Can US Congress control the abuse of AI in the 2024 election?

This is The Guardian.

Back in January, as I was travelling around New Hampshire talking to residents about how they were likely to vote in the primary election there, I heard a bizarre story. Voters in New Hampshire received a robocall right before they headed to the polls. When they picked up the phone, voters heard what they thought was President Joe Biden telling them not to vote in that Tuesday's election. It's important that you save your vote for the November election. Voting this Tuesday only enables the Republicans in their quest to elect Donald Trump again.

Now that sounds like a plea from Joe Biden, except that it isn't. And now it's under investigation by the New Hampshire Attorney General.

In response to this scam, just a couple of weeks later, the US government outlawed robocalls that use voices generated by artificial intelligence. But experts are warning that this story is just one example why 2024 will be a year of unprecedented election disinformation in the US and all around the world. So why are people so worried about the influence of AI on November's presidential election? And what can the politicians do to catch up? I'm Jonathan Friedland, columnist for The Guardian, and this is Politics Weekly America.

I made my family all do it. We have a various mix of people who work in politics, who are very politically savvy, and then some who, just like, kind of lightly pay attention to the news. And the results were all over the place. It's hard to tell what's going on once you really start thinking about it. You really have to pay attention to a lot of context clues.

OK, well, I think we should plunge right in with clip number one. So let's listen to that.

She stroked his pretty skin again and kissed him and held him tight. But instead of saying, thank you, that snake gave her a vicious bite.

I think I've got a sort of privileged access to this one because I've actually heard Donald Trump, indeed in New Hampshire, doing the snake. It's a bit of his stump speech. now. I agree. If I hadn't heard it myself, I would assume it's so bizarre that it would have to be faked.

That's correct.

OK, so let's go on to clip number two. No, you're a lying dog face pony soldier.

You said you were, but now you got to be honest. I'm going to be honest with you.

Lying dog face pony soldier. I mean, you'd think that was made up. But weirdly, it's the sort of thing you can imagine Joe Biden saying. because it's so random. I'm going to say that's real.

It is real.

All right. So two out of two so far. Now let's go on to clip number three.

I fixed the economy.

We were all rich.

And then they came after me.

Oh, Donald Trump definitely heard him express those sentiments, that laugh at the end. I'm not so sure that's real. He so rarely laughs. I'm going to take a risk and say that's fake.

Yes, that is fake.

Amazing. Amazing, that that's fake because there's nothing in it that would make you think. I mean, it sounds absolutely like him. So the technology can make his voice say that. Journalist, I don't think he would ever call that out as being or even be suspicious if someone passed that off.

No, I mean, it's. it's definitely what he says. You know, it's just. he didn't say that specifically.

Yeah. Let's go on to clip number four.

As my football coach used to say, lots of luck in your senior year.

So, Joe Biden, as my football coach, you say lots of luck in your senior year. I'm going to say that's fake.

That one is real.

Ah, and in a way, what would you have to go on? Let's go to number five now.

That one is fake.

OK. And again, my only guide, really nothing glitchy in the voice or anything, was. I think if he'd said it, we would know about it. It would be one of those memorable clips that's played again.

What an incredible score.

And I've got to say, the ones there that were fake, even when I guessed them right, there was nothing in it that would give it away.

And I think you exhibited some of what starts to happen. is you really just start to question what is real. Your brain kind of goes a little buggy. These are things that plausibly someone could have said. It sounds just like them.

Completely. So this is audio. What about video? If we were doing this test with video, is the video AI technology at a similarly advanced stage?

Not from the free or like pretty inexpensive tools. Audio is more advanced, just in that you kind of lose that video dimension. You lose one of your senses. And for that reason, it's more plausibly real when it's fake. And when I tried to create video clips on the same AI app that I used to create the audio ones, they just look so obviously fake.

That's, I mean, itself, extraordinary. This is a few years ago, it would have cost millions to do what you've just done with a $40 app. There is a practical reason why you and I just had fun with this quiz. And that was the story in New Hampshire, which I referred to, just before. Voters getting a call from a voice AI generated, which sounded just like Joe Biden, telling them not to vote, a so-called robo call.

We do know now that it was someone who had worked and like a strategist in an affiliated group with Dean Phillips, the U.

And you mentioned Dean Phillips, who finally suspended his campaign the other day. Did he know about this?

He said he did not, and he condemned it. And it was not his campaign. It was a separate group that had been working on getting voter messages out. So he did not have any knowledge of it, he has said.

