wisely navigate a digital age.

THACKER: We all really have to be part of this together. And that sense of pursuing wisdom together as we navigate a lot of this kind of post truth age that we live in.

Reporting for WORLD, I’m Mary Muncy.

WORLD Radio transcripts are created on a rush deadline. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of WORLD Radio programming is the audio record.