WEBVTT - The Story: Liar, Liar, Deepfakes on Fire w/ Hany Farid 0:00:00.040 --> 0:00:02.640 Thanks for tuney into tech Stuff. If you don't recognize 0:00:02.680 --> 0:00:05.440 my voice, my name is Osva Looshan, and I'm here 0:00:05.480 --> 0:00:08.600 because the inimitable Jonathan Strickland has passed the baton to 0:00:08.720 --> 0:00:12.160 Cara Price and myself to host Tech Stuff. The show 0:00:12.160 --> 0:00:15.040 will remain your home for all things tech, and all 0:00:15.080 --> 0:00:20.400 the old episodes will remain available in this feed. Welcome 0:00:20.400 --> 0:00:24.200 to tech Stuff. This is the story. Every Wednesday, we 0:00:24.280 --> 0:00:26.880 bring you an in depth interview with someone at the 0:00:26.920 --> 0:00:30.600 forefront of technology or someone who can unknock a world 0:00:30.640 --> 0:00:34.000 where tech is that it's most fascinating. This week it's 0:00:34.040 --> 0:00:40.640 Harney for Reid. He's a professor of electrical engineering and 0:00:40.680 --> 0:00:45.040 computer science at the University of California, Berkeley, with a 0:00:45.120 --> 0:00:51.640 CSI sounding specialization digital forensics. His focus is on image 0:00:51.640 --> 0:00:55.480 analysis and human perception, so he's the guy you call 0:00:55.920 --> 0:00:58.000 when you need to know whether or not you're confronting 0:00:58.040 --> 0:01:01.720 at deep fake, and many do. He's constantly talking to 0:01:01.840 --> 0:01:05.760 journalists to help them determine what's real and what's fake online. 0:01:06.440 --> 0:01:08.959 In his lab at UC Berkeley, he and his students 0:01:09.000 --> 0:01:13.319 study the various ways misinformation is created and spread and 0:01:13.360 --> 0:01:15.480 how it erodes trust. In our institutions. 0:01:16.200 --> 0:01:16.880 And one more thing. 0:01:17.440 --> 0:01:20.200 For Reid is the founder and chief science officer of 0:01:20.240 --> 0:01:24.320 get Real Labs, where he consults with businesses, news organizations, 0:01:24.360 --> 0:01:28.760 and law enforcement to authenticate digital content. You might be 0:01:28.760 --> 0:01:31.600 wondering how far Reed got into this field. If so, 0:01:32.080 --> 0:01:32.840 you're not alone. 0:01:33.120 --> 0:01:34.800 Somebody said to me the other day, Oh, you were 0:01:34.800 --> 0:01:37.120 so prescient. I'm like, no, we weren't. We were just 0:01:37.160 --> 0:01:37.960 screwing around. 0:01:38.520 --> 0:01:42.920 Farred first started pondering the implications of digital images back 0:01:42.920 --> 0:01:43.960 in nineteen ninety seven. 0:01:44.280 --> 0:01:47.680 This is really pre digital Almost film was still the 0:01:47.720 --> 0:01:50.080 dominant source of media that we took photographs on the 0:01:50.120 --> 0:01:53.280 Internet was nothing right. There was no social media, and 0:01:53.360 --> 0:01:56.440 everything was very nascent. You could see the trends, you 0:01:56.520 --> 0:01:59.240 knew things. Something was bubbling up with the Internet and 0:01:59.240 --> 0:02:02.400 with digital technolog Freed was a postdoc at the time. 0:02:02.720 --> 0:02:05.840 I was at the library getting a book, which now 0:02:05.960 --> 0:02:09.720 just seems quaint, and I was waiting in line, and 0:02:09.760 --> 0:02:11.679 there was a return card, and on the return cart 0:02:11.800 --> 0:02:14.320 was a big book called the Federal Rules of Evidence. 0:02:14.919 --> 0:02:17.200 I'm not a legal scholar, I'm not a lawyer, but 0:02:17.240 --> 0:02:19.040 I was bored and I flipped it open to a 0:02:19.120 --> 0:02:23.160 random page and it was titled introducing Photographs into Evidence 0:02:23.160 --> 0:02:25.760 in a Court of Law. And I liked taking photographs. 0:02:25.760 --> 0:02:27.960 I was working with digital images, but nothing to do 0:02:28.000 --> 0:02:29.679 with this topic, and I thought, I wonder what the 0:02:29.760 --> 0:02:32.040 rules are, and so I read it and there was 0:02:32.040 --> 0:02:35.640 almost a footnote that said, there's this digital format and 0:02:35.720 --> 0:02:39.040 we're going to treat digital the same way we treat analog. 0:02:40.080 --> 0:02:42.400 And I just remember thinking, I don't know anything, but 0:02:42.520 --> 0:02:43.919 that seems like a bad idea. 0:02:44.600 --> 0:02:47.560 This passage really stuck with him, and for years he 0:02:47.600 --> 0:02:51.040 couldn't stop thinking about the implications of a digital world, 0:02:51.760 --> 0:02:54.880 the fact that digital manipulation would change our perception of 0:02:54.880 --> 0:03:00.320 what's real because the photographic medium had fundamentally shifted. What 0:03:00.440 --> 0:03:03.600 surprised him was that few others were taking note. 0:03:04.000 --> 0:03:07.040 It's really unusual in an academic life where you start 0:03:07.080 --> 0:03:09.080 thinking about a problem and you go into the academic 0:03:09.160 --> 0:03:13.000 literature and there is nothing. It was just crickets, because 0:03:13.000 --> 0:03:15.440 there was no reason to be thinking about the problem. 0:03:15.960 --> 0:03:19.160 Two years later, as a professor of computer science at Dartmouth, 0:03:19.600 --> 0:03:22.799 he was playing around in photoshop creating a comic image 0:03:22.800 --> 0:03:25.440 of his friend when he had an epiphany. 0:03:26.040 --> 0:03:31.079 Mathematically, I just did something very interesting. I introduced pixels 0:03:31.320 --> 0:03:35.240 that have been synthesized by photoshop to make the image bigger, right, 0:03:35.280 --> 0:03:37.920 because they didn't exist, and I remember thinking, oh, I 0:03:37.920 --> 0:03:39.160 should be able to detect that. 0:03:40.240 --> 0:03:43.880 In that moment, he started writing code and actually developed 0:03:43.920 --> 0:03:48.120 programs to detect digital manipulation. The world woke up to 0:03:48.120 --> 0:03:50.240 the importance of this work, and he started getting asked 0:03:50.280 --> 0:03:53.120 to chime in on serious cases for the Associated Press, 0:03:53.440 --> 0:03:55.800 for law enforcement, for national security. 0:03:56.240 --> 0:04:00.400 And then twenty fifteen, sixteen seventeen AI hit and the 0:04:00.400 --> 0:04:03.920 world exploded. But it exploded for a few reasons because one, 0:04:04.320 --> 0:04:06.680 at least with Photoshop, there was a barrier to entry. 0:04:06.920 --> 0:04:09.160 You had to actually know how to use photoshop. But 0:04:09.200 --> 0:04:11.400 then when AI came around, you just go to chat 0:04:11.480 --> 0:04:13.800 cheepetiting type, give me an image of X, right, and 0:04:13.840 --> 0:04:15.520 give me an image of Y, give me a video 0:04:15.520 --> 0:04:16.960 of this, give me an audio of this. And so 0:04:17.000 --> 0:04:21.080 suddenly there's no barrier to entry. But more importantly, social 0:04:21.120 --> 0:04:24.600 media dominates the landscape. We went from a few million 0:04:24.720 --> 0:04:28.000 users to a few billion users, and so now not 0:04:28.080 --> 0:04:31.719 only could people easily with no barrier to entry, create 0:04:32.120 --> 0:04:36.400 fake content, they could distribute it to the masses and 0:04:37.160 --> 0:04:42.080 it gets amplified because the algorithms amplify the most outrageous things. 0:04:42.440 --> 0:04:46.280 People want things that conform to the worldview. We are hyperpartisan, 0:04:46.400 --> 0:04:51.480 both here and abroad, and that was the perfect storm create, distribute, amplify, 0:04:51.839 --> 0:04:56.040 rinse and repeat. And so now through the AI revolution, 0:04:56.680 --> 0:04:58.560 it's bizarre what's happening. 0:04:59.120 --> 0:05:01.520 We'll dive into the world Careed does on deep fakes 0:05:01.520 --> 0:05:03.960 in a bit, but first I had to ask you 0:05:04.080 --> 0:05:10.960 about something seemingly completely unrelated, death bods. So you were 0:05:11.040 --> 0:05:14.359 quoted in this Atlantic article about death bods with the 0:05:14.400 --> 0:05:18.440 headline no one is ready for digital immortality, So nie it. 0:05:18.560 --> 0:05:20.560 It'll be good to define on terms like what do 0:05:20.600 --> 0:05:23.080 we mean by this idea of digital immortality? 0:05:23.720 --> 0:05:26.080 Yeah, I don't know that it's a well established term. 0:05:26.120 --> 0:05:30.320 But here's my definition. Is that your likeness, the way 0:05:30.360 --> 0:05:32.719 you think, the way you talk, the way you look, 0:05:33.200 --> 0:05:37.000 lives on an eternity in a digital form through a 0:05:37.120 --> 0:05:41.600 version of AI that embodies how I write, how I think, 0:05:41.720 --> 0:05:44.800 how I talk in order to interact with other people. 0:05:45.000 --> 0:05:47.359 It's interactive, that's the key, but it's dynamic. 0:05:47.680 --> 0:05:50.120 What got you interested in this topic and why did 0:05:50.160 --> 0:05:51.520 you agree to be a source in the story. 0:05:51.960 --> 0:05:55.359 So this is almost a philosophical and legal question, and 0:05:55.440 --> 0:05:58.320 I'm neither of those things. But I got to say 0:05:58.440 --> 0:06:03.120 I've been thinking a lot about it, technically, personally, philosophically. 0:06:03.880 --> 0:06:06.600 Here's why I've been thinking about it. So one is, 0:06:06.640 --> 0:06:08.840 I'm a professor. I've been a professor for twenty five years. 0:06:08.880 --> 0:06:11.480 I love teaching. I love my students. I hate them 0:06:11.480 --> 0:06:15.000 some days, but I usually love them. They're amazing and 0:06:15.040 --> 0:06:18.200 weird and wonderful in many ways. So is there a 0:06:18.279 --> 0:06:21.880 story here where I can keep teaching after I die? 0:06:22.560 --> 0:06:25.640 Like there's something sort of magical about that. I think 0:06:25.680 --> 0:06:27.480 about it for my parents. Both my parents are now 0:06:27.480 --> 0:06:29.560 in their late eighties. One of them will die first, 0:06:29.600 --> 0:06:31.599 almost certainly, And what does it mean for the one 0:06:31.760 --> 0:06:35.360 who They've been together for fifty years? So there's parts 0:06:35.400 --> 0:06:38.039 of it. I think this is wonderful, this idea that 0:06:38.200 --> 0:06:40.240 one of my parents can wake up and open up 0:06:40.240 --> 0:06:42.479 their iPad and have a conversation with the person that 0:06:42.480 --> 0:06:45.039 they spent fifty years of their lives with. On the 0:06:45.080 --> 0:06:47.680 other hand, if that happens early in life, is that 0:06:47.720 --> 0:06:50.479 healthy for somebody? If a thirty year old loses their spouse, 0:06:50.560 --> 0:06:53.359 is that good? That they never sort of physically move on. 0:06:53.880 --> 0:06:55.919 I also think about it from a technical perspective, what 0:06:55.960 --> 0:06:58.520 would that look like for somebody who's famous where there's 0:06:58.520 --> 0:07:01.000 a big digital footprint. I think we have all the 0:07:01.040 --> 0:07:04.039 pieces to do that. We have the large language models, 0:07:04.480 --> 0:07:07.919 we have voice, we have likeness, we have video, and 0:07:07.920 --> 0:07:11.040 you're already seeing people do this creating digital avatars of 0:07:11.160 --> 0:07:13.640 both people who are with us and not with us, 0:07:13.960 --> 0:07:16.120 so that you can interact with them. I can go 0:07:16.240 --> 0:07:18.440 scrape every single piece of writing that Martin Luther King 0:07:18.520 --> 0:07:20.600 Junior wrote. I can grab his speeches, I can grab 0:07:20.640 --> 0:07:22.200 his likeness, I can grab his voice, and I could 0:07:22.240 --> 0:07:24.520 create an avatar of him that I could interact with. 0:07:24.760 --> 0:07:27.200 Well, it reminds me of your work on deep fis 0:07:27.240 --> 0:07:29.840 in some sense, because, as you said, exactly, all the 0:07:29.840 --> 0:07:33.800 pieces are there technically and otherwise. Yeah, but society's clearly 0:07:33.800 --> 0:07:34.239 not ready. 0:07:34.880 --> 0:07:37.040 I don't think we're ready. But look a lot of things, 0:07:37.320 --> 0:07:39.720 if you look at the last two, three, four or 0:07:39.760 --> 0:07:42.880 five decades from technology we weren't ready for and we 0:07:42.920 --> 0:07:43.720 became ready for it. 0:07:43.800 --> 0:07:43.960 Right. 