And for the average voter, hearing that robo call after you've been told it's fake, I think the brain adjusts and then says, OK, well, now that I know that, yeah, you can hear the things that are wrong. And you can imagine that in the busy household, you know, the kids are making a noise, the kettle is boiling, whatever it is. You're not just analyzing the voice, even like I was during your quiz there. People would have just passed that. Are we seeing more and more of these kind of scams in just regular American life before we get onto politics?

For sure, in that in that dimension, it's far more common and advanced than in U.

The bad guys can now clone a voice using only a couple of seconds worth of actual sound.

You know, these scammers can kind of clone the voice of someone's grandchild or child and call them and be like, I need money, I need your help. You need to wire money right away. And it's emotionally alarming. And there is a worry that that element of it being able to clone like my voice to my grandparent could be used in U.S. politics, because then I'm a more trusted messenger to a grandparent than, you know, any politician would be.

Oh, wow. So the grandparent scam could become a political device where your grandparent hears you and saying, you know, vote Donald Trump. That's what I'm doing, grandma. And they might fall for it.

That would be the most extreme worry and how that technology could be used. And there's no, you know, reason to say it couldn't be. Aside from that, it's just time consuming.

Yeah. And the general, we should say this just as a bit of public service, the advice people always give about these things is if ever you do get a call, especially if it's somebody asking you for money, you should just call them back on their own number. And then, you know, then you're not back because they won't phone. you, call the grandchild or whatever and say, was that really you just now?

Yeah, this is an incredible election year. And that alone actually means there is more, you know, potential for mayhem just because of the sheer numbers, but then also because our attention gets really dispersed. You know, there's only so many people who can keep an eye on things in this many places. So in Slovakia, there was actually a fake recording created by AI, where one of the candidates had boasted that he'd rigged the election. And the belief is that that fake call swayed the election to a pro-Russia candidate in that country.

And when you say pro-Russia, that does point at least a sort of hint of a finger at who may have been messing with the AI. And people did say that about Russia after 2016, that, you know, they were just at the very beginnings of the technology that would soon be available. And here we are, eight years later. This has got politicians, lawmakers in the United States, sufficiently worried that they have moved to ban the practice. What exactly have they done?

So the Federal Communications Commission, the FCC, which kind of regulates phones and other technologies, they have banned it. But there aren't regulations yet on the Federal Election Commission side on what can be used in campaign ads or the kind of like TV ads that we would see, whether those use AI, that there's not anything on that right now. And, you know, we're in the midst of the election. So we're already behind on this issue. The other thing I'll note is that with FCC rules, there's already a do not call list where, you know, you can sign up and say people who are trying to sell something shouldn't call me.

Yeah, I think this is such a good point. If you're dealing with people who, anyway, are operating dishonestly, a change in the law might not help. Let's talk about 2024 specifically. What other interventions or deployments of AI have you spotted already in this election that could mess with how people see reality?

There are videos where, I mean, the RNC had done a video kind of showing this is what Biden's America looks like, sort of a dystopian kind of vibe to it, right? And it's very clearly AI. Everything just looks a little too shiny is what I would say. Like. people kind of get really smooth looking like their skin is very smooth and their hands and everything.

Yeah, that's what a lot of the AI campaigners are demanding. is just labeling, truth in labeling on all kinds of products, so that you at least know. What about the image of Donald Trump with his African-American friends?

Right, yeah, that was a classic example, a very recent example of, you know, not affiliated with the Trump campaign, but people pushing for, you know, black voters to go for Trump have just put him into photos with black people, basically to say that he is, you know, a candidate who is working for black voters, that kind of thing. And they're not real photos, but they look pretty real if you were to just kind of glance at them.

The only thing there would be. the clue would be perhaps the political context, although we've been reporting on the podcast before, and we're going to get into it in future weeks, there is a growing body of African-American votes that is moving to Donald Trump. And so, in a way, it's not a wholly implausible reality, but there we are faking the imagery. So the politicians are trying to move against this. as so often with tech, they are miles behind, aren't they?

Congress just started a task force that is looking at this issue, and hopefully we'll have some sort of recommendations on what to do with it. Again, we're sitting here in March, so it could have been done previously. We all knew this was coming. It's not like. this is something that's surprising.

And people have been quite struck that this is one of those rare, and they're becoming increasingly rare, areas of political life where it's quite bipartisan, Democrats and Republicans working together. By way of illustration, I feel that for the second time in quite just a few weeks, we need to mention Taylor Swift on the podcast. Tell us about how she plays into this.

