Today, Explained - I see fake people

Episode Date: June 19, 2019

Doctored videos of Nancy Pelosi and Mark Zuckerberg have Congress worried about the nation’s grasp on reality. Drew Harwell from the Washington Post explains how “deepfakes” might corrupt upcoming elections. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Before we get to the show, a note about MailChimp's all-in-one marketing platform. It allows you to manage more of your marketing activities from one place, so you can market smarter and grow faster. It'll help you know who to talk to, what to say, when to say it, and the best channel to deliver the message. Who couldn't use that? Learn more at MailChimp.com. We want to give this president the opportunity to do something historic

for our country. This video of Nancy Pelosi made the rounds online a few weeks ago. While there are those in our family who think, why would you work with him if he, you know, go. The video had been doctored, slowed down. This is what the real video sounded like. We want to give this president the opportunity to do something historic for our country.

While there are those in our family who think, why would you work with him if he, you know, and basically he's saying back to me, why would I work with you if you're investigating me? But the fact is, something happened there. We've all seen and heard enough slow motion in our lives to know that if you slow down video, the audio drops down. Your voice lowers and deepens. But that didn't happen as it should have in the Pelosi video that was doctored. It was manipulated to make it seem like it hadn't been. To make Nancy Pelosi look sloppy drunk while talking about the president.

It was very, very, very strange. This video renewed concerns over fake videos as we head towards big, consequential elections. People were saying, oh my gosh, look at her. Like, can you just believe how she sounds? Like, this is crazy. Drew Harwell wrote about it for the Washington Post. By the time we went to Facebook to ask them about this,

the video had been shared almost a million times, had tens of thousands of comments and likes, and most of these comments were people who were totally believing what they were seeing. Even when people started to fact-check, Rudy Giuliani, President Trump's attorney, shared the video and said, wow, doesn't she look bad? Did anyone step in and try to stop this video from spreading? When we talked to YouTube about the Pelosi video, Did anyone step in and try to stop this video from spreading? video or a totally false caption. YouTube says they take those videos down. So the Pelosi video came offline on YouTube. But on Facebook, it was a totally different story. Facebook

asked their independent fact checker partners to rate the video and they ultimately rated it false. And they put a little sort of information box next to the video that gave a fact check, but they kept the video online. And then people went after Facebook, right? Yeah, totally. People were saying, well, maybe if the video was made of Mark Zuckerberg, the company would react differently. So people tested that hypothesis and they made a fake video of Mark Zuckerberg saying all sorts of crazy things. Imagine this for a second. One man with total control of billions of people's stolen data, all their secrets, their lives, their futures. Whoever controls the data controls the future.

The voice wasn't spot on, but the lip syncing was perfect. And it was a video of him like looking directly into the camera. So it was really just like a showcase of what people could make and what could happen if you really did want to target somebody specifically and kind of undermine them in that way. The Zuckerberg video was different from the Pelosi video, right? It wasn't just slowed down. Yeah, it was a deepfake. Deepfakes are computer-generated fake videos. They look like the real thing, but they are morphed in some way by the artificial intelligence into looking very deceivingly convincing and a lot like the real thing. Right, yeah, we talked about this last year in the context of fake celebrity porn videos circulating online,

but now there's this doctored video of the Speaker of the House circulating. Is the government getting involved? Not really. I mean, there have been some bills that have been proposed at the congressional level, but deepfakes are a tough problem to solve. And this was sort of brought out in a hearing of a House Intelligence Committee last week, actually, where, you know, a lot of lawmakers had these AI experts come to the stand and talk about deepfakes. I'm sure that members of the committee are familiar with their

run-ins with the media and know just how awkward it can be to have words put in your mouth that you didn't say. So deepfakes take this problem and potentially accelerate it. There's a fine line between a deepfake that's meant to discredit someone and a deepfake that's satirical or a joke. Did anyone at the hearing address that? Yeah, you know, deepfake videos do sort of fall into that gray area that we've always protected as a part of free speech and free expression. One of the classic articles about former Vice President Biden comes from The Onion, that he was waxing his Camaro in the driveway of the White House.

It was a comedy bit and it had manipulated content on it. If we went to that extreme, we would have a country where everything that's ever been changed or modified for any reason would have to be policed. And we'd be asking a private sector company to police that. You know, First Amendment, we've always seen as protecting parodies and protecting satire and political speech. But with deepfakes, is that protected political speech? I think some of them could be. And you could make the argument that, you know, the Mark Zuckerberg video may be political speech. It's making a broader point about people in power, right?