0:07:44.320 --> 0:07:46.400 Look, you can go back to in vitual fertilization. When 0:07:46.400 --> 0:07:49.840 it first started, people were freaked out by that completely 0:07:50.000 --> 0:07:52.160 normal now, right, And by the way, this could also 0:07:52.200 --> 0:07:55.200 be generational. I can imagine some of my students here 0:07:55.200 --> 0:07:59.400 at UC Berkeley think sure, who cares, right, And I'm 0:07:59.440 --> 0:08:01.000 an older guy and I'm like, ah, that seems a 0:08:01.000 --> 0:08:03.560 little weird. So this may just go away generationally, which 0:08:03.600 --> 0:08:04.600 is usually how this happens. 0:08:04.640 --> 0:08:07.160 By the way, do you think we'll see a fundamental 0:08:07.400 --> 0:08:09.440 shift in our society in this case in terms of 0:08:10.400 --> 0:08:11.640 how we think about death. 0:08:12.560 --> 0:08:16.920 I think this idea of a digital immortality is really profound. 0:08:17.600 --> 0:08:21.200 And look, I don't know where this AI revolution is 0:08:21.200 --> 0:08:24.120 going right now. I don't think anybody really does. But 0:08:24.200 --> 0:08:27.800 something is happening. There is something here that is quite dramatic. 0:08:28.240 --> 0:08:30.320 I think it's going to reshape society. I think it's 0:08:30.360 --> 0:08:32.520 going to reshape education. I think it's going to reshape 0:08:32.520 --> 0:08:34.800 the workforce. I think it's going to reshape a lot 0:08:34.800 --> 0:08:37.920 of things. And I do think your likeness or your 0:08:37.920 --> 0:08:40.080 being or your essence or whatever you want to call 0:08:40.120 --> 0:08:42.800 that can live on and you can interact with people. 0:08:43.000 --> 0:08:45.880 You can continue to have a podcast after you die, 0:08:46.320 --> 0:08:47.840 you can keep interviewing people. 0:08:49.320 --> 0:08:53.280 When we come back. How deep fakes impact everyone even 0:08:53.320 --> 0:09:02.000 if you don't know it. There's an interesting point of 0:09:02.040 --> 0:09:06.280 intersection between so death bots and your more core field 0:09:06.280 --> 0:09:10.240 of study, and that's this Indian politician who a parliamentary candidate, 0:09:10.280 --> 0:09:14.760 who created a video of his deceased father endorsing him 0:09:14.760 --> 0:09:15.760 as his rightful heir. 0:09:16.080 --> 0:09:17.440 Yeah. 0:09:17.520 --> 0:09:19.240 I mean this is kind of a world's collide moment 0:09:19.280 --> 0:09:22.599 between misinformation, deep fakes, and digital immortality. 0:09:22.840 --> 0:09:25.640 Yeah. Yeah, So for people who didn't see it, India 0:09:25.679 --> 0:09:28.520 had an election this year, big one and you know, 0:09:29.200 --> 0:09:33.480 billion plus people voting. It was chaotic, and a politician 0:09:33.480 --> 0:09:36.320 did exactly this. His father was a well known politician, 0:09:36.360 --> 0:09:38.600 and he created a digital recreation with his voice and 0:09:38.600 --> 0:09:40.480 his likeness and he was talking and endorsing his son. 0:09:41.040 --> 0:09:43.520 So I have a couple of thoughts on that right now, 0:09:43.559 --> 0:09:46.559 in this particular moment, as we're still grappling, I think 0:09:46.559 --> 0:09:50.080 there should be two rules, which are consent and disclosure. 0:09:51.240 --> 0:09:52.920 And it's really simple, like, if you're going to use 0:09:52.920 --> 0:09:55.840 somebody's likeness, you should have consent, and if you're going 0:09:55.880 --> 0:09:59.000 to distribute it, you should have disclosure. Now, consent is 0:09:59.000 --> 0:10:01.520 difficult when somebody is dead. But if I want to 0:10:01.520 --> 0:10:04.880 get an endorsement from somebody, who's living. I have to 0:10:04.880 --> 0:10:08.280 get their consent yep. And if I distribute that, it 0:10:08.320 --> 0:10:11.880 has to be very clearly labeled and disclosed as this 0:10:11.920 --> 0:10:14.640 is AI generated. I'll give you a really nice example 0:10:14.679 --> 0:10:15.880 of this where it was sort of cool. I was 0:10:15.960 --> 0:10:19.800 during the Olympics. One of the newscasters, well known and 0:10:20.000 --> 0:10:22.559 I'm just blinking out his name right now, was creating 0:10:22.600 --> 0:10:26.080 AI generated personalized summaries. So my wife was watching the 0:10:26.080 --> 0:10:28.880 Olympics and she would get these personalized summaries from the broadcaster. 