Why candidate deepfakes are so hard to debunk

HostWe have all seen those clips online where a famous face says something totally wild that they would never actually say. Usually it's just for a laugh, but when a big vote is coming up, these fakes can turn into a massive headache for everyone involved. What's it about these clips that makes them so hard to shut down before people head to the polls?

GuestThe biggest thing is time. In the world of news and social media, a lie can reach millions of people in a few minutes, but a solid proof that it's a lie takes hours or even days to put together. Think about the days right before an election. If a video drops on a Friday night and the vote is on Sunday, there's a tiny window to react. By the time a news team or a tech company can prove the video is a fake, the damage is already done. People have already seen it, felt angry about it, and maybe even shared it with all their friends. You're trying to put out a fire with a cup of water while someone else is using a hose to spray gas on it.

HostBut we have all these smart tools now. If I see a video that looks fishy, why can I not just run it through a check to see if a computer made it?

GuestIt sounds like it should be that simple, but the tools we use to catch fakes are always one step behind the tools used to make them. It's a cat and mouse game. Every time we find a way to spot a fake, like looking for weird blinking or strange skin tones, the people making the fakes find a way to fix that. Also, these detection tools aren't perfect. Sometimes they say a real video is fake, which is a huge problem. If a news site uses a tool that gets it wrong even once, they lose all their trust. So they have to be very careful and move slowly, which plays right into the hands of the people spreading the fake stuff.

HostI would think video is the main worry because it looks so real, but I have heard that sound is actually a bigger mess to deal with. Is it really harder to debunk a voice than a face?

GuestActually, audio is much more dangerous. With a video, you have thousands of tiny details to get right, like how shadows move or how hair flows. But with a voice, a computer only needs a few minutes of real talking to build a perfect clone. Think about how many times you have heard a candidate speak on the news. There's plenty of data out there to train a bot. And when we listen to a clip, we don't have as many clues to look for. If the voice sounds a bit flat, you might just think they're tired or the phone line is bad. In a recent vote in Europe, a fake sound clip of a leader talking about rigging the vote came out just two days before the polls opened. It was just two people talking, no video at all, and it was almost impossible to prove it was a fake in time to matter.

HostThat sounds like a nightmare for the person running. But if they just come out and say it's a fake, does that not solve the problem?

GuestYou would think so, but there's a side effect to all of this that we call the liar’s dividend. Because everyone knows deepfakes exist, a candidate who gets caught doing something bad in a real video can just point at it and say it's a fake. It creates this fog where nobody knows what to believe. If everything could be a lie, then nothing feels like the truth. It gives a free pass to people who actually did something wrong. They can just blame the tech. So, the fake videos don't just spread lies, they also make it harder for the truth to land.

HostWait, so you're saying the existence of fakes makes real evidence less powerful? That feels like a huge shift from how things used to work.

GuestIt's a total shift. In the past, if you had a tape of someone saying something bad, that was the end of the story. Now, the tape is just the start of a long fight about whether the tape itself is real. And this brings us to another wall, which is how our own brains work. We tend to believe things that fit what we already think. If you already dislike a candidate and a video comes out of them saying something mean, you're way more likely to believe it's real without checking. Even if a group of experts says it's a fake later, that first hit of anger stays with you. You might forget the fact-check, but you'll remember the feeling that the person is bad.

HostIt feels like the tech companies should be doing more to stop this at the source. If they see a fake, why not just pull it down right away?

GuestThey try, but they're stuck between a rock and a hard place. If they pull something down too fast and it turns out to be a real video of a politician, they get accused of messing with the election or hiding the truth. If they leave it up and it's a fake, they get blamed for spreading lies. Most of these companies have rules about AI, but the sheer volume of stuff being posted every second is more than any team of humans can handle. They rely on bots to flag things, and as we talked about, those bots aren't always right. It creates a lag. And in the final hours of a vote, a three-hour lag is basically an eternity.

HostSo even if the tech gets better at spotting the fakes, the human side of the loop is still the weak link.

GuestThe real trick is that these fakes don't need to be perfect to work. They just need to be good enough to make people doubt what they see for a day or two.

HostA voter sitting at a kitchen table with a phone is now the final judge of what's real, long before any expert can weigh in.

GuestThat person is the one who has to decide if the voice they hear is a real person or just a very good ghost in the machine.

HostThe old rule about seeing being believing has never felt quite so shaky as it does right now.