1 00:00:03,120 --> 00:00:11,160 Speaker 1: Bloomberg Audio Studios, podcasts, radio news. In the small town 2 00:00:11,240 --> 00:00:15,400 Speaker 1: of Pushkar in northwest India, Deviandra Singh Jadoun is hard 3 00:00:15,400 --> 00:00:18,360 Speaker 1: at work in his studio. The Vibe is kind of 4 00:00:18,360 --> 00:00:21,160 Speaker 1: similar to a Silicon Valley startup. In its early days, 5 00:00:21,720 --> 00:00:24,360 Speaker 1: Jaduna set up a small workshop where he and a 6 00:00:24,400 --> 00:00:29,160 Speaker 1: handful of coworkers create digital content. He's gone from making 7 00:00:29,240 --> 00:00:32,680 Speaker 1: Bollywood spoofs to creating videos like this one. 8 00:00:32,680 --> 00:00:38,840 Speaker 2: Made Piai Hit Jane made Piai Man Media, Piati Abishek 9 00:00:38,960 --> 00:00:41,880 Speaker 2: Samson made Pagan Johan. 10 00:00:42,600 --> 00:00:46,200 Speaker 1: It appears to be Narendra Modi, the Prime Minister of India, 11 00:00:46,280 --> 00:00:50,519 Speaker 1: but it's actually a deep fake. Jadoun, who calls himself 12 00:00:50,640 --> 00:00:54,400 Speaker 1: the Indian deep Faker, is using AI technology to make 13 00:00:54,440 --> 00:00:58,320 Speaker 1: it seem like Mody is speaking directly to individual voters. 14 00:00:58,960 --> 00:01:02,240 Speaker 1: In that clip, you hear him sub in four different names. 15 00:01:02,560 --> 00:01:05,920 Speaker 1: He refers to each of them as my dear. The 16 00:01:06,000 --> 00:01:08,920 Speaker 1: largest election in history just got under way in India, 17 00:01:09,000 --> 00:01:12,640 Speaker 1: and Jadun is in high demand. He makes videos like 18 00:01:12,720 --> 00:01:16,440 Speaker 1: these and AI chatbots for candidates looking to capitalize on 19 00:01:16,520 --> 00:01:20,360 Speaker 1: a largely unregulated technology to reach a wider range of 20 00:01:20,400 --> 00:01:24,560 Speaker 1: constituents quickly and what Jadun is doing is changing how 21 00:01:24,680 --> 00:01:29,640 Speaker 1: politicians campaign in the world's largest democracy, and that could 22 00:01:29,640 --> 00:01:34,560 Speaker 1: have implications worldwide as countries grapple with the disinformation risks 23 00:01:34,600 --> 00:01:38,520 Speaker 1: that come with AI. This year, the stakes are really high. 24 00:01:38,720 --> 00:01:42,160 Speaker 1: There are national elections in more than sixty countries, including 25 00:01:42,160 --> 00:01:47,280 Speaker 1: the US, and this technology doesn't recognize borders. Here's another 26 00:01:47,360 --> 00:01:50,800 Speaker 1: video from Jadun. You might recognize this voice even if 27 00:01:50,800 --> 00:01:54,640 Speaker 1: you've never heard him speak in Hindi. Trump Nehihu raji 28 00:01:54,680 --> 00:01:58,400 Speaker 1: shtumkaka rehu may I slid Trump. Nahihu Tunita tum kaka 29 00:01:58,480 --> 00:02:01,640 Speaker 1: rehu may I slid Trump. That was a deep fake 30 00:02:01,720 --> 00:02:05,240 Speaker 1: of Donald Trump from the same public facing highlight reel 31 00:02:05,360 --> 00:02:09,480 Speaker 1: Jadoon made that featured the Modi video. In it, Donald 32 00:02:09,560 --> 00:02:13,320 Speaker 1: Trump is also addressing different people by name and noting 33 00:02:13,840 --> 00:02:19,680 Speaker 1: he is not the real Donald Trump. On today's show, 34 00:02:19,760 --> 00:02:23,080 Speaker 1: what a deep fake operation in India tells us about 35 00:02:23,080 --> 00:02:25,920 Speaker 1: the year ahead when there are elections all over the 36 00:02:25,960 --> 00:02:29,680 Speaker 1: world and the challenges of regulating this new and fast 37 00:02:29,720 --> 00:02:33,360 Speaker 1: growing technology. This is the big take from Bloomberg News. 38 00:02:33,560 --> 00:02:44,639 Speaker 1: I'm David Gerra, really, I promise Diveandra Jadoon is having 39 00:02:44,760 --> 00:02:47,880 Speaker 1: a pretty busy year. The thirty one year old has 40 00:02:47,880 --> 00:02:51,880 Speaker 1: built a business around making deep fakes. At first, his 41 00:02:52,000 --> 00:02:54,399 Speaker 1: work was mostly entertainment focused. 