The Chaser Report - Biden Deepfake Inaccurately Speaks In Full Sentences

Episode Date: February 26, 2024

The American Presidential Election 2024 has already been plagued by instances of AI causing havoc and confusion. A totally real and not AI generated Charles and Dom bring you up to speed. Hosted on Acast. See acast.com/privacy for more information.

The Chaser Report is recorded on Gatigal Land. Striving for mediocrity in a world of excellence, this is The Chaser Report. Hello and welcome to the Chaser Report with Dom and Charles. Hello, Charles. I called it. For once, my prediction came true. AI has been used to disrupt the US election already. There's been fake robocalls of Joe Biden in New Hampshire already before the primary there. Apparently the robot voice was far more awake and actually knew it they were talking about. So it was pretty easy to pick.

But was it by his enemy or his friend? By his enemy. Here's what it sounded like. We know the value of voting Democratic on our votes count. 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. Wow.

Yeah, because I just assumed that if you inputted Joe Biden into an AI, it would come out with incoherent nonsense. Yeah, no, that was far more coherent and focused than I've heard Biden for a while. There was no meandering anecdote about someone who used to know in the 1940s. It was, that how it was spotted? I think so.

Yeah, I think absolutely, yeah. No one fell for it. Stuck out. Look, we've used AI to generate the rest of this episode. Let's have a listen after this. Hello, Charles. How are you? I really like working with you.

I am very well, Dom. This is. Such a great tool to sound like a tool. It really is. So, yeah, so it's already started. And you can just imagine how much more of this is going to happen. Deepfakes on election day when it's too late to debunk is my theory. And every phone in America is going to get, I don't know,

a video of Joe Biden personally welcoming migrants over the border or Donald Trump. I'm hoping the PP tape actually gets enacted by AI so we can all, you know, satisfy curiosity on that front or is that just me. He wants that. My prediction, though, is that this will. be the only election ever where AI is used convincingly to smear campaigns and campaigners and things like that because I think people will immediately grow very wary of it. And what AI will actually enable is when people are actually really caught up in something, like court saying

something absolutely horrific or, you know, doing something on video. Yeah. Incredibly embarrassing. it will be used as a way to defend themselves by going, oh, no, no, that was AI. So, for instance, the Donald Trump, the recording, I think was, was it Billy Bush or something before the 2016 election, the infamous, I come up to them and grab them, you know. Nowadays, they'd say, oh, that's not, that wasn't real, that was just AI. That was just AI.

And that's all you know, like, basically we're at the threshold of no accountability for any of your actions ever. Yep, that sounds amazing. Well, just, just illustrate how easy this. is to do, Charles. The audio was made, apparently, by a New Orleans magician, like someone who was working as a magician and therefore desperately needed money, he said, because he couldn't make money out of doing magic because he can't, you know, it doesn't actually work.

And he'd use a product called 11 Labs to create the fake Biden audio and a potential Mino Democratic candidate paid him to do that. And so the guy goes, there was no malicious intent. I didn't know how it was going to be distributed. Yeah, right. making a fake audio of Joe Biden telling people not to vote. How could that possibly be used maliciously? Well, are you sure that was the magician saying that or some AI saying? I don't know. It also says here it's a bit like some sort of Christopher Nolan film.

It is even more so because this is from New York magazine. This guy, Paul Carpenter, has world records in fork bending and straight jacket attempts. So that's the kind of, you don't exactly need to be a Silicon Valley boffin. He has no fixed address and he was paid $150 on Venmo and it cost him $1 to make the fake audio and it took him 20 minutes and so he later quoted as saying it's so scary that it's this easy to do people aren't ready for it and yeah it went to 25,000 registered voters look I think that that really is an uplifting tale of American ingenuity absolutely where somebody manages to get paid $120 for a minute That's all I heard.

All I heard was one minute of work, $120. I also heard that even if you're good enough at magic to hold a fork bending world record, you can still be of no fixed address in America. Why don't they value artists? Why does nobody value the artists, Charles? Yes, the lost art of bending for, you know. It's sort of like, because that was huge in the 80s, wasn't it? Oh, yeah.