0:10:29.600 --> 0:10:32.240 So the content was personalized to her based on what 0:10:32.320 --> 0:10:36.079 she was watching. And then the voice being generated was his, 0:10:36.920 --> 0:10:39.760 and the script was being AI generated. Everything was with 0:10:39.840 --> 0:10:41.920 his permission, and it was disclosed to her that it 0:10:41.960 --> 0:10:44.640 was AI generated and summarized. And I think that was 0:10:44.679 --> 0:10:47.960 really well done in terms of the things that were 0:10:48.000 --> 0:10:50.240 made clear of what you were getting and how it 0:10:50.280 --> 0:10:51.200 was being delivered to you. 0:10:52.120 --> 0:10:54.840 That's sort of a high watermark for how this stuff works. 0:10:54.840 --> 0:10:57.040 When it works well. Yeah, Do you think as a 0:10:57.040 --> 0:11:00.920 society we're more likely to move toward that high water 0:11:01.040 --> 0:11:06.079 mark through collective demand or through regulation or through some 0:11:06.120 --> 0:11:08.600 decision from the tech overlords like what gets us there? 0:11:08.679 --> 0:11:12.000 More broadly, yeah. I mean, there's nothing in the last 0:11:12.040 --> 0:11:14.880 twenty years or twenty five years that gives me confident 0:11:14.960 --> 0:11:16.800 that our tech overlords are going to do the right thing. 0:11:17.520 --> 0:11:20.120 They're going to do the thing that maximizes their profits. 0:11:20.120 --> 0:11:23.040 And we know this. Let's stop pretending otherwise that Silicon 0:11:23.120 --> 0:11:24.880 Value is anything other than it is. It's a modern 0:11:24.960 --> 0:11:27.480 day Wall Street in some ways, by the way, even 0:11:27.559 --> 0:11:31.480 more powerful, right, because they control information, not just money, 0:11:31.559 --> 0:11:34.800 and that arguably is much more powerful. I don't think 0:11:34.840 --> 0:11:38.319 this comes from consumers, because we're not customers, we're the product. 0:11:39.160 --> 0:11:42.480 We as users, I should say, have almost no power 0:11:42.520 --> 0:11:45.920 at all. And so the media we tried, right, we 0:11:45.960 --> 0:11:49.040 tried criticizing and embarrassing, and we tried dragging them in 0:11:49.040 --> 0:11:53.840 front of Congress. Nothing effects change. So what does good regulation? 0:11:54.320 --> 0:11:56.560 We got to put guardrails on this and look, there's 0:11:56.720 --> 0:12:00.120 nothing there is nothing in our physical world that it 0:12:00.160 --> 0:12:04.400 is not subject to regulation to make products safe and reasonable. 0:12:04.679 --> 0:12:07.240 But somehow we've abandoned that for the last twenty five 0:12:07.320 --> 0:12:09.920 years because it's the Internet. So I do think it's 0:12:09.960 --> 0:12:11.400 going to have to come I don't think it's going 0:12:11.480 --> 0:12:14.040 to come from the US. It is coming from the UK. 0:12:14.200 --> 0:12:16.720 It is coming from the EU, it is coming from Australia, 0:12:17.120 --> 0:12:19.160 and I think those are going to be the leaders 0:12:19.160 --> 0:12:21.160 in this space. And you saw this with GDPR with 0:12:21.200 --> 0:12:24.040 the privacy rules in many ways that I don't think 0:12:24.040 --> 0:12:26.160 it solved the privacy problem around the world, but it 0:12:26.240 --> 0:12:29.160 moved the needle on the problem. And the EU and 0:12:29.200 --> 0:12:32.280 the UK have moved very aggressively on AI safety, on 0:12:32.360 --> 0:12:36.439 digital safety, and on misuse of monopolies, and I think 0:12:36.440 --> 0:12:38.280 it's going to have to come at that level. 0:12:38.840 --> 0:12:40.880 I want to talk about some of the more personal 0:12:40.960 --> 0:12:44.040 ways in which we can experience deep fakes. I think 0:12:44.040 --> 0:12:47.040 a lot of people think maybe it only touches politicians 0:12:47.120 --> 0:12:50.120 or celebrities. But there was an NPR story about a 0:12:50.160 --> 0:12:54.280 case you worked someone that involved a Baltimore teacher. Can 0:12:54.320 --> 0:12:55.440 you talk about what happened there? 