42 00:02:55,080 --> 00:02:58,600 Speaker 2: During COVID, he told me that he started using a 43 00:02:58,600 --> 00:03:04,880 Speaker 2: lot of these technolology tools where he started making videos 44 00:03:05,320 --> 00:03:13,880 Speaker 2: and superimposing other celebrities on Bollywood dance videos or superimposing 45 00:03:14,360 --> 00:03:18,400 Speaker 2: celebrities from Bollywood onto Hollywood movie clips. 46 00:03:18,639 --> 00:03:21,919 Speaker 1: That's Bloomberg. Sarita Rye, she covers AI in Asia from 47 00:03:21,919 --> 00:03:24,320 Speaker 1: her home base in Bangalore, having. 48 00:03:24,120 --> 00:03:27,880 Speaker 2: Really clunky little videos. He showed me a few examples. 49 00:03:27,880 --> 00:03:30,519 Speaker 2: They weren't really very technically advanced. 50 00:03:31,280 --> 00:03:35,000 Speaker 1: Sretha says. The advent of Open AI's chat GPT made 51 00:03:35,000 --> 00:03:38,280 Speaker 1: it easier for Jadune to make higher quality videos, and 52 00:03:38,440 --> 00:03:42,640 Speaker 1: soon he saw an opportunity to branch into politics in India, 53 00:03:42,760 --> 00:03:46,600 Speaker 1: where there's a big demand for AI generated videos. There 54 00:03:46,640 --> 00:03:49,680 Speaker 1: are broadly speaking, two kinds of deep fakes we're seeing 55 00:03:49,760 --> 00:03:52,880 Speaker 1: in elections this year. There are deep fakes from campaigns 56 00:03:52,920 --> 00:03:55,600 Speaker 1: who want the next generation AI driven version of a 57 00:03:55,720 --> 00:03:59,280 Speaker 1: text message tailored to each voter. The other kind are 58 00:03:59,400 --> 00:04:03,600 Speaker 1: intentionally misleading videos in which someone's likeness is used without 59 00:04:03,640 --> 00:04:07,200 Speaker 1: that person's knowledge or consent to influence the outcome. Of 60 00:04:07,240 --> 00:04:10,680 Speaker 1: an election and told Serita he has ground rules for 61 00:04:10,720 --> 00:04:12,200 Speaker 1: what kind of work he takes on. 62 00:04:12,840 --> 00:04:17,640 Speaker 2: He tells me that he doesn't accept contracts and commissions 63 00:04:17,640 --> 00:04:21,440 Speaker 2: that come to him directly from candidates. He would rather 64 00:04:21,920 --> 00:04:25,520 Speaker 2: it come from a digital agency, or an AI agency, 65 00:04:25,600 --> 00:04:29,159 Speaker 2: or a contractor or even a political party. But he 66 00:04:29,200 --> 00:04:33,120 Speaker 2: says usually they come through these intermediaries that are contracted 67 00:04:33,160 --> 00:04:37,240 Speaker 2: by the mainstream parties. He's definitely not playing sides in 68 00:04:37,320 --> 00:04:39,640 Speaker 2: terms of his work. He's told me that he's working 69 00:04:39,720 --> 00:04:42,120 Speaker 2: for practically all the major political parties. 70 00:04:42,520 --> 00:04:44,360 Speaker 1: What is the operation that he's built up? 71 00:04:44,360 --> 00:04:46,960 Speaker 2: Look like he started off as a one man shop, 72 00:04:47,839 --> 00:04:50,359 Speaker 2: but he since expanded because all of the work that 73 00:04:50,480 --> 00:04:52,800 Speaker 2: is coming to him. The first time I spoke to him, 74 00:04:52,800 --> 00:04:54,640 Speaker 2: he said he had a couple of workers, and that 75 00:04:54,800 --> 00:04:57,440 Speaker 2: was a few months ago, and then he told me 76 00:04:57,560 --> 00:05:00,479 Speaker 2: now he has five other employees who helped they make 77 00:05:00,520 --> 00:05:01,159 Speaker 2: these videos. 78 00:05:01,520 --> 00:05:05,120 Speaker 1: Making those personalized videos, like the ones of Mody and Trump, 79 00:05:05,279 --> 00:05:08,760 Speaker 1: involve feeding an AI model audio or video of a 80 00:05:08,800 --> 00:05:12,640 Speaker 1: person to capture their voice, facial expressions, and movements. 81 00:05:13,080 --> 00:05:15,479 Speaker 2: It takes a few days up to a week to 82 00:05:15,600 --> 00:05:20,080 Speaker 2: train the AI models, but once trained, he can produce 83 00:05:20,200 --> 00:05:24,640 Speaker 2: these very quickly, like instantly, almost like within minutes, he 84 00:05:24,680 --> 00:05:26,479 Speaker 2: can make these videos. 85 00:05:26,880 --> 00:05:30,080 Speaker 1: Sriita says that in India, politicians are using deep fake 86 00:05:30,160 --> 00:05:35,120 Speaker 1: videos like jededuns for voter outreach. It's a relatively inexpensive 87 00:05:35,160 --> 00:05:38,360 Speaker 1: way to appeal to people all over this huge country 88 00:05:38,720 --> 00:05:42,240 Speaker 1: directly in a way they can't at big campaign rallies, 89 00:05:42,760 --> 00:05:45,360 Speaker 1: and that's valuable in India, where there are some nine 90 00:05:45,400 --> 00:05:48,520 Speaker 1: hundred and sixty eight million eligible voters. 91 00:05:48,560 --> 00:05:52,600 Speaker 2: The level of hyper personalization that is possible with generative AI. 92 00:05:53,120 --> 00:05:57,359 Speaker 2: You can just train the model to say the main 93 00:05:57,440 --> 00:06:00,960 Speaker 2: message and then feed it with house and thousands of 94 00:06:01,080 --> 00:06:06,320 Speaker 2: memes and it will instantly reproduce that same video and 95 00:06:06,480 --> 00:06:12,080 Speaker 2: sync it perfectly with Narenermodi calling out each person by name. 96 00:06:12,680 --> 00:06:15,880 Speaker 1: The videos are so good, Sarita says, it can be 97 00:06:16,080 --> 00:06:19,760 Speaker 1: very difficult to discern their deep fakes, especially in a 98 00:06:19,800 --> 00:06:23,200 Speaker 1: country where technological literacy varies so widely. 99 00:06:23,720 --> 00:06:29,120 Speaker 2: These are millions and millions who have their first encounter 100 00:06:29,520 --> 00:06:34,279 Speaker 2: of any technological device because they have access to a 101 00:06:34,360 --> 00:06:38,560 Speaker 2: cheap smartphone. They're the ones who have never experienced the 102 00:06:38,560 --> 00:06:44,280 Speaker 2: Internet except for experiencing it or accessing it through this 103 00:06:44,400 --> 00:06:47,159 Speaker 2: device for the first time in the last few years. 104 00:06:47,960 --> 00:06:51,840 Speaker 2: So the kind of impact that AI would have in 105 00:06:51,839 --> 00:06:57,240 Speaker 2: India is certainly magnified because people are so much more 106 00:06:57,320 --> 00:07:02,039 Speaker 2: vulnerable to the kind of deceptions that AI can be 107 00:07:02,400 --> 00:07:05,560 Speaker 2: used for. I think that really is the power of 108 00:07:06,200 --> 00:07:12,400 Speaker 2: these AI technologies colliding with the ubiquity of smartphones and 109 00:07:12,600 --> 00:07:17,520 Speaker 2: the cheap broadband in a country like India where otherwise 110 00:07:18,040 --> 00:07:22,880 Speaker 2: it would be a really expensive exercise to get out 111 00:07:23,040 --> 00:07:24,720 Speaker 2: and get the message to the waters. 112 00:07:24,960 --> 00:07:26,760 Speaker 1: You said that he can work quickly and it's not 113 00:07:26,840 --> 00:07:27,760 Speaker 1: extremely expensive. 114 00:07:27,880 --> 00:07:32,440 Speaker 2: I'd certainly say it's peanuts compared with what physical campaigning 115 00:07:32,480 --> 00:07:37,320 Speaker 2: will cost some of these politicians. For instance, if Modi 116 00:07:37,440 --> 00:07:41,600 Speaker 2: flies is a private plane to get to a remote 117 00:07:41,600 --> 00:07:45,240 Speaker 2: corner of the country, here morey can address all of 118 00:07:45,280 --> 00:07:51,240 Speaker 2: those voters with a deep faith personalized, hyper personalized message 119 00:07:51,640 --> 00:07:54,880 Speaker 2: for a few for maybe six or seven thousand dollars. 120 00:07:55,280 --> 00:07:58,320 Speaker 1: Are you able to ascertain the effect that this kind 121 00:07:58,320 --> 00:08:00,920 Speaker 1: of technology is having and indeed it is likely to 122 00:08:00,960 --> 00:08:02,120 Speaker 1: have on the election? 123 00:08:02,920 --> 00:08:07,360 Speaker 2: Elections were a magical time when all kinds of performers, 124 00:08:07,440 --> 00:08:10,920 Speaker 2: all kinds of theater, all kinds of stuff came by 125 00:08:11,120 --> 00:08:14,280 Speaker 2: onto your neighborhood because that was part of the whole, 126 00:08:14,600 --> 00:08:17,280 Speaker 2: as we call it in India, the election tamasha, which 127 00:08:17,320 --> 00:08:22,960 Speaker 2: is the election theater, almost very theatrical stuff. But when 128 00:08:22,960 --> 00:08:27,360 Speaker 2: I look out today outside in my neighborhood, there's hardly 129 00:08:27,480 --> 00:08:30,960 Speaker 2: any visible sign of an election that is coming up. 130 00:08:31,000 --> 00:08:33,319 Speaker 2: And in fact Bangalore where I'm based, is going to 131 00:08:33,400 --> 00:08:37,480 Speaker 2: vote in a few days, and there is absolutely no 132 00:08:37,600 --> 00:08:43,079 Speaker 2: sign of any large scale campaigning or any overt display 133 00:08:43,200 --> 00:08:47,800 Speaker 2: of election material. There are no banners, there are no posters, 134 00:08:47,880 --> 00:08:51,200 Speaker 2: there is nothing of that sort, which was literally what 135 00:08:51,360 --> 00:08:54,040 Speaker 2: it used to be like plastered all over the streets, 136 00:08:54,400 --> 00:08:57,480 Speaker 2: hung on the buildings and everything. None of it. None 137 00:08:57,520 --> 00:09:01,120 Speaker 2: of that is happening today. What I and I see 138 00:09:01,160 --> 00:09:04,640 Speaker 2: that this is a change that technology is bringing about 139 00:09:05,080 --> 00:09:08,200 Speaker 2: that a lot of the campaigning David is going to 140 00:09:08,240 --> 00:09:13,079 Speaker 2: be really personalized, in fact hyper personalized to the electors. 141 00:09:14,480 --> 00:09:18,400 Speaker 1: On Friday, India began seven rounds of phased voting. The 142 00:09:18,440 --> 00:09:21,560 Speaker 1: election will continue until June first, and we expect to 143 00:09:21,600 --> 00:09:25,320 Speaker 1: learn the outcome three days after that. Mody expects to 144 00:09:25,320 --> 00:09:28,560 Speaker 1: be re elected and Jadoun is anticipating a lot of 145 00:09:28,600 --> 00:09:31,920 Speaker 1: his work will come days before voting opens. In tighter 146 00:09:32,080 --> 00:09:33,400 Speaker 1: state races. 147 00:09:33,320 --> 00:09:35,559 Speaker 2: A lot of it is going to reach them via 148 00:09:35,920 --> 00:09:39,600 Speaker 2: their funds, and a lot of it will be AI 149 00:09:39,720 --> 00:09:41,240 Speaker 2: generated material. 150 00:09:42,440 --> 00:09:46,400 Speaker 1: After the break. The challenges of regulating the technology behind 151 00:09:46,440 --> 00:09:49,800 Speaker 1: deep fase not just in India but all over the world. 