Yuri Geller and all that, yeah. Yuri Geller, you know, he used to go on primetime television, you know, bending spoons. I think his thing was spoons, not fours. Boones, more than forks, yeah. Maybe that's why the guy that's got the fork bending world record. There's no one bothered because it's sharp.

But you know who's to blame for that, just as a sidebar, Charles. Who got rid of all the proper forks? Craig Rookastle. It's Craig's fault that we all have wooden forks. Disposable bamboo wooden forks. Thanks for nothing, Roocastle. Can't bend those.

So I've got a bit of an AI tale and how good AI is in some ways. So we're doing this show, Wankanomics at the moment. I'm so glad you mentioned it. Sort of completely sold out in Adelaide. We've added two extra shows, but they'll sell out by the time you listen to this. It's a huge triumph of the human spirit. There is one gag, though, that we just could not nail for the first sort of eight or nine shows that we did, right?

It's very simple, but it structurally needs to be there, right? So the way the show works is we pick on one audience member and we construct a whole. corporation for them, right, during the show. But one of the things we've got to do is we've got to come up with a brand name for this, you know, corporation that we're setting up, right? And we just couldn't think of a funny way to do it, right? Like, it just, you know, we riff off their name and sort of them, whatever, like. Oh, yeah, like Charles Corp or something.

We tried to sort of misspell it. You know how modern brands or misspelled? So you can get the website. So your one would be like, sure. Yeah, yeah, exactly. Choles.com. But it all felt, every different way that we tried, always fell a bit flat and it was sort of like, oh, that's, you know, that's an observation. Right.

So then a couple of nights ago, we went, why don't we just let AI solve this problem, right? What a good idea. And the thing is, in the show, we go, oh, well, obviously, this is 2024, you've sacked all your human stuff. you're going to need to use AI to find your brand name, right? And we sort of make it look like it's part of the show. It's actually because we couldn't work out what a good way to come up with the brand name is. And we just pretended that, you know, it was a joke, right? So anyway, and then what you can do with chat GPT is I've created a Wankanomics GPT engine, right?

Which I think we're going to put on our Wainconomics website. Yes, I've seen, I looked at this recently where you can actually train the AI to use particular input. rather than the whole of the internet. Yes, exactly. So, and to put it on guardrail, so you're getting exactly what you need out of, out of the engine, right? And so I just set up this little chatbot that we, you know,

like what we need is we need a sort of corporate jargonistic, we need a suggestion for a name based on somebody's own personal name for a large company. But then we want a jargonistic corporate, you know, set of reasonings around it. It's pretty snappy, right? It works so fucking well. Like, it is so, it is scary

because they're all tailored to each person's name. It looks like a magic tree. It looks like the modern version of spoonbending. And I was saying to James, who had had the show with, like, there is a very small window of time

where that particular thing will look impressive on stage, right? Absolutely. In 12 months' time, in 18 months' time, And everyone's going, what's that? Of course it's good. That's like using Google, you know. Well, we'll be using Google.

Yeah. But you can hear the audible gasps from the audience going, oh my God, how did they do that? We're in the window now. We're in this brief window where if you want amusingly mediocre results for anything, AI is just perfect. I used it to generate on radio not long ago. I used it to generate poems for different suburbs.

like a sea shanty about Mossman or something and you know it came up with a result in one second or a hip hop song about Eastwood or whatever it is like a rap battle and you know it does it very fast they're not good but they're amusingly not good enough to sort of read out on the radio

at this point but soon they'll just be oh that's actually quite good well done chat GPT long term listeners to this podcast though will remember that I once replaced you with chat GPT I do remember and And long-term listeners will also remember that that was the worst episode of the podcast ever. I've never felt so valued.

And that one listener even wrote in to say that I had to promise never to do it again because listening to Charles be outwitted by an AI was excruciating. Speaking excruciating, let's come back in a second. The Chaser Report, less news. That's great that you found a way to use it on stage before in a year or two, you know, Wankonomics is presented by, I don't know, AI avatars and you don't get any money because it's just trained itself on all of your jokes. Yeah, no, exactly.