Yeah, there were AI-generated images of Taylor Swift nude. And, you know, there's a whole subset of the, you know, the bad Internet is what I always call it, that does a lot of the fake pornography. And they will use, you know, publicly sourced images, imagery of famous people, but also sometimes just everyday people, and make them look as if they took nude photos. I mean, imagine that that was you, right? That is so violating and fake.

In fact, we should direct listeners to a whole episode in our new Guardian miniseries, Black Box, which zeroes in on exactly this phenomenon presented by my colleague, Michael Safi, so people can get more on all of this. But in terms of the threat to democracy rather than just, you know, personal privacy, would it be fair to say that Congress and politicians haven't fully woken up to the scale of this as a threat, don't actually take it ultimately that seriously?

I think that's fair to say. And I will say, too, like I was skeptical that the technology was far enough along that it could meaningfully trick anybody, until I started playing around with the audio tools and realized just how easy it is. And then, really, the deepfake call in New Hampshire to me was a moment, an eye-opening moment, just to realize that this will be a feature of this election and in future elections, you know, and especially as we go from a national level, the more local that we get, the more concerned I am about it, because there are just fewer journalists paying attention. You know, if there's a national fake photo or fake call, you think of how quickly it became debunked, right? You think of how quickly in New Hampshire we were able to see that it was fake.

Yeah, I mean, absolutely. If you're faking something about somebody running for the local school board, who's going to be around to call that out? And before you know it, it's spread and everyone believes it. We talk about how politicians can be quite slow. The tech companies themselves, usually they can be much faster and they are promising.

Google's chatbot is doing similar to what others have done, which is to say that if there are political questions asked, election related questions, it won't answer it or will direct people to more trusted sources. Other chatbots have directed to voter registration sites, Secretary of State. Basically, they'll just say, we can't answer that right now and direct you to somewhere that can. I think that's a good idea. A lot of times.

All of this, like I mentioned, is pursued in the serious black box that my colleague Michael Safi is fronting, including this question of how people are going to be able to tell fact from fiction. What's your advice to individuals about when, you know, when they are confronted with, crucially, they won't know if it's AI they're looking at. But just in terms of watching things, listening things, seeing things, what tools or pointers could you give for how they could protect themselves against this kind of abuse, this darker side of AI?

I think the first thing is remember what you trust in terms of sources. So if you have a news organization you trust, you know, a think tank, what have you, who do you go to when you're trying to figure out something is real? And when you see something that feels either it's too good to be true, right, like it sort of makes someone look better than you thought, figure out what the trusted sources are saying about it. And that might take a little bit of time, but don't jump to any conclusions while you're waiting to see, you know, what researchers or journalists are learning about something that seems, you know, questionable. The other thing I'll say is what I've noticed when I was creating the fake audio is that people sound too good in it.

Yeah, it's so interesting what you're saying about limbs and hands, because that fuss that was kicked up particularly over here about the picture with Kate Middleton, Princess Catherine and her children, released for Mother's Day. A whole lot of people were looking at the fingers, saying they just don't look right. You know, it's alleged that perhaps there was some AI manipulation there. Kate Middleton herself, the princess, admitting that she had definitely sort of tweaked the picture. But lots of people had a sort of hawk eye for those fingers and thumbs and hands because of exactly what you said about AI.

It's unclear what happens in the Senate. There are some senators who've expressed that they have concerns with this over free speech, First Amendment concerns, which makes sense. This is a platform that people definitely use for speech. They use it for all kinds of speech, for political speech, and just for fun. And a whole ban on it, it just seems so harsh to some people.

And Republicans in the Senate, I mean, they're not a particularly young group. I mean, are they going to be worried about alienating younger voters by doing this? Or do they think, yeah, it's important to strike a blow at the Chinese government?

I would say the latter. I don't think that their base of voters really uses TikTok. It's much more people on the left who are using it. Than people on the right. So they definitely, their base is more interested in Chinese influence over the U.S.

So spare a thought for Joe Biden, who is the latest in the TikTok generation to get on the platform. We will now have to hastily get off it. Rachel Leungang, thank you so much for talking to us for Politics Weekly, America.

Thank you. It was great to be here.

And that is all from me for this week. After Tuesday's primary contests in Georgia, Mississippi and Washington state clinched once and for all the party nominations for both Joe Biden and Donald Trump, our colleague Lauren Gambino spoke with the Guardian Australia podcast Full Story to discuss the presidential rematch. no one wants. So search for that wherever you get your podcasts. And, like I've mentioned, you really have to check out the Guardian's new miniseries, Black Box.