152 00:09:56,679 --> 00:10:00,200 Speaker 1: The explosive growth of generative AI and its potential have 153 00:10:00,280 --> 00:10:03,839 Speaker 1: made it a challenge to regulate its proponents, to out 154 00:10:03,880 --> 00:10:07,120 Speaker 1: its potential to boost productivity and, in the case of elections, 155 00:10:07,120 --> 00:10:10,800 Speaker 1: to reach more voters. But its critics note its pitfalls 156 00:10:10,880 --> 00:10:14,599 Speaker 1: in an age of heightened disinformation and concerns about cybersecurity, 157 00:10:15,280 --> 00:10:18,480 Speaker 1: and when it comes to regulation, many governments are unable 158 00:10:18,640 --> 00:10:22,320 Speaker 1: or reluctant to move fast in the technologies infancy. In 159 00:10:22,360 --> 00:10:25,760 Speaker 1: the US, AI legislation is tied up in Congress. The 160 00:10:25,880 --> 00:10:29,840 Speaker 1: UK is also debating legislation. China does have rules in 161 00:10:29,880 --> 00:10:33,360 Speaker 1: place for AI companies, but there's little transparency into how 162 00:10:33,360 --> 00:10:37,120 Speaker 1: they work in practice. The European Union has passed some 163 00:10:37,160 --> 00:10:40,640 Speaker 1: of the most comprehensive regulations. These could be a blueprint 164 00:10:40,679 --> 00:10:45,400 Speaker 1: for lawmakers everywhere. Last year, it adopted the Artificial Intelligence Act, 165 00:10:45,520 --> 00:10:49,440 Speaker 1: which placed reporting requirements and restrictions on how companies operating 166 00:10:49,440 --> 00:10:52,120 Speaker 1: in the EU are able to develop and use AI. 167 00:10:52,800 --> 00:10:56,439 Speaker 1: Blomberg's Gillian Deutsch covered the debate over that law from Brussels. 168 00:10:57,320 --> 00:10:58,920 Speaker 3: I think the best way to look at the way 169 00:10:59,040 --> 00:11:01,520 Speaker 3: the use approaching it is to see that they're not 170 00:11:01,559 --> 00:11:05,319 Speaker 3: really regulating the technology itself. The regulating the uses of 171 00:11:05,600 --> 00:11:09,880 Speaker 3: the technology. They're required to submit risk assessments to your opinion, 172 00:11:10,080 --> 00:11:13,520 Speaker 3: and then these kind of more very serious harms pushed 173 00:11:13,520 --> 00:11:17,480 Speaker 3: by AI, for example, assigning social scores to citizens based 174 00:11:17,480 --> 00:11:21,280 Speaker 3: on their behavior, or even using emotional recognition technology in 175 00:11:21,320 --> 00:11:24,120 Speaker 3: the workplace or in schools, those are fed out banned 176 00:11:24,360 --> 00:11:26,800 Speaker 3: in the EU. These are the kind of good enough 177 00:11:26,880 --> 00:11:29,920 Speaker 3: rules to actually make sure that this kind of scary 178 00:11:30,080 --> 00:11:33,840 Speaker 3: and obviously very promising technology, the generative AI. Those are 179 00:11:33,840 --> 00:11:35,520 Speaker 3: the correct guard rills that we should put in place. 180 00:11:35,800 --> 00:11:38,120 Speaker 1: But Jillian says, in spite of how this bill is 181 00:11:38,160 --> 00:11:41,960 Speaker 1: framed and it's splashy rollout, the AI Act is still 182 00:11:42,080 --> 00:11:46,720 Speaker 1: limited in scope. There's disagreement over how strict regulation should be, 183 00:11:47,120 --> 00:11:50,680 Speaker 1: and there's fear that regulation will stifle innovation. 184 00:11:50,960 --> 00:11:53,640 Speaker 3: There's also a very kind of quick sudden pushback from 185 00:11:53,679 --> 00:11:56,560 Speaker 3: countries like Germany and France saying actually that those are 186 00:11:56,559 --> 00:11:58,640 Speaker 3: too many rules and actually Europe is going to shoot 187 00:11:58,640 --> 00:12:01,439 Speaker 3: itself in the foot. We are the first allegates to 188 00:12:02,000 --> 00:12:04,280 Speaker 3: place really kind of what we view is very strict 189 00:12:04,360 --> 00:12:08,160 Speaker 3: requirements on genetic A companies. They will just go elsewhere. 190 00:12:08,160 --> 00:12:10,760 Speaker 3: They're not going to invest on this continent. And we 191 00:12:10,800 --> 00:12:13,480 Speaker 3: also need to gain the benefits of AI, not just 192 00:12:14,120 --> 00:12:16,920 Speaker 3: over regulate based on the concerns. So what ended up 193 00:12:16,920 --> 00:12:20,679 Speaker 3: happening is really we have more basic transparency requirements. 194 00:12:20,920 --> 00:12:24,079 Speaker 1: Other countries are studying Europe's approach, but many of them 195 00:12:24,160 --> 00:12:28,120 Speaker 1: are calling on companies to self regulate. I asked reporter 196 00:12:28,160 --> 00:12:32,000 Speaker 1: Sareita Rye about this. What is the regulatory landscape when 197 00:12:32,000 --> 00:12:33,880 Speaker 1: it comes to AI look like in India today? 198 00:12:34,160 --> 00:12:37,520 Speaker 2: India has no regulation at all when it comes to AI. 199 00:12:38,559 --> 00:12:42,319 Speaker 2: Certainly there have been attempts to regulate deep fakes on 200 00:12:42,360 --> 00:12:45,920 Speaker 2: social media. That's kind of a roundabout way of getting there. 201 00:12:46,679 --> 00:12:52,040 Speaker 2: What India's government has done is asked social media platforms 202 00:12:52,080 --> 00:12:56,960 Speaker 2: such as Facebook and Google to regulate their own platforms. 203 00:12:57,080 --> 00:13:02,079 Speaker 2: The government has told these platforms that when these deep 204 00:13:02,120 --> 00:13:06,160 Speaker 2: fakes or AI generated content hits these platforms and people 205 00:13:06,800 --> 00:13:09,280 Speaker 2: bring it to their notice, they have only twenty four 206 00:13:09,320 --> 00:13:11,040 Speaker 2: hours to take the content down. 207 00:13:12,400 --> 00:13:16,080 Speaker 1: Representatives from Meta Alphabet and other tech companies say they're 208 00:13:16,120 --> 00:13:21,240 Speaker 1: trying to address AI's potential problems. Jillian says, experts assert 209 00:13:21,360 --> 00:13:24,120 Speaker 1: it's hard to know how big a role deep fakes 210 00:13:24,120 --> 00:13:27,720 Speaker 1: will play in determining the outcome of elections in twenty 211 00:13:27,760 --> 00:13:28,240 Speaker 1: twenty four. 212 00:13:28,880 --> 00:13:32,400 Speaker 3: We don't want to overemphasize how widespread or how impactful 213 00:13:32,559 --> 00:13:34,880 Speaker 3: de fakes are, but it obviously has captured a lot 214 00:13:34,880 --> 00:13:37,920 Speaker 3: of people's attention because this is something very new. It's 215 00:13:38,000 --> 00:13:40,840 Speaker 3: much cheaper to make, they're much faster to make, they're 216 00:13:40,880 --> 00:13:43,679 Speaker 3: much more sophisticated than they used to be, and they 217 00:13:43,679 --> 00:13:46,240 Speaker 3: don't have to influence every single person in a country. 218 00:13:46,800 --> 00:13:49,280 Speaker 3: You know, a lot of these governments, all of these 219 00:13:49,280 --> 00:13:52,080 Speaker 3: elections are decided on very small percentile differences. 220 00:13:52,400 --> 00:13:54,800 Speaker 1: For the time being, voters will need to bring a 221 00:13:54,840 --> 00:13:59,160 Speaker 1: critical eye to everything they see online. Jillian points out 222 00:13:59,200 --> 00:14:02,280 Speaker 1: some deep fakes are harder to detect as fakes than others. 223 00:14:02,720 --> 00:14:06,120 Speaker 1: Many videos still have telltale signs. There could be too 224 00:14:06,160 --> 00:14:10,280 Speaker 1: many fingers on someone's hand, or inconsistencies and how clothing 225 00:14:10,440 --> 00:14:11,520 Speaker 1: or accessories look. 226 00:14:12,160 --> 00:14:14,800 Speaker 3: Audio deep figs are even trickier to identify. We have 227 00:14:15,480 --> 00:14:17,960 Speaker 3: far fewer social cues to see that it's in correct, 228 00:14:18,000 --> 00:14:18,880 Speaker 3: or that it's been faked. 229 00:14:19,200 --> 00:14:21,800 Speaker 1: Soretha says this played out in New Hampshire during the 230 00:14:21,840 --> 00:14:25,160 Speaker 1: run up to the presidential primaries there in January. 231 00:14:25,080 --> 00:14:32,440 Speaker 2: Robocalls using voice cloning technology impersonating Biden asking to vote 232 00:14:32,480 --> 00:14:35,200 Speaker 2: for a candidate that certainly he wasn't endorsing. 233 00:14:35,560 --> 00:14:40,080 Speaker 1: Increasingly, deep fakes are being seen as a serious disinformation threat. 234 00:14:40,600 --> 00:14:43,440 Speaker 1: There is a fear about how this technology could be 235 00:14:43,520 --> 00:14:47,560 Speaker 1: used or exploited in American politics. Is there the same 236 00:14:47,960 --> 00:14:51,440 Speaker 1: apprehension or anxiety in India about the way that this 237 00:14:51,520 --> 00:14:54,360 Speaker 1: is being used or could be used, or is there 238 00:14:54,640 --> 00:14:58,440 Speaker 1: just a fascination with kind of the whizbang nature of this. 239 00:14:58,800 --> 00:15:01,880 Speaker 2: Certainly, I think the world over there are apprehensions of 240 00:15:02,240 --> 00:15:07,000 Speaker 2: about how these defates can be used to fool or 241 00:15:07,080 --> 00:15:11,720 Speaker 2: deceive voters. He's seen it happen in multiple countries, right 242 00:15:11,880 --> 00:15:15,280 Speaker 2: from Indonesia to Pakistan and even in the US. 243 00:15:15,800 --> 00:15:20,720 Speaker 1: But deep fake creators like Diveandra Jadun see opportunity. Sarita says, 244 00:15:21,160 --> 00:15:25,680 Speaker 1: his business is booming. Does he have designs or a 245 00:15:25,720 --> 00:15:29,600 Speaker 1: desire to take this work to other countries and get 246 00:15:29,640 --> 00:15:31,120 Speaker 1: involved in other countries politics. 247 00:15:31,360 --> 00:15:35,080 Speaker 2: Divindras told me that He has agents working for him 248 00:15:35,160 --> 00:15:38,200 Speaker 2: in multiple countries, such as Canada, which is going to 249 00:15:38,240 --> 00:15:41,560 Speaker 2: an election in twenty twenty five. I think he certainly 250 00:15:41,640 --> 00:15:45,760 Speaker 2: is ambitious. India has always been known for technology services outsourcing, 251 00:15:46,280 --> 00:15:50,640 Speaker 2: but AI and election deep fake outsourcing may now see 252 00:15:50,680 --> 00:15:53,320 Speaker 2: a new lease of life with the likes of Viviendra 253 00:15:53,480 --> 00:15:56,160 Speaker 2: and this might be a new type of outsourcing. So 254 00:15:56,240 --> 00:15:59,680 Speaker 2: I think he's building a new business model. 255 00:16:02,240 --> 00:16:04,840 Speaker 1: This is the Big Take from Bloomberg News. I'm David Gura. 256 00:16:05,080 --> 00:16:08,000 Speaker 1: Today's episode was produced by Alex Sagura. It was edited 257 00:16:08,000 --> 00:16:10,920 Speaker 1: by Aaron Edwards and Tom O'Sullivan. It was mixed by 258 00:16:11,000 --> 00:16:14,720 Speaker 1: Veronica Rodriguez. It was fact checked by David Fox. Naomi 259 00:16:14,760 --> 00:16:18,520 Speaker 1: Shavin is our senior producer. Elizabeth Ponso is our senior editor. 260 00:16:18,840 --> 00:16:22,200 Speaker 1: Nicole Beemster Bor is our executive producer. Sage Bauman is 261 00:16:22,240 --> 00:16:25,160 Speaker 1: our head of podcasts. Please follow and review The Big 262 00:16:25,200 --> 00:16:28,320 Speaker 1: Take wherever you listen to podcasts. It helps new listeners 263 00:16:28,320 --> 00:16:31,040 Speaker 1: find the show. Thanks for listening. We'll see you next week.