0:12:55.960 --> 0:12:58.640 This case is I'm fascinated by, and I still don't 0:12:58.640 --> 0:13:00.199 think we've gotten to the end of it. Tell you, 0:13:00.240 --> 0:13:03.319 first of all, your listeners, what the case is. Baltimore 0:13:03.320 --> 0:13:06.520 Public School audio of the principle saying things that were 0:13:06.640 --> 0:13:10.920 racist was leaked, and it was leaked to some news outlet, 0:13:11.400 --> 0:13:13.360 and it was bad and it was if you listen 0:13:13.440 --> 0:13:17.040 to it it's pretty bad, and the principal said, this 0:13:17.160 --> 0:13:21.960 isn't me, this is AI generated, And we analyze the audio. 0:13:22.120 --> 0:13:26.679 Several labs analyze the audio. There is alteration to the audio. 0:13:26.840 --> 0:13:29.800 That is, we can hear and see that it's been 0:13:29.880 --> 0:13:34.440 spliced together five or six segments, but when we analyze 0:13:34.440 --> 0:13:37.560 the individual segments, it is not one hundred percent clear 0:13:37.600 --> 0:13:39.760 to us that it is AI generated. It could be 0:13:40.600 --> 0:13:42.640 that he said these things, but they were sort of 0:13:42.720 --> 0:13:45.160 stitched together in a way that put them out of context, 0:13:45.440 --> 0:13:48.000 which would be deceptive. It could be that it's AI 0:13:48.040 --> 0:13:50.800 generated and our tools simply didn't detect it. It could 0:13:50.840 --> 0:13:52.640 be that this is a case of the liar's dividend, 0:13:52.640 --> 0:13:54.760 where the principle really did say this, but he's claiming 0:13:54.800 --> 0:13:55.520 he didn't say it. 0:13:55.640 --> 0:13:58.520 Honey, can you explain exactly what the liar's dividend is? 0:13:58.960 --> 0:14:01.480 The liar's divining go something like this. It says, when 0:14:01.520 --> 0:14:04.840 you live in a world or anything can be manipulated. 0:14:04.880 --> 0:14:06.839 Any image can be fake, any audio can be fake, 0:14:06.880 --> 0:14:09.520 any video can be fake. Nothing has to be real. 0:14:09.840 --> 0:14:13.240 I get to use the fact that fake things exist 0:14:13.280 --> 0:14:16.520 as an excuse for what I've done. But this case 0:14:16.559 --> 0:14:21.960 is a really good example of how dangerous this technology 0:14:22.040 --> 0:14:25.160 is for two reasons. One is, with twenty to thirty 0:14:25.160 --> 0:14:28.920 seconds of your voice, I don't need hours. I can 0:14:28.960 --> 0:14:31.880 clone your voice. I can upload it to an AI 0:14:32.040 --> 0:14:34.200 tool that I use and that I can type and 0:14:34.240 --> 0:14:37.680 have you say anything I want. That means anybody with 0:14:37.800 --> 0:14:41.480 twenty seconds of their voice available has a vulnerability. So 0:14:41.520 --> 0:14:45.200 this is not for movie stars and podcasters. This is everybody. 0:14:45.360 --> 0:14:49.840 Number one. Number two is anybody who's caught saying or 0:14:49.880 --> 0:14:53.440 doing something that they don't want to take ownership of 0:14:53.560 --> 0:14:56.760 can say it's fake. Yep, the dog ate my homework, 0:14:57.040 --> 0:14:59.840 all right, this is easy. And so both of those 0:14:59.880 --> 0:15:03.720 are problematic because where's our shared sense of reality. It 0:15:03.800 --> 0:15:07.000 used to be when you had images and video, despite 0:15:07.000 --> 0:15:09.280 the fact that there was photoshop, despite the fact that 0:15:09.520 --> 0:15:13.240 Hollywood could could manipulate videos, we had a pretty reasonable 0:15:13.240 --> 0:15:16.040 confidence in what we read and saw and heard. And 0:15:16.080 --> 0:15:18.880 you can't say that anymore. This is why I spent 0:15:18.960 --> 0:15:21.640 so much time talking to journalists and fact checkers and 0:15:21.760 --> 0:15:25.560 lawyers and law enforcement. So on this particular case, it 0:15:25.640 --> 0:15:28.640 really showed how this has trickled all the way down 0:15:28.920 --> 0:15:30.359 to high school teachers. 0:15:30.600 --> 0:15:33.800 Zooming out from the individuals to the collective. One of 0:15:33.840 --> 0:15:37.480 the interesting things that happens is whenever there's like a 0:15:37.520 --> 0:15:40.720 world event that everyone's paying attention to, you get this 0:15:40.840 --> 0:15:44.280 fire hose of fake images. I remember in the early 0:15:44.400 --> 0:15:47.720 days of the conflict in Gaza, there was this aerial 0:15:47.840 --> 0:15:52.160 image with what was supposed to be Palestinian tents making 0:15:52.200 --> 0:15:55.520 the word help us or you know. Right out to 0:15:55.560 --> 0:15:58.280 the La fires began, there were these images of the 0:15:58.320 --> 0:16:01.280 Hollywood Sign on fire. I don't know how many people 0:16:01.320 --> 0:16:04.880 believe these images were actually true or in some ways, 0:16:04.880 --> 0:16:07.720 what the harm is if they did. But what's going 0:16:07.760 --> 0:16:08.160 on here? 0:16:08.880 --> 0:16:10.960 So let's start with the La fires. First of all, 0:16:11.200 --> 0:16:15.480 many images coming out of those fires were fake. What's 0:16:15.520 --> 0:16:18.200 the harm, Well, this one's easy. If people believe there's 0:16:18.240 --> 0:16:21.600 fire in this neighborhood, that is very bad. Fire departments 0:16:21.600 --> 0:16:23.440 are going to get distracted. First responders are going to 0:16:23.440 --> 0:16:26.000 get distracted. People are scared that their neighborhood is on fire. 0:16:26.000 --> 0:16:27.920 They're going to get distracted. So I do think there 0:16:27.960 --> 0:16:30.480 is real harm I think in the Gosspt images. Also, 0:16:30.920 --> 0:16:34.360 this is a complicated conflict, and we are all trying 0:16:34.360 --> 0:16:36.680 to get our heads around this thing and figure it out, 0:16:36.920 --> 0:16:40.880 and meanwhile people are fanning the flames, trying to push 0:16:40.880 --> 0:16:43.240 a particular narrative on either side, and I don't think 0:16:43.280 --> 0:16:47.960 that's healthy. Look, we can have serious discussions about how 0:16:48.000 --> 0:16:50.680 to combat climate change, we can have serious discussions about 0:16:50.720 --> 0:16:53.640 how to resolve the Israeli Palestinian conflict. We can have 0:16:53.680 --> 0:16:56.640 serious discussions about a lot of things, but we've got 0:16:56.680 --> 0:16:59.080 to start with a set of facts. And when you 0:16:59.120 --> 0:17:03.160 pollute the higher information ecosystem, we are at a loss. 0:17:03.560 --> 0:17:05.920 You could say, okay, well somebody believe the fake image 0:17:05.920 --> 0:17:08.960 of the tense. Okay, who cares? But here's why you care, 0:17:09.440 --> 0:17:11.840 Because then when the real images come out showing human 0:17:11.920 --> 0:17:16.360 rights violations, showing people being killed, people being bombed, how 0:17:16.359 --> 0:17:19.840 do I believe it? When you pollute the information ecosystem, 0:17:20.119 --> 0:17:23.479 everything is in doubt. And suddenly you have people who 0:17:23.520 --> 0:17:26.560 are denying that anybody's died, You have people denying that 0:17:26.600 --> 0:17:28.960 the fires exist, you have people denying that people are 0:17:29.040 --> 0:17:32.480 dying from COVID. Because this is how untrusting we have become, 0:17:33.040 --> 0:17:35.280 and that I have a real problem with, because look, 0:17:35.840 --> 0:17:39.359 no matter what side of the political or ideological aisle 0:17:39.440 --> 0:17:42.360 you are on, can we at least agree that if 0:17:42.359 --> 0:17:45.720 we don't have a shared factual system, a shared sense 0:17:45.720 --> 0:17:49.720 of reality, we do not have a society or democracy. 0:17:49.760 --> 0:17:52.160 We can't be arguing about whether one plus one is two. 0:17:52.480 --> 0:17:54.960 And I would argue that this problem started well before 0:17:55.080 --> 0:17:58.440 deep fakes. Social media is the one that is amplifying 0:17:58.920 --> 0:18:04.160 and encouraging this type of behavior because it engages users, 0:18:04.520 --> 0:18:08.400 drives ad drives attention, drives profits. The problem is not 0:18:08.480 --> 0:18:11.800 just the creation side, it's the distribution side, and that, 0:18:12.040 --> 0:18:14.359 I would argue, is the bigger problem here than the 0:18:14.400 --> 0:18:14.880 deep fake. 0:18:16.920 --> 0:18:19.679 Coming up. Harney for Reid on what it takes to 0:18:19.720 --> 0:18:29.639 identify a deep fake stay with us. When we first spoke, 0:18:29.720 --> 0:18:33.240 it was just five years ago in twenty nineteen. The 0:18:33.280 --> 0:18:36.760 big question at the time was is there going to 0:18:36.760 --> 0:18:42.919 be a causal piece of fake media that measurably sways 0:18:43.040 --> 0:18:46.399 the outcome of an election? And some people say the 0:18:46.400 --> 0:18:47.920 answer to that is no. I mean the New yorkd 0:18:47.920 --> 0:18:51.399 apiece in twenty twenty three saying basically, you know that 0:18:51.440 --> 0:18:55.199 the deep fakes haven't had that characterismic effect that some 0:18:55.240 --> 0:18:58.920 people thought they would. The Atlantic run a story recently 0:18:59.000 --> 0:19:01.919 under the headline AI's fingerprints were all over the election, 0:19:02.040 --> 0:19:05.480 but deep fakes and information weren't the main issue, and 0:19:05.720 --> 0:19:08.359 the kind of the point about both pieces was that 0:19:08.840 --> 0:19:10.919 what deep fakes are really being used for is to 0:19:10.960 --> 0:19:14.439 create memes and satire rather than to directly trick people. 0:19:14.960 --> 0:19:17.600 And the second point was quote to growing numbers of people, 0:19:18.200 --> 0:19:21.040 everything is fake now except what they know or other feel. 0:19:21.600 --> 0:19:21.959 Yeah. 0:19:22.000 --> 0:19:27.600 So has this been less explosively destructive than people thought 0:19:27.640 --> 0:19:29.560 it would be? Or are the New York and the 0:19:29.560 --> 0:19:31.520 Atlantic slightly missing the point in your view? 0:19:32.359 --> 0:19:35.720 I agree and disagree with them. I agree that there 0:19:35.760 --> 0:19:39.760 was no single atomic bomb that got dropped, and that 0:19:39.800 --> 0:19:41.560 you can draw a line from me to be saying 0:19:41.560 --> 0:19:44.400 this change in election, But nobody thought that was going 0:19:44.480 --> 0:19:45.879 to be the case. So I think that's a little 0:19:45.880 --> 0:19:47.320 bit of a straw man argument. 0:19:47.640 --> 0:19:47.800 Right. 0:19:47.880 --> 0:19:50.680 Okay, here's the other reason I disagree. Go talk to 0:19:50.720 --> 0:19:53.480 the people in Slovakia, because what they will tell you 0:19:54.320 --> 0:19:58.320 is that forty hours before election, there were two candidates, 0:19:58.359 --> 0:20:00.600 a Pronato and a pro Putin candidate, and the pro 0:20:00.720 --> 0:20:04.199 NATO candidate was up four points. A deep fake of 0:20:04.240 --> 0:20:06.320 the Pro NATO candidate was released saying We're going to 0:20:06.400 --> 0:20:09.560 rig the election, and two days later the pro Plutin 0:20:09.640 --> 0:20:12.800 candidate won by four points. There was an eight point 0:20:12.920 --> 0:20:16.400 swing in the polls in forty eight hours. Now were 0:20:16.440 --> 0:20:19.119 the polls wrong? Possibly? Did it have anything to do 0:20:19.200 --> 0:20:22.480 with the deep fake? Don't know, but this could have 0:20:22.600 --> 0:20:25.160 been the first example, just a couple of years ago, 0:20:25.600 --> 0:20:28.520 of where a deep fake was a tipping point. So 0:20:29.000 --> 0:20:31.560 I'm not sure i'd buy that story. I think this 0:20:31.640 --> 0:20:34.880 is more about death by a thousand cuts than by 0:20:35.119 --> 0:20:38.800 dropping an atomic bomb. I think that when you keep 0:20:38.840 --> 0:20:44.040 polluting the information ecosystem, everybody loses trust because you don't 0:20:44.119 --> 0:20:46.120 trust NPR, you don't trust in your times, you don't 0:20:46.119 --> 0:20:48.320 trust an end Who do you trust? Well, you trust 0:20:48.320 --> 0:20:50.879