But where does it get into the kind of speech that you don't want to support? And some people have also gone to Facebook and said, you know, there's a difference between freedom of speech and what they call freedom of reach. The power that the tech platforms offer to videos and creators like this, where they're supporting their ability to go viral almost immediately. So I think there's going to be more talk on the lawmaker level about what they can do. Can they craft a piece of legislation that would not go too far in violating people's rights to free expression, but would also prevent active BS video campaigns from going viral online. I saw something from John Herman from the New York Times online that said, one of the real problems here might not just be the proliferation of fake video that is

indistinguishable from the real, but the fact that the existence of so much fake video could lead to the point where politicians or anyone else could just say like, oh, that video of me, that was fake when it was in fact real. Totally. And that's actually one of the bigger concerns around some of the researchers who look at this. You know, if you're captured on camera saying something or doing something you'd rather have not been recorded, you can say, oh, it was fake. And, you know, that is not, that's not an unprecedented problem, right? Because we saw with Donald Trump, after the Access Hollywood video, he was even saying, he was telling his

people around him, you know, this video was doctored, it was inauthentic, who are you going to believe? You know, that's the problem right there. Do you think all this confusion might just make people check out of the news? Yeah, so people call this reality apathy, where they say, like, if everything online is fake, and I can't believe anything, then screw it. I'm just not going to read the news. And so if we get to a point where we're not believing videos or we're just saying to hell with it, I think that's going to be a huge and very subtle change to how we encounter the problems we deal with and end up causing more harm than we're trying to avoid. And you can imagine a

situation in which reality apathy leads to serious voter apathy in, say, the next election. You know, I think videos are probably just going to be one part of the misinformation machine that will be churning during the 2020 campaign. We're getting very close to the vote. Maybe we will see these, you know, blockbuster deepfakes that are really

convincing and that, you know, may be fact-checked in due time and we may all realize that they're fake in enough time. But in the interim, there'll be enough to maybe convince people to vote the other way or maybe just convince people not to vote at all. I'm Deborah Johnson. I'm an emeritus professor at the University of Virginia in the School of Engineering. I worked with Nick Diakopoulos, who is a professor at Northwestern University, and together we came up with a set of scenarios that would make it clear the potential harm that can be done through deep fakes. Scenario one, supporting a candidate. So imagine that it's 2020. Seth Moulton is running in the primary,

and he is one of the only candidates with military experience, which a PAC wants to use to build up his campaign. They might develop a video that used both real footage and synthesized footage in which the candidate had been a hero. Some people would be claiming that it was fake. Other people wouldn't even know that it was being claimed to be fake, and it would just spread across the Internet. And even though he might have had a heroic record, it's now being called into question. Even though it might have been intended to help his campaign, it's actually hurting his campaign. Scenario two, hurting a candidate.

In this case, we can imagine Elizabeth Warren increasingly doing well, even among white men. One of her competitors in the primary, or if she were the candidate, the incumbent, fabricates a video which depicts her as making disparaging comments about white men. Even if her campaign could convince the major networks to spread the fact that this was a fake, a good deal of damage would be done just by the thing being out there and circulating. Scenario three, undermining the election process.

It's the night after the debate between the incumbent and a competitor. Let's say it's Biden. He did very well in the debate, and suddenly a synthesized video appears, which suggests that the person who was in charge of the debate is talking to a member of Biden's staff and asking something about whether the questions were okay or got there in time, with the implication that Biden had been given the questions before the debate. And even though there might be time for Biden to kind of counteract the claim, the whole debating process would have been undermined. The three scenarios just suggest really serious ways in which candidates can be hurt and the whole election process can be undermined so

people don't know what to believe. And this casts a doubt on, you know, the major way in which we decide what's true. It really threatens the integrity of our election system. After the break, can anything be done about deepfakes, or should we all just start prepping for the apocalypse? Do you have a business? Because if so, MailChimp would like to help you grow it. They've got an all-in-one marketing platform that allows you to manage more of your marketing activities from one place so you can market smarter, grow faster. The platform eliminates

the need for multiple tools by giving you everything you need to create, publish, manage, and measure your multi-channel marketing campaigns if If you have them, maybe you want them. This is the time. MailChimp's all-in-one platform is powered by a marketing CRM, which I recall means customer relationship management. So you can collect, organize, understand, and act on all of your audience data. So you'll know who to talk to what to say when to say it and the best channel to deliver the news the complete marketing platform has everything you

need to start marketing your business today explained and they'll be there to help you as your business grows and needs new capabilities let MailChimp help you market smarter learn more at MailChimp.com. Thank you. Drew, I know we're mostly talking about fake video here, but I'm thinking if it's getting easier to manipulate video, it's got to be way easier to manipulate audio, right? So fake audio is a big issue too. And these AI researchers are using some of the similar systems that they use with fake video to do audio matching

where you can put in a bunch of audio from an original speaker and train the AI system to be able to change that person's voice and say whatever you want. The voices are surprisingly convincing. Like, if you're just hearing it at a glance, you might be fooled. Yeah, I saw that these two Facebook engineers used AI to create audio clips of a bunch of famous people saying things they never said. And the Bill Gates ones were something. Two plus seven is less than 10.

