31-year-old Divindra Singh Jaden may seem like an ordinary programmer, but in India's cyberspace, he has another nickname that many people know – the "King of Indian Counterfeiting."
It only takes three days for his team to produce a video of the election that is realistic enough to build momentum for the party he supports. The total cost of counterfeiting is only 600,000 rupees (about 52,200 yuan). What's more, if the quality requirements are not high, a fake video can be generated in a few minutes to discredit the opponent.
The phenomenon of AI-generated fake content storming elections is not unique to India, and similar situations are occurring in the United States and Europe. Keeni Bardans, an observer at the Alliance for Democratic Progress in the Software Industry, believes that "what is happening in India is like a canary cry in a coal mine, sounding the alarm bells around the world".
AI fraud is rampant
It has become a "semi-open secret" in India
On April 19, India's unprecedented general election officially kicked off, involving a total of 969 million voters in the country's 28 states, and the number of eligible voters is more than 10% of the world's total population. Due to the sheer size of the election, the election will be held in seven phases over a total of 44 days.
In this "extravaganza", Jaden's team is also playing a role. Jaden, who was previously an ordinary programmer, was stuck at home during the pandemic and had to make ends meet by taking odd jobs. At that time, he mainly provided some rough special effects for Bollywood production companies, such as "face swapping" and other ways to produce scenes of Bollywood movie stars singing and dancing.
▲ "Indian Counterfeiting King" Jaden
Now, with the explosive growth of AI tools, Jaden and a small team of several people have settled in the small town of Pashtunka in Rajasthan, where they are known as the "King of Indian Counterfeiters". The team created two video samples to demonstrate the technical capabilities of the "King of Indian Counterfeiters". In one of the videos, "Modi" waves to the audience and can call out each viewer's name, as if Modi himself was actually talking one-on-one to the audience. Another video shows a Hindi-speaking "fake Trump" enthusiastically saying, "I'm 'Trump Fake One,' I'm not the real Trump, how are you Deepp."
▲ The fake video of "Trump" speaking Hindi generated by Jaden's team
Mr. Jaden said he was politically neutral and did not want to be associated with a particular party, and that he did not work directly with candidates through channels such as advertising agencies and lobbying firms. During this election, Jaden's business has been in a state of "explosive orders", and almost all mainstream political parties have asked him to customize AI fake videos.
Jaden said that his team can produce a selected video in three days, with an offer of around Rs 600,000. If it's a long-term client, they'll spend a week building a database of images of the politician and letting the AI model learn from it. In post-production, if they don't need particularly high precision, they can generate a video of that candidate in a matter of minutes.
The use of generative AI has been a "semi-open secret" during India's general election. Some political parties use AI technology to "resurrect" two deceased party veterans to evoke memories of their supporters. Jaden revealed that he turned down a lot of "immoral" order requests, such as a political party approaching him through an agent and asking him to use AI to make a "sex tape" video of a competitor.
AI legislation needs to be improved
Expert: Counterfeiting has never been easier
With more than 60 countries and regions holding elections this year, experts have warned that because "technology knows no borders", India's proliferation of AI fraud is likely to spread around the world.
In January, a fake audio synthesized from Biden's voice went viral, and even became the ringing of many people's phones. In this audio, "Biden" calls on Democratic voters not to vote in the primary election and instead "save the votes for the real election." Technical experts say that fake audio files are smaller and more difficult to identify and traceable to the source than AI-generated videos and images. The incident also led the White House to tighten legislation on the use of AI technology in elections.
And on the eve of the Slovak elections last October, a fake audio of a "conversation" between Progressive Party leader Michal Simeka and local journalists went viral. In the AI-generated fake audio, Simeka claims to use money to buy votes from gypsies, and if elected, she wants to raise the price of local beer, and makes lewd jokes.
▲ Slovakia appeared before last year's election with AI-generated fake audio
Randy James, co-founder of Stanford University's Institute for Human-Centered Artificial Intelligence, believes that AI fraud has "never been easier" and that voters around the world should be on high alert in 2024. At present, the fake videos produced by AI have crossed the barrier of language translation and can be specially customized, so that candidates seem to be talking alone with netizens in front of a computer screen "one-on-one". Due to the extremely low cost of counterfeiting, tens of thousands of fake videos can cause great disruption to the normal order of the election.
Marijay Shaq, a political scientist at Stanford University, quipped: "Big companies are talking about the existential threat of AI to humanity, but they're scrambling to launch their own big models." While many voters feel they are smart enough to discern AI-generated disinformation, "elections tend to be won or lost in the range of a few percentage points," and as long as the disinformation is widespread enough, it can affect the middle voters who can decide the outcome of the outcome, according to Marij.
At present, the Biden administration has formed a working group to study AI legislation, but it will take at least several years for specific laws to be implemented. At present, the technology giants in the United States have also introduced some means to limit the dissemination of AI information. Nick Craig, Meta's president of global affairs, said that software such as Facebook and Whatsapp will recognize and annotate AI-generated information. In addition, companies such as Google are restricting users from asking specific and tendentious questions about ongoing elections in large AI models.
(Source: Red Star News)