So we've got to, we've got to use AI, but then also sue the fuck out of them for using all our writing as training data. 100%. It's a sort of pints a movement between using it and suing them. That's really funny. Can I just circle back, as you say? Yeah. Because there are some other ways in which AI is being used in the presidential election

as a bit of a trip down the rabbit hole. The first thing to note is that exactly what you said is already happening. So the window is closed. Donald Trump is using the claim that footage of... So you know the Lincoln Project, the Republican anti-Trump group that puts out quite effective video ads. They put an ad on Fox News that had basically a video, a compile of his gaffes. And he says that they're all AI generated, that they're not real. But they actually are.

Of course they're real, yeah. Yeah, but I remember seeing a few of them. But yeah, it's not true. It's not true. The fake news used artificial intelligence. But it's still earlier this month, after one of his lawsuits, he claimed that the fake news used AI to create an image of him that made him look fat. So that's a little bit awkward.

Yeah, right. Wow. So, yeah, so what he's claiming is that essentially every appearance that he has, where he's a bit fat, he's actually just AI-generated. Yeah, yeah, and he's really skinny. I'm going to use that one myself, actually. There's some bizarre ones here. So, there were images doing the rounds of Donald Trump aboard Jeffrey Epstein's plane.

And Mark Ruffalo, you know the Hulk from the Avengers movies? He shared them, and he had to apologize because they were fake. Wow. So many other things. In a super PAC ad for Ron DeSantis, they used a Trump AI to read Trump's own posts on Truths Social. They put Joe Biden in a bubble suit to make him look like he, but, you know, didn't want to do anything. Ron DeSantis had Donald Trump

passion, Dr. Anthony Fauci. So, yeah, pretty amazing stuff. But the one that's the most interesting kind of moral dilemma is, you know this new Google Gemini AI. There's an example being doing the rounds that Elon Musk sounded off about. When you ask it to do, like,

a variety of founding fathers, or you say, I think, a diverse range of founding fathers, it will produce images that look like the traditional American founding fathers. fathers in terms of the wigs and the outfits and so on, but have racial diversity, which, of course, they didn't have. So, of course, Elon Musk's using it as an example of the woke mind virus, because Gemini doesn't realize that diverse doesn't apply in the 1700s.

Doesn't it just suggest that Google used their Hamilton as part of their training data? Probably. It probably exactly what happened. There's another candidate who trained at chatbot based on all of his campaign materials to chat to voters and that got banned by OpenA.I. But that's not a bad idea. We could train a chatbot based on all of your podcast contributions. Yeah, that's a great idea. So rather than having whatever you did to generate me, it would actually be based on your own flaws in coherence. Yes, exactly. What we should do is we should have a second stream. So we have the Chaser Report

podcast. We have another feed, which is the AI generated chaser report, where we use the training data from all the years that we've done it and it just gives the sort of statistically most likely episode that day of what the Chaser report will be. Well, I mean, we could do it as a second feeder or we could just make it the main feed. It'd be much more cost-effective.

Not yet. Well, that's worth experimenting with. Or has it already happened? Yes. Oh, it is the real us. I like the idea that the reason why you can tell that it is actually the real us is because

in the training one, we would have already like in the AI version, we would have already referred to Bluetooth. Yes, that's true. Because on average, it's in every episode. That's right. Oh no, but now it is. So now it's indistinguishable.

Coming soon to you, the AI Powered Chaser Report. If you made it this far, tell us what you think. We should train it on podcast at chaser.com. Are you or if you've got any other brilliant time-saving and very slap-dash methods? Surely we should train it on the Joe Rogan experience,

the most popular podcast in the world. Yes, we should definitely make a fake. In fact, you know what? People already have. They already use a fake AI, Joe Rogan podcast. Because I remember looking for it on Apple Podcasts, and that was when it was exclusive to Spotify. And the only thing you could find was the AI one,

which was even worse. I don't think this is possible, but it's even worse than the real thing. Okay. What a wonderful world this is. Shall we stop there and get dabbling, get training the AI? Good luck with your show. Well, have a good day.

And, yeah, I suppose I'll see you tomorrow. Well, hopefully never speak to you again, because we'll just get AI to do it. Our gear is remote. We are part of the Aconiclass Network. Catch you tomorrow, or will we? Or will we?