Poured as a strong wine with a smoky taste. The glow deepened in the eyes of the sweet girl. I mean, it wasn't perfect, but it was sort of wild. And even fake text, like you can invent these whole fake news articles where you read it and it seems like a human wrote it, like the words make sense, it's convincing enough for potentially enough people. So all of these things are in the stew of misinformation and all of them are getting better way quicker than

anybody really expected. Are there any technical solutions, any ways people can detect fakes? There's actually a research team led by this guy, Hani Fareed, in California, where they'll take these really precise scans of, in this case, world leaders' faces. And so they'll have this computer model of how somebody like Barack Obama's facial features correlate with each other. You know, when Barack Obama smiles, he tends to tilt his head in this way. When he talks, his cheek moves this way and his upper lip raises this way.

And so then when they see a video later on, whether it's like a human impersonator trying to satirize them or whether it's like a total AI-generated deep fake, they can compare the two and say, these facial tics don't really line up. But the issue is that if you're not one of these world leaders who has had this intimate biometric model made of your face, you're out of luck. Is this video detection tool public? Can anyone use it? Yeah. Honey Freed's team has like made their

research public now. They've given up some details of what they're looking at, but the specific tool that they have made to like test videos, they're really protective of. They don't want to just put it online for everybody to test because then they feel like the deep fake creators will find ways to poke holes in it, right? So they want to maybe give it to media outlets and fact-checking groups to be able to use it as news video comes out. But they're really afraid of, like, putting it out there more widely or else it'll be kind of misused

to the point where it's almost useless. What happens when the deepfakes get so good that even a program like that can't identify them? Is that a possibility? I think it is a possibility. And some of the researchers think that's only maybe a couple years away till we get a video that is so realistic that it convinces most, if not all, people who watch it that it was the real thing. And, you know, this isn't just like a political thing. This could affect like

criminal justice. I mean, video that's submitted into evidence in the courtroom. What if somebody has a fake video made of them doing something they never did? That's going to be a huge burden on the police to prove it, the jury to understand it, let alone like the people who are being accused of the crime in the first place. But this isn't necessarily some wait-and-see thing that's coming in the future. We're already at a place where kids are following digitally created models on Instagram and don't even realize they're fake. We're at a point where people are sharing this Nancy Pelosi video and thinking that it's real and continue to believe that it's real

even after they're told that it's fake. I mean, it is really alarming that our own reality is already being questioned. I think it is. And these are like pretty big moments that people should be in agreement on, right? Like we all heard the Access Hollywood tape on our own, right? Like we all think we knew what happened there. And yet even there, there's disagreement.

And with the Pelosi video too, like we know what happened there. And yet even there, there's disagreement. And with the Pelosi video too, like we know it's fake. And yet the brain acts in interesting ways when we encounter even fake stuff. Like a fake video like the Pelosi video can get its message across even when you know it's fake. Like the message has been conveyed and it's important for people to be careful, but I almost think that that's not enough. Like we have to really engage with this idea that video at some point could be compromised and what we see with our eyes may not always be like something we can truthfully believe. When video first came around and images came around,

it was something like you were given a portal into this other world where what you were looking at was real. It was a part of society somewhere else. You know, videos from Tiananmen Square, videos from the Vietnam War, these were like pivotal moments in how we understood the things around us and they helped push the world to be a little bit different too. And so if we get to a point

where even that like window into the world is just another tool to be used by people who want to deceive you, that's a huge problem. And I'm worried we're going to be there sooner than we'd like. Drew Harwell is a self-proclaimed Washington Post human covering AI and the algorithms changing our lives. I'm Sean Ramos-Verm, a Vox human covering the news every day, but not on the weekends.

The rest of the Today Explained humans are Irene Noguchi, Afim Shapiro, Bridget McCarthy, Noam Hassenfeld, Amina Alsadi, and Halima Shah. We've got two human interns right now, Will Reed and Alex Pena. The mysterious Breakmaster Cylinder is obviously not a human. And by the by, Today Explained is produced in association with Stitcher or part of the Vox Media Podcast Network. Thanks to MailChimp for supporting the show today.

Once again, their all-in-one marketing platform will really revolutionize your life because it will tell you who to talk to, what to say, when to say it, and the best channel to deliver the message. I mean, this is about marketing, not about like your social activities, but who knows, maybe it'll teach you a thing or two about that too. Learn more at MailChimp.com.