They’re not TV anchors, they’re avatars: How Venezuela is using AI-generated propaganda

A blond-haired man introduces a news story for House of News, an alleged newscast in which an English-speaking anchor attempts to demonstrate that Venezuela’s economy is not “really destroyed” as many claim, because hotel occupancy for the Carnival period is reportedly sky-high thanks to Venezuelans eager to spend their money on the Caribbean beaches. In another news segment, a Black news anchor discusses the profits generated by the Caribbean Series, a professional baseball tournament held a week ago in Caracas: $10 million in tickets to watch baseball games, $7 million in food bought by fans. The data seems surprising, considering that the government has not even revealed how much it cost to build two stadiums in record time with the country’s depleted finances – caused, as authorities are constantly repeating, by international sanctions.

The alleged journalists are Noah and Daren, two avatars created with artificial intelligence from the Synthesia software’s catalog of more than a hundred multiracial faces. Like Noah and Daren, there are avatars dressed as television hosts, but there is also Dave who can be made to look like a doctor or an executive, Carlo who wears a construction helmet, a woman in a hijab, a chef and even Santa Claus. A few months ago, some of them, also in their role as reporters, were used in a pro-China disinformation campaign, as The New York Times reported a few days ago.

The videos of the fake presenters talking about Venezuela had hundreds of thousands of views on YouTube, went viral on social media apps like TikTok and were inserted as paid advertising on that platform. In addition, they were aired on the state broadcaster Venezolana de Televisión, the Nicolás Maduro administration’s main broadcasting mouthpiece.

Chavismo – supporters of the government style associated with former Venezuelan president Hugo Chávez – had already made use of bots and armies of paid tweeters to promote certain hashtags and rarefy the online conversation. Now, while the world is still trying to grasp the pros and cons of ChatGPT, the complex propaganda apparatus orbiting the Venezuelan crisis has quickly incorporated artificial intelligence into its digital artillery.

With a subscription of just a few dollars a month, Synthesia clients can input a written script for the software to generate an ultra-realistic video with voices available in more than 100 languages and accents, and made to sync with the mouth of the avatar. The company was created in 2017 by entrepreneurs and researchers from universities in the United States and Europe, according to its website. The catalog of avatars is made up of “digital twins,” actors who have provided the images and have received payment for it, they say in their frequently asked questions section. They promote “synthetic media,” which they describe as “one of the most exciting developments enabled by recent progress in deep learning,” with which they seek to empower people for the creation of content for commercial purposes only, as “political, sexual, personal or discriminatory content” is not tolerated, they say. But this recent progress is also tied to the deepfake that lies behind this technology and adds fuel to the great epidemic of disinformation.

“Artificial intelligence is being democratized and made accessible. Many people have free trial periods for this kind of software, or it is left open for the public to train the programs with a view to making these videos increasingly realistic,” explains Héctor Mazarri, from Cazadores de Fake News, an organization that analyzes and verifies Venezuela-related news. This is the group that dissected the videos of Noah and Daren talking about Venezuela and alerted about a new state-sponsored disinformation campaign by Chavismo and the spread of misleading data.

“In this case, what we are seeing is an organized attempt to push certain narratives favorable to the government. Although a trained user can spot the errors, this is done so that no one is exempt from not believing it,” adds the journalist. “Disinformation is a global problem and its purpose is to make us mistrust everything and create polarization, which is why it is necessary to create a community of infocitizens so that people can understand how this works and put much of the information that comes to them in quarantine in order to contain the impulse to share it or believe it.”

For Mazarri, this is just the beginning of what is to come, which includes voice cloning, a technique that poses great challenges to hoax-hunting. “Development and investment has grown in artificial intelligence, but it is the public that has encouraged this boom, because the more often it is used, the more reliable it becomes and this generates great competition between technology companies.”

Conquering the conversation

Since 2018, the ProBox Observatory has been monitoring sociopolitical trends in social media in Venezuela, Cuba, Nicaragua and El Salvador. Closely following the propaganda machinery of Venezuela’s ruling party has allowed this group to identify patterns. Every day, early in the day, the Twitter account of the Ministry of Communication and Information of Venezuela puts out the hashtag that will be positioned that day thanks to an indeterminate army of real tweeters and automated accounts – the so-called troops – who will receive cash bonuses through Sistema Patria, a digital platform through which pensions and some government salaries are distributed.

“This is done with the intention of making international public opinion believe certain narratives, such as the hashtag this week #LasSancionesMatanSalario [SanctionsKillWages], which already has a million messages; but if you compare it with independent media you know that it is not so. Mainly it is about generating propaganda, diverting attention, changing the narrative,” explains Estefanía Da Silva, general coordinator at ProBox.

During the past week, according to this group, more than 10 million messages associated with Chavismo were generated. “[Some] 96.5% are manipulated messages that come from automated accounts or possibly bots. Everything moves through coordinated accounts in an inauthentic way.” In another case, under the hashtag #LasSancionesSonContraElPueblo (SanctionsAreAgainstThePeople), 1,350,000 messages were generated through 15,392 users, with an additional 5,750 potential bots and automated accounts. This produces an artificial behavior of the trend. Da Silva adds that they have also noticed participation by accounts from Nicaragua and Cuba to position specific hashtags, which leads them to conclude that these are coordinated actions between the propaganda apparatuses of those governments.

In a Telegram chat group that brings together 740 tweeters, there was a discussion of the amount of this week’s bonus, said to range between 96 and 120 bolivars (between $3 and $5). The Venezuelan government has established categories to reward work artificially promoting hashtags, as if they were the Olympics: bronze, silver and gold. “You must bear in mind that tweeters become spammers for the microblog network. This is why work must be paused so that your account is not banned or suspended,” recommended a tutorial shared several times in the group, which is called Tuiteros Activos.

The strategy has turned Twitter into a minefield for Venezuelans. ProBox monitors messages in the sociopolitical spectrum of the ruling party, but also from the opposition, civil society, anonymous accounts and other actors. Digital activism for human rights or, in recent weeks, to demand a salary increase, only grows when the protest also has its counterpart in the streets. In general, adds Da Silva, “the ruling party is conquering conversations,” even if this is achieved with bots, paid tweeters and now, too, with AI-generated